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         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 324
 325         /*
 326          * We don't want to deadlock with somebody trying to allocate a new
 327          * extent for the extent root while also trying to search the extent
 328          * root to add free space.  So we skip locking and search the commit
 329          * root, since its read-only
 330          */
 331         path->skip_locking = 1;
 332         path->search_commit_root = 1;
 333         path->reada = 2;
 334
 335         key.objectid = last;
 336         key.offset = 0;
 337         key.type = BTRFS_EXTENT_ITEM_KEY;
 338 again:
 339         mutex_lock(&caching_ctl->mutex);
 340         /* need to make sure the commit_root doesn't disappear */
 341         down_read(&fs_info->extent_commit_sem);
 342
 343         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 344         if (ret < 0)
 345                 goto err;
 346
 347         leaf = path->nodes[0];
 348         nritems = btrfs_header_nritems(leaf);
 349
 350         while (1) {
 351                 smp_mb();
 352                 if (fs_info->closing > 1) {
 353                         last = (u64)-1;
 354                         break;
 355                 }
 356
 357                 if (path->slots[0] < nritems) {
 358                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 359                 } else {
 360                         ret = find_next_key(path, 0, &key);
 361                         if (ret)
 362                                 break;
 363
 364                         caching_ctl->progress = last;
 365                         btrfs_release_path(extent_root, path);
 366                         up_read(&fs_info->extent_commit_sem);
 367                         mutex_unlock(&caching_ctl->mutex);
 368                         if (btrfs_transaction_in_commit(fs_info))
 369                                 schedule_timeout(1);
 370                         else
 371                                 cond_resched();
 372                         goto again;
 373                 }
 374
 375                 if (key.objectid < block_group->key.objectid) {
 376                         path->slots[0]++;
 377                         continue;
 378                 }
 379
 380                 if (key.objectid >= block_group->key.objectid +
 381                     block_group->key.offset)
 382                         break;
 383
 384                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 385                         total_found += add_new_free_space(block_group,
 386                                                           fs_info, last,
 387                                                           key.objectid);
 388                         last = key.objectid + key.offset;
 389
 390                         if (total_found > (1024 * 1024 * 2)) {
 391                                 total_found = 0;
 392                                 wake_up(&caching_ctl->wait);
 393                         }
 394                 }
 395                 path->slots[0]++;
 396         }
 397         ret = 0;
 398
 399         total_found += add_new_free_space(block_group, fs_info, last,
 400                                           block_group->key.objectid +
 401                                           block_group->key.offset);
 402         caching_ctl->progress = (u64)-1;
 403
 404         spin_lock(&block_group->lock);
 405         block_group->caching_ctl = NULL;
 406         block_group->cached = BTRFS_CACHE_FINISHED;
 407         spin_unlock(&block_group->lock);
 408
 409 err:
 410         btrfs_free_path(path);
 411         up_read(&fs_info->extent_commit_sem);
 412
 413         free_excluded_extents(extent_root, block_group);
 414
 415         mutex_unlock(&caching_ctl->mutex);
 416         wake_up(&caching_ctl->wait);
 417
 418         put_caching_control(caching_ctl);
 419         atomic_dec(&block_group->space_info->caching_threads);
 420         btrfs_put_block_group(block_group);
 421
 422         return 0;
 423 }
 424
 425 static int cache_block_group(struct btrfs_block_group_cache *cache,
 426                              struct btrfs_trans_handle *trans,
 427                              struct btrfs_root *root,
 428                              int load_cache_only)
 429 {
 430         struct btrfs_fs_info *fs_info = cache->fs_info;
 431         struct btrfs_caching_control *caching_ctl;
 432         struct task_struct *tsk;
 433         int ret = 0;
 434
 435         smp_mb();
 436         if (cache->cached != BTRFS_CACHE_NO)
 437                 return 0;
 438
 439         /*
 440          * We can't do the read from on-disk cache during a commit since we need
 441          * to have the normal tree locking.  Also if we are currently trying to
 442          * allocate blocks for the tree root we can't do the fast caching since
 443          * we likely hold important locks.
 444          */
 445         if (!trans->transaction->in_commit &&
 446             (root && root != root->fs_info->tree_root)) {
 447                 spin_lock(&cache->lock);
 448                 if (cache->cached != BTRFS_CACHE_NO) {
 449                         spin_unlock(&cache->lock);
 450                         return 0;
 451                 }
 452                 cache->cached = BTRFS_CACHE_STARTED;
 453                 spin_unlock(&cache->lock);
 454
 455                 ret = load_free_space_cache(fs_info, cache);
 456
 457                 spin_lock(&cache->lock);
 458                 if (ret == 1) {
 459                         cache->cached = BTRFS_CACHE_FINISHED;
 460                         cache->last_byte_to_unpin = (u64)-1;
 461                 } else {
 462                         cache->cached = BTRFS_CACHE_NO;
 463                 }
 464                 spin_unlock(&cache->lock);
 465                 if (ret == 1) {
 466                         free_excluded_extents(fs_info->extent_root, cache);
 467                         return 0;
 468                 }
 469         }
 470
 471         if (load_cache_only)
 472                 return 0;
 473
 474         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 475         BUG_ON(!caching_ctl);
 476
 477         INIT_LIST_HEAD(&caching_ctl->list);
 478         mutex_init(&caching_ctl->mutex);
 479         init_waitqueue_head(&caching_ctl->wait);
 480         caching_ctl->block_group = cache;
 481         caching_ctl->progress = cache->key.objectid;
 482         /* one for caching kthread, one for caching block group list */
 483         atomic_set(&caching_ctl->count, 2);
 484
 485         spin_lock(&cache->lock);
 486         if (cache->cached != BTRFS_CACHE_NO) {
 487                 spin_unlock(&cache->lock);
 488                 kfree(caching_ctl);
 489                 return 0;
 490         }
 491         cache->caching_ctl = caching_ctl;
 492         cache->cached = BTRFS_CACHE_STARTED;
 493         spin_unlock(&cache->lock);
 494
 495         down_write(&fs_info->extent_commit_sem);
 496         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 497         up_write(&fs_info->extent_commit_sem);
 498
 499         atomic_inc(&cache->space_info->caching_threads);
 500         btrfs_get_block_group(cache);
 501
 502         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 503                           cache->key.objectid);
 504         if (IS_ERR(tsk)) {
 505                 ret = PTR_ERR(tsk);
 506                 printk(KERN_ERR "error running thread %d\n", ret);
 507                 BUG();
 508         }
 509
 510         return ret;
 511 }
 512
 513 /*
 514  * return the block group that starts at or after bytenr
 515  */
 516 static struct btrfs_block_group_cache *
 517 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 518 {
 519         struct btrfs_block_group_cache *cache;
 520
 521         cache = block_group_cache_tree_search(info, bytenr, 0);
 522
 523         return cache;
 524 }
 525
 526 /*
 527  * return the block group that contains the given bytenr
 528  */
 529 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 530                                                  struct btrfs_fs_info *info,
 531                                                  u64 bytenr)
 532 {
 533         struct btrfs_block_group_cache *cache;
 534
 535         cache = block_group_cache_tree_search(info, bytenr, 1);
 536
 537         return cache;
 538 }
 539
 540 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 541                                                   u64 flags)
 542 {
 543         struct list_head *head = &info->space_info;
 544         struct btrfs_space_info *found;
 545
 546         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 547                  BTRFS_BLOCK_GROUP_METADATA;
 548
 549         rcu_read_lock();
 550         list_for_each_entry_rcu(found, head, list) {
 551                 if (found->flags & flags) {
 552                         rcu_read_unlock();
 553                         return found;
 554                 }
 555         }
 556         rcu_read_unlock();
 557         return NULL;
 558 }
 559
 560 /*
 561  * after adding space to the filesystem, we need to clear the full flags
 562  * on all the space infos.
 563  */
 564 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 565 {
 566         struct list_head *head = &info->space_info;
 567         struct btrfs_space_info *found;
 568
 569         rcu_read_lock();
 570         list_for_each_entry_rcu(found, head, list)
 571                 found->full = 0;
 572         rcu_read_unlock();
 573 }
 574
 575 static u64 div_factor(u64 num, int factor)
 576 {
 577         if (factor == 10)
 578                 return num;
 579         num *= factor;
 580         do_div(num, 10);
 581         return num;
 582 }
 583
 584 static u64 div_factor_fine(u64 num, int factor)
 585 {
 586         if (factor == 100)
 587                 return num;
 588         num *= factor;
 589         do_div(num, 100);
 590         return num;
 591 }
 592
 593 u64 btrfs_find_block_group(struct btrfs_root *root,
 594                            u64 search_start, u64 search_hint, int owner)
 595 {
 596         struct btrfs_block_group_cache *cache;
 597         u64 used;
 598         u64 last = max(search_hint, search_start);
 599         u64 group_start = 0;
 600         int full_search = 0;
 601         int factor = 9;
 602         int wrapped = 0;
 603 again:
 604         while (1) {
 605                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 606                 if (!cache)
 607                         break;
 608
 609                 spin_lock(&cache->lock);
 610                 last = cache->key.objectid + cache->key.offset;
 611                 used = btrfs_block_group_used(&cache->item);
 612
 613                 if ((full_search || !cache->ro) &&
 614                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 615                         if (used + cache->pinned + cache->reserved <
 616                             div_factor(cache->key.offset, factor)) {
 617                                 group_start = cache->key.objectid;
 618                                 spin_unlock(&cache->lock);
 619                                 btrfs_put_block_group(cache);
 620                                 goto found;
 621                         }
 622                 }
 623                 spin_unlock(&cache->lock);
 624                 btrfs_put_block_group(cache);
 625                 cond_resched();
 626         }
 627         if (!wrapped) {
 628                 last = search_start;
 629                 wrapped = 1;
 630                 goto again;
 631         }
 632         if (!full_search && factor < 10) {
 633                 last = search_start;
 634                 full_search = 1;
 635                 factor = 10;
 636                 goto again;
 637         }
 638 found:
 639         return group_start;
 640 }
 641
 642 /* simple helper to search for an existing extent at a given offset */
 643 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 644 {
 645         int ret;
 646         struct btrfs_key key;
 647         struct btrfs_path *path;
 648
 649         path = btrfs_alloc_path();
 650         BUG_ON(!path);
 651         key.objectid = start;
 652         key.offset = len;
 653         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 654         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 655                                 0, 0);
 656         btrfs_free_path(path);
 657         return ret;
 658 }
 659
 660 /*
 661  * helper function to lookup reference count and flags of extent.
 662  *
 663  * the head node for delayed ref is used to store the sum of all the
 664  * reference count modifications queued up in the rbtree. the head
 665  * node may also store the extent flags to set. This way you can check
 666  * to see what the reference count and extent flags would be if all of
 667  * the delayed refs are not processed.
 668  */
 669 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 670                              struct btrfs_root *root, u64 bytenr,
 671                              u64 num_bytes, u64 *refs, u64 *flags)
 672 {
 673         struct btrfs_delayed_ref_head *head;
 674         struct btrfs_delayed_ref_root *delayed_refs;
 675         struct btrfs_path *path;
 676         struct btrfs_extent_item *ei;
 677         struct extent_buffer *leaf;
 678         struct btrfs_key key;
 679         u32 item_size;
 680         u64 num_refs;
 681         u64 extent_flags;
 682         int ret;
 683
 684         path = btrfs_alloc_path();
 685         if (!path)
 686                 return -ENOMEM;
 687
 688         key.objectid = bytenr;
 689         key.type = BTRFS_EXTENT_ITEM_KEY;
 690         key.offset = num_bytes;
 691         if (!trans) {
 692                 path->skip_locking = 1;
 693                 path->search_commit_root = 1;
 694         }
 695 again:
 696         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 697                                 &key, path, 0, 0);
 698         if (ret < 0)
 699                 goto out_free;
 700
 701         if (ret == 0) {
 702                 leaf = path->nodes[0];
 703                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 704                 if (item_size >= sizeof(*ei)) {
 705                         ei = btrfs_item_ptr(leaf, path->slots[0],
 706                                             struct btrfs_extent_item);
 707                         num_refs = btrfs_extent_refs(leaf, ei);
 708                         extent_flags = btrfs_extent_flags(leaf, ei);
 709                 } else {
 710 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 711                         struct btrfs_extent_item_v0 *ei0;
 712                         BUG_ON(item_size != sizeof(*ei0));
 713                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 714                                              struct btrfs_extent_item_v0);
 715                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 716                         /* FIXME: this isn't correct for data */
 717                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 718 #else
 719                         BUG();
 720 #endif
 721                 }
 722                 BUG_ON(num_refs == 0);
 723         } else {
 724                 num_refs = 0;
 725                 extent_flags = 0;
 726                 ret = 0;
 727         }
 728
 729         if (!trans)
 730                 goto out;
 731
 732         delayed_refs = &trans->transaction->delayed_refs;
 733         spin_lock(&delayed_refs->lock);
 734         head = btrfs_find_delayed_ref_head(trans, bytenr);
 735         if (head) {
 736                 if (!mutex_trylock(&head->mutex)) {
 737                         atomic_inc(&head->node.refs);
 738                         spin_unlock(&delayed_refs->lock);
 739
 740                         btrfs_release_path(root->fs_info->extent_root, path);
 741
 742                         mutex_lock(&head->mutex);
 743                         mutex_unlock(&head->mutex);
 744                         btrfs_put_delayed_ref(&head->node);
 745                         goto again;
 746                 }
 747                 if (head->extent_op && head->extent_op->update_flags)
 748                         extent_flags |= head->extent_op->flags_to_set;
 749                 else
 750                         BUG_ON(num_refs == 0);
 751
 752                 num_refs += head->node.ref_mod;
 753                 mutex_unlock(&head->mutex);
 754         }
 755         spin_unlock(&delayed_refs->lock);
 756 out:
 757         WARN_ON(num_refs == 0);
 758         if (refs)
 759                 *refs = num_refs;
 760         if (flags)
 761                 *flags = extent_flags;
 762 out_free:
 763         btrfs_free_path(path);
 764         return ret;
 765 }
 766
 767 /*
 768  * Back reference rules.  Back refs have three main goals:
 769  *
 770  * 1) differentiate between all holders of references to an extent so that
 771  *    when a reference is dropped we can make sure it was a valid reference
 772  *    before freeing the extent.
 773  *
 774  * 2) Provide enough information to quickly find the holders of an extent
 775  *    if we notice a given block is corrupted or bad.
 776  *
 777  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 778  *    maintenance.  This is actually the same as #2, but with a slightly
 779  *    different use case.
 780  *
 781  * There are two kinds of back refs. The implicit back refs is optimized
 782  * for pointers in non-shared tree blocks. For a given pointer in a block,
 783  * back refs of this kind provide information about the block's owner tree
 784  * and the pointer's key. These information allow us to find the block by
 785  * b-tree searching. The full back refs is for pointers in tree blocks not
 786  * referenced by their owner trees. The location of tree block is recorded
 787  * in the back refs. Actually the full back refs is generic, and can be
 788  * used in all cases the implicit back refs is used. The major shortcoming
 789  * of the full back refs is its overhead. Every time a tree block gets
 790  * COWed, we have to update back refs entry for all pointers in it.
 791  *
 792  * For a newly allocated tree block, we use implicit back refs for
 793  * pointers in it. This means most tree related operations only involve
 794  * implicit back refs. For a tree block created in old transaction, the
 795  * only way to drop a reference to it is COW it. So we can detect the
 796  * event that tree block loses its owner tree's reference and do the
 797  * back refs conversion.
 798  *
 799  * When a tree block is COW'd through a tree, there are four cases:
 800  *
 801  * The reference count of the block is one and the tree is the block's
 802  * owner tree. Nothing to do in this case.
 803  *
 804  * The reference count of the block is one and the tree is not the
 805  * block's owner tree. In this case, full back refs is used for pointers
 806  * in the block. Remove these full back refs, add implicit back refs for
 807  * every pointers in the new block.
 808  *
 809  * The reference count of the block is greater than one and the tree is
 810  * the block's owner tree. In this case, implicit back refs is used for
 811  * pointers in the block. Add full back refs for every pointers in the
 812  * block, increase lower level extents' reference counts. The original
 813  * implicit back refs are entailed to the new block.
 814  *
 815  * The reference count of the block is greater than one and the tree is
 816  * not the block's owner tree. Add implicit back refs for every pointer in
 817  * the new block, increase lower level extents' reference count.
 818  *
 819  * Back Reference Key composing:
 820  *
 821  * The key objectid corresponds to the first byte in the extent,
 822  * The key type is used to differentiate between types of back refs.
 823  * There are different meanings of the key offset for different types
 824  * of back refs.
 825  *
 826  * File extents can be referenced by:
 827  *
 828  * - multiple snapshots, subvolumes, or different generations in one subvol
 829  * - different files inside a single subvolume
 830  * - different offsets inside a file (bookend extents in file.c)
 831  *
 832  * The extent ref structure for the implicit back refs has fields for:
 833  *
 834  * - Objectid of the subvolume root
 835  * - objectid of the file holding the reference
 836  * - original offset in the file
 837  * - how many bookend extents
 838  *
 839  * The key offset for the implicit back refs is hash of the first
 840  * three fields.
 841  *
 842  * The extent ref structure for the full back refs has field for:
 843  *
 844  * - number of pointers in the tree leaf
 845  *
 846  * The key offset for the implicit back refs is the first byte of
 847  * the tree leaf
 848  *
 849  * When a file extent is allocated, The implicit back refs is used.
 850  * the fields are filled in:
 851  *
 852  *     (root_key.objectid, inode objectid, offset in file, 1)
 853  *
 854  * When a file extent is removed file truncation, we find the
 855  * corresponding implicit back refs and check the following fields:
 856  *
 857  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 858  *
 859  * Btree extents can be referenced by:
 860  *
 861  * - Different subvolumes
 862  *
 863  * Both the implicit back refs and the full back refs for tree blocks
 864  * only consist of key. The key offset for the implicit back refs is
 865  * objectid of block's owner tree. The key offset for the full back refs
 866  * is the first byte of parent block.
 867  *
 868  * When implicit back refs is used, information about the lowest key and
 869  * level of the tree block are required. These information are stored in
 870  * tree block info structure.
 871  */
 872
 873 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 874 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 875                                   struct btrfs_root *root,
 876                                   struct btrfs_path *path,
 877                                   u64 owner, u32 extra_size)
 878 {
 879         struct btrfs_extent_item *item;
 880         struct btrfs_extent_item_v0 *ei0;
 881         struct btrfs_extent_ref_v0 *ref0;
 882         struct btrfs_tree_block_info *bi;
 883         struct extent_buffer *leaf;
 884         struct btrfs_key key;
 885         struct btrfs_key found_key;
 886         u32 new_size = sizeof(*item);
 887         u64 refs;
 888         int ret;
 889
 890         leaf = path->nodes[0];
 891         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 892
 893         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 894         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 895                              struct btrfs_extent_item_v0);
 896         refs = btrfs_extent_refs_v0(leaf, ei0);
 897
 898         if (owner == (u64)-1) {
 899                 while (1) {
 900                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 901                                 ret = btrfs_next_leaf(root, path);
 902                                 if (ret < 0)
 903                                         return ret;
 904                                 BUG_ON(ret > 0);
 905                                 leaf = path->nodes[0];
 906                         }
 907                         btrfs_item_key_to_cpu(leaf, &found_key,
 908                                               path->slots[0]);
 909                         BUG_ON(key.objectid != found_key.objectid);
 910                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 911                                 path->slots[0]++;
 912                                 continue;
 913                         }
 914                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 915                                               struct btrfs_extent_ref_v0);
 916                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 917                         break;
 918                 }
 919         }
 920         btrfs_release_path(root, path);
 921
 922         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 923                 new_size += sizeof(*bi);
 924
 925         new_size -= sizeof(*ei0);
 926         ret = btrfs_search_slot(trans, root, &key, path,
 927                                 new_size + extra_size, 1);
 928         if (ret < 0)
 929                 return ret;
 930         BUG_ON(ret);
 931
 932         ret = btrfs_extend_item(trans, root, path, new_size);
 933         BUG_ON(ret);
 934
 935         leaf = path->nodes[0];
 936         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 937         btrfs_set_extent_refs(leaf, item, refs);
 938         /* FIXME: get real generation */
 939         btrfs_set_extent_generation(leaf, item, 0);
 940         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 941                 btrfs_set_extent_flags(leaf, item,
 942                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 943                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 944                 bi = (struct btrfs_tree_block_info *)(item + 1);
 945                 /* FIXME: get first key of the block */
 946                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 947                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 948         } else {
 949                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 950         }
 951         btrfs_mark_buffer_dirty(leaf);
 952         return 0;
 953 }
 954 #endif
 955
 956 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 957 {
 958         u32 high_crc = ~(u32)0;
 959         u32 low_crc = ~(u32)0;
 960         __le64 lenum;
 961
 962         lenum = cpu_to_le64(root_objectid);
 963         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 964         lenum = cpu_to_le64(owner);
 965         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 966         lenum = cpu_to_le64(offset);
 967         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 968
 969         return ((u64)high_crc << 31) ^ (u64)low_crc;
 970 }
 971
 972 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 973                                      struct btrfs_extent_data_ref *ref)
 974 {
 975         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 976                                     btrfs_extent_data_ref_objectid(leaf, ref),
 977                                     btrfs_extent_data_ref_offset(leaf, ref));
 978 }
 979
 980 static int match_extent_data_ref(struct extent_buffer *leaf,
 981                                  struct btrfs_extent_data_ref *ref,
 982                                  u64 root_objectid, u64 owner, u64 offset)
 983 {
 984         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 985             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 986             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 987                 return 0;
 988         return 1;
 989 }
 990
 991 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 992                                            struct btrfs_root *root,
 993                                            struct btrfs_path *path,
 994                                            u64 bytenr, u64 parent,
 995                                            u64 root_objectid,
 996                                            u64 owner, u64 offset)
 997 {
 998         struct btrfs_key key;
 999         struct btrfs_extent_data_ref *ref;
1000         struct extent_buffer *leaf;
1001         u32 nritems;
1002         int ret;
1003         int recow;
1004         int err = -ENOENT;
1005
1006         key.objectid = bytenr;
1007         if (parent) {
1008                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1009                 key.offset = parent;
1010         } else {
1011                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1012                 key.offset = hash_extent_data_ref(root_objectid,
1013                                                   owner, offset);
1014         }
1015 again:
1016         recow = 0;
1017         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1018         if (ret < 0) {
1019                 err = ret;
1020                 goto fail;
1021         }
1022
1023         if (parent) {
1024                 if (!ret)
1025                         return 0;
1026 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1027                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1028                 btrfs_release_path(root, path);
1029                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1030                 if (ret < 0) {
1031                         err = ret;
1032                         goto fail;
1033                 }
1034                 if (!ret)
1035                         return 0;
1036 #endif
1037                 goto fail;
1038         }
1039
1040         leaf = path->nodes[0];
1041         nritems = btrfs_header_nritems(leaf);
1042         while (1) {
1043                 if (path->slots[0] >= nritems) {
1044                         ret = btrfs_next_leaf(root, path);
1045                         if (ret < 0)
1046                                 err = ret;
1047                         if (ret)
1048                                 goto fail;
1049
1050                         leaf = path->nodes[0];
1051                         nritems = btrfs_header_nritems(leaf);
1052                         recow = 1;
1053                 }
1054
1055                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1056                 if (key.objectid != bytenr ||
1057                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1058                         goto fail;
1059
1060                 ref = btrfs_item_ptr(leaf, path->slots[0],
1061                                      struct btrfs_extent_data_ref);
1062
1063                 if (match_extent_data_ref(leaf, ref, root_objectid,
1064                                           owner, offset)) {
1065                         if (recow) {
1066                                 btrfs_release_path(root, path);
1067                                 goto again;
1068                         }
1069                         err = 0;
1070                         break;
1071                 }
1072                 path->slots[0]++;
1073         }
1074 fail:
1075         return err;
1076 }
1077
1078 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1079                                            struct btrfs_root *root,
1080                                            struct btrfs_path *path,
1081                                            u64 bytenr, u64 parent,
1082                                            u64 root_objectid, u64 owner,
1083                                            u64 offset, int refs_to_add)
1084 {
1085         struct btrfs_key key;
1086         struct extent_buffer *leaf;
1087         u32 size;
1088         u32 num_refs;
1089         int ret;
1090
1091         key.objectid = bytenr;
1092         if (parent) {
1093                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1094                 key.offset = parent;
1095                 size = sizeof(struct btrfs_shared_data_ref);
1096         } else {
1097                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1098                 key.offset = hash_extent_data_ref(root_objectid,
1099                                                   owner, offset);
1100                 size = sizeof(struct btrfs_extent_data_ref);
1101         }
1102
1103         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1104         if (ret && ret != -EEXIST)
1105                 goto fail;
1106
1107         leaf = path->nodes[0];
1108         if (parent) {
1109                 struct btrfs_shared_data_ref *ref;
1110                 ref = btrfs_item_ptr(leaf, path->slots[0],
1111                                      struct btrfs_shared_data_ref);
1112                 if (ret == 0) {
1113                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1114                 } else {
1115                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1116                         num_refs += refs_to_add;
1117                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1118                 }
1119         } else {
1120                 struct btrfs_extent_data_ref *ref;
1121                 while (ret == -EEXIST) {
1122                         ref = btrfs_item_ptr(leaf, path->slots[0],
1123                                              struct btrfs_extent_data_ref);
1124                         if (match_extent_data_ref(leaf, ref, root_objectid,
1125                                                   owner, offset))
1126                                 break;
1127                         btrfs_release_path(root, path);
1128                         key.offset++;
1129                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1130                                                       size);
1131                         if (ret && ret != -EEXIST)
1132                                 goto fail;
1133
1134                         leaf = path->nodes[0];
1135                 }
1136                 ref = btrfs_item_ptr(leaf, path->slots[0],
1137                                      struct btrfs_extent_data_ref);
1138                 if (ret == 0) {
1139                         btrfs_set_extent_data_ref_root(leaf, ref,
1140                                                        root_objectid);
1141                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1142                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1143                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1144                 } else {
1145                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1146                         num_refs += refs_to_add;
1147                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1148                 }
1149         }
1150         btrfs_mark_buffer_dirty(leaf);
1151         ret = 0;
1152 fail:
1153         btrfs_release_path(root, path);
1154         return ret;
1155 }
1156
1157 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1158                                            struct btrfs_root *root,
1159                                            struct btrfs_path *path,
1160                                            int refs_to_drop)
1161 {
1162         struct btrfs_key key;
1163         struct btrfs_extent_data_ref *ref1 = NULL;
1164         struct btrfs_shared_data_ref *ref2 = NULL;
1165         struct extent_buffer *leaf;
1166         u32 num_refs = 0;
1167         int ret = 0;
1168
1169         leaf = path->nodes[0];
1170         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1171
1172         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1173                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1174                                       struct btrfs_extent_data_ref);
1175                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1176         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1177                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1178                                       struct btrfs_shared_data_ref);
1179                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1180 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1181         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1182                 struct btrfs_extent_ref_v0 *ref0;
1183                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1184                                       struct btrfs_extent_ref_v0);
1185                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1186 #endif
1187         } else {
1188                 BUG();
1189         }
1190
1191         BUG_ON(num_refs < refs_to_drop);
1192         num_refs -= refs_to_drop;
1193
1194         if (num_refs == 0) {
1195                 ret = btrfs_del_item(trans, root, path);
1196         } else {
1197                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1198                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1199                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1200                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1201 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1202                 else {
1203                         struct btrfs_extent_ref_v0 *ref0;
1204                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1205                                         struct btrfs_extent_ref_v0);
1206                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1207                 }
1208 #endif
1209                 btrfs_mark_buffer_dirty(leaf);
1210         }
1211         return ret;
1212 }
1213
1214 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1215                                           struct btrfs_path *path,
1216                                           struct btrfs_extent_inline_ref *iref)
1217 {
1218         struct btrfs_key key;
1219         struct extent_buffer *leaf;
1220         struct btrfs_extent_data_ref *ref1;
1221         struct btrfs_shared_data_ref *ref2;
1222         u32 num_refs = 0;
1223
1224         leaf = path->nodes[0];
1225         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1226         if (iref) {
1227                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1228                     BTRFS_EXTENT_DATA_REF_KEY) {
1229                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1230                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1231                 } else {
1232                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1233                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1234                 }
1235         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1236                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1237                                       struct btrfs_extent_data_ref);
1238                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1239         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1240                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1241                                       struct btrfs_shared_data_ref);
1242                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1243 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1244         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1245                 struct btrfs_extent_ref_v0 *ref0;
1246                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1247                                       struct btrfs_extent_ref_v0);
1248                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1249 #endif
1250         } else {
1251                 WARN_ON(1);
1252         }
1253         return num_refs;
1254 }
1255
1256 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1257                                           struct btrfs_root *root,
1258                                           struct btrfs_path *path,
1259                                           u64 bytenr, u64 parent,
1260                                           u64 root_objectid)
1261 {
1262         struct btrfs_key key;
1263         int ret;
1264
1265         key.objectid = bytenr;
1266         if (parent) {
1267                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1268                 key.offset = parent;
1269         } else {
1270                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1271                 key.offset = root_objectid;
1272         }
1273
1274         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1275         if (ret > 0)
1276                 ret = -ENOENT;
1277 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1278         if (ret == -ENOENT && parent) {
1279                 btrfs_release_path(root, path);
1280                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1281                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1282                 if (ret > 0)
1283                         ret = -ENOENT;
1284         }
1285 #endif
1286         return ret;
1287 }
1288
1289 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1290                                           struct btrfs_root *root,
1291                                           struct btrfs_path *path,
1292                                           u64 bytenr, u64 parent,
1293                                           u64 root_objectid)
1294 {
1295         struct btrfs_key key;
1296         int ret;
1297
1298         key.objectid = bytenr;
1299         if (parent) {
1300                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1301                 key.offset = parent;
1302         } else {
1303                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1304                 key.offset = root_objectid;
1305         }
1306
1307         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1308         btrfs_release_path(root, path);
1309         return ret;
1310 }
1311
1312 static inline int extent_ref_type(u64 parent, u64 owner)
1313 {
1314         int type;
1315         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1316                 if (parent > 0)
1317                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1318                 else
1319                         type = BTRFS_TREE_BLOCK_REF_KEY;
1320         } else {
1321                 if (parent > 0)
1322                         type = BTRFS_SHARED_DATA_REF_KEY;
1323                 else
1324                         type = BTRFS_EXTENT_DATA_REF_KEY;
1325         }
1326         return type;
1327 }
1328
1329 static int find_next_key(struct btrfs_path *path, int level,
1330                          struct btrfs_key *key)
1331
1332 {
1333         for (; level < BTRFS_MAX_LEVEL; level++) {
1334                 if (!path->nodes[level])
1335                         break;
1336                 if (path->slots[level] + 1 >=
1337                     btrfs_header_nritems(path->nodes[level]))
1338                         continue;
1339                 if (level == 0)
1340                         btrfs_item_key_to_cpu(path->nodes[level], key,
1341                                               path->slots[level] + 1);
1342                 else
1343                         btrfs_node_key_to_cpu(path->nodes[level], key,
1344                                               path->slots[level] + 1);
1345                 return 0;
1346         }
1347         return 1;
1348 }
1349
1350 /*
1351  * look for inline back ref. if back ref is found, *ref_ret is set
1352  * to the address of inline back ref, and 0 is returned.
1353  *
1354  * if back ref isn't found, *ref_ret is set to the address where it
1355  * should be inserted, and -ENOENT is returned.
1356  *
1357  * if insert is true and there are too many inline back refs, the path
1358  * points to the extent item, and -EAGAIN is returned.
1359  *
1360  * NOTE: inline back refs are ordered in the same way that back ref
1361  *       items in the tree are ordered.
1362  */
1363 static noinline_for_stack
1364 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1365                                  struct btrfs_root *root,
1366                                  struct btrfs_path *path,
1367                                  struct btrfs_extent_inline_ref **ref_ret,
1368                                  u64 bytenr, u64 num_bytes,
1369                                  u64 parent, u64 root_objectid,
1370                                  u64 owner, u64 offset, int insert)
1371 {
1372         struct btrfs_key key;
1373         struct extent_buffer *leaf;
1374         struct btrfs_extent_item *ei;
1375         struct btrfs_extent_inline_ref *iref;
1376         u64 flags;
1377         u64 item_size;
1378         unsigned long ptr;
1379         unsigned long end;
1380         int extra_size;
1381         int type;
1382         int want;
1383         int ret;
1384         int err = 0;
1385
1386         key.objectid = bytenr;
1387         key.type = BTRFS_EXTENT_ITEM_KEY;
1388         key.offset = num_bytes;
1389
1390         want = extent_ref_type(parent, owner);
1391         if (insert) {
1392                 extra_size = btrfs_extent_inline_ref_size(want);
1393                 path->keep_locks = 1;
1394         } else
1395                 extra_size = -1;
1396         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1397         if (ret < 0) {
1398                 err = ret;
1399                 goto out;
1400         }
1401         BUG_ON(ret);
1402
1403         leaf = path->nodes[0];
1404         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1405 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1406         if (item_size < sizeof(*ei)) {
1407                 if (!insert) {
1408                         err = -ENOENT;
1409                         goto out;
1410                 }
1411                 ret = convert_extent_item_v0(trans, root, path, owner,
1412                                              extra_size);
1413                 if (ret < 0) {
1414                         err = ret;
1415                         goto out;
1416                 }
1417                 leaf = path->nodes[0];
1418                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1419         }
1420 #endif
1421         BUG_ON(item_size < sizeof(*ei));
1422
1423         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1424         flags = btrfs_extent_flags(leaf, ei);
1425
1426         ptr = (unsigned long)(ei + 1);
1427         end = (unsigned long)ei + item_size;
1428
1429         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1430                 ptr += sizeof(struct btrfs_tree_block_info);
1431                 BUG_ON(ptr > end);
1432         } else {
1433                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1434         }
1435
1436         err = -ENOENT;
1437         while (1) {
1438                 if (ptr >= end) {
1439                         WARN_ON(ptr > end);
1440                         break;
1441                 }
1442                 iref = (struct btrfs_extent_inline_ref *)ptr;
1443                 type = btrfs_extent_inline_ref_type(leaf, iref);
1444                 if (want < type)
1445                         break;
1446                 if (want > type) {
1447                         ptr += btrfs_extent_inline_ref_size(type);
1448                         continue;
1449                 }
1450
1451                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1452                         struct btrfs_extent_data_ref *dref;
1453                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1454                         if (match_extent_data_ref(leaf, dref, root_objectid,
1455                                                   owner, offset)) {
1456                                 err = 0;
1457                                 break;
1458                         }
1459                         if (hash_extent_data_ref_item(leaf, dref) <
1460                             hash_extent_data_ref(root_objectid, owner, offset))
1461                                 break;
1462                 } else {
1463                         u64 ref_offset;
1464                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1465                         if (parent > 0) {
1466                                 if (parent == ref_offset) {
1467                                         err = 0;
1468                                         break;
1469                                 }
1470                                 if (ref_offset < parent)
1471                                         break;
1472                         } else {
1473                                 if (root_objectid == ref_offset) {
1474                                         err = 0;
1475                                         break;
1476                                 }
1477                                 if (ref_offset < root_objectid)
1478                                         break;
1479                         }
1480                 }
1481                 ptr += btrfs_extent_inline_ref_size(type);
1482         }
1483         if (err == -ENOENT && insert) {
1484                 if (item_size + extra_size >=
1485                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1486                         err = -EAGAIN;
1487                         goto out;
1488                 }
1489                 /*
1490                  * To add new inline back ref, we have to make sure
1491                  * there is no corresponding back ref item.
1492                  * For simplicity, we just do not add new inline back
1493                  * ref if there is any kind of item for this block
1494                  */
1495                 if (find_next_key(path, 0, &key) == 0 &&
1496                     key.objectid == bytenr &&
1497                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1498                         err = -EAGAIN;
1499                         goto out;
1500                 }
1501         }
1502         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1503 out:
1504         if (insert) {
1505                 path->keep_locks = 0;
1506                 btrfs_unlock_up_safe(path, 1);
1507         }
1508         return err;
1509 }
1510
1511 /*
1512  * helper to add new inline back ref
1513  */
1514 static noinline_for_stack
1515 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1516                                 struct btrfs_root *root,
1517                                 struct btrfs_path *path,
1518                                 struct btrfs_extent_inline_ref *iref,
1519                                 u64 parent, u64 root_objectid,
1520                                 u64 owner, u64 offset, int refs_to_add,
1521                                 struct btrfs_delayed_extent_op *extent_op)
1522 {
1523         struct extent_buffer *leaf;
1524         struct btrfs_extent_item *ei;
1525         unsigned long ptr;
1526         unsigned long end;
1527         unsigned long item_offset;
1528         u64 refs;
1529         int size;
1530         int type;
1531         int ret;
1532
1533         leaf = path->nodes[0];
1534         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1535         item_offset = (unsigned long)iref - (unsigned long)ei;
1536
1537         type = extent_ref_type(parent, owner);
1538         size = btrfs_extent_inline_ref_size(type);
1539
1540         ret = btrfs_extend_item(trans, root, path, size);
1541         BUG_ON(ret);
1542
1543         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1544         refs = btrfs_extent_refs(leaf, ei);
1545         refs += refs_to_add;
1546         btrfs_set_extent_refs(leaf, ei, refs);
1547         if (extent_op)
1548                 __run_delayed_extent_op(extent_op, leaf, ei);
1549
1550         ptr = (unsigned long)ei + item_offset;
1551         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1552         if (ptr < end - size)
1553                 memmove_extent_buffer(leaf, ptr + size, ptr,
1554                                       end - size - ptr);
1555
1556         iref = (struct btrfs_extent_inline_ref *)ptr;
1557         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1558         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1559                 struct btrfs_extent_data_ref *dref;
1560                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1561                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1562                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1563                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1564                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1565         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1566                 struct btrfs_shared_data_ref *sref;
1567                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1568                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1569                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1570         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1571                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1572         } else {
1573                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1574         }
1575         btrfs_mark_buffer_dirty(leaf);
1576         return 0;
1577 }
1578
1579 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1580                                  struct btrfs_root *root,
1581                                  struct btrfs_path *path,
1582                                  struct btrfs_extent_inline_ref **ref_ret,
1583                                  u64 bytenr, u64 num_bytes, u64 parent,
1584                                  u64 root_objectid, u64 owner, u64 offset)
1585 {
1586         int ret;
1587
1588         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1589                                            bytenr, num_bytes, parent,
1590                                            root_objectid, owner, offset, 0);
1591         if (ret != -ENOENT)
1592                 return ret;
1593
1594         btrfs_release_path(root, path);
1595         *ref_ret = NULL;
1596
1597         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1598                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1599                                             root_objectid);
1600         } else {
1601                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1602                                              root_objectid, owner, offset);
1603         }
1604         return ret;
1605 }
1606
1607 /*
1608  * helper to update/remove inline back ref
1609  */
1610 static noinline_for_stack
1611 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1612                                  struct btrfs_root *root,
1613                                  struct btrfs_path *path,
1614                                  struct btrfs_extent_inline_ref *iref,
1615                                  int refs_to_mod,
1616                                  struct btrfs_delayed_extent_op *extent_op)
1617 {
1618         struct extent_buffer *leaf;
1619         struct btrfs_extent_item *ei;
1620         struct btrfs_extent_data_ref *dref = NULL;
1621         struct btrfs_shared_data_ref *sref = NULL;
1622         unsigned long ptr;
1623         unsigned long end;
1624         u32 item_size;
1625         int size;
1626         int type;
1627         int ret;
1628         u64 refs;
1629
1630         leaf = path->nodes[0];
1631         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1632         refs = btrfs_extent_refs(leaf, ei);
1633         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1634         refs += refs_to_mod;
1635         btrfs_set_extent_refs(leaf, ei, refs);
1636         if (extent_op)
1637                 __run_delayed_extent_op(extent_op, leaf, ei);
1638
1639         type = btrfs_extent_inline_ref_type(leaf, iref);
1640
1641         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1642                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1643                 refs = btrfs_extent_data_ref_count(leaf, dref);
1644         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1645                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1646                 refs = btrfs_shared_data_ref_count(leaf, sref);
1647         } else {
1648                 refs = 1;
1649                 BUG_ON(refs_to_mod != -1);
1650         }
1651
1652         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1653         refs += refs_to_mod;
1654
1655         if (refs > 0) {
1656                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1657                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1658                 else
1659                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1660         } else {
1661                 size =  btrfs_extent_inline_ref_size(type);
1662                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1663                 ptr = (unsigned long)iref;
1664                 end = (unsigned long)ei + item_size;
1665                 if (ptr + size < end)
1666                         memmove_extent_buffer(leaf, ptr, ptr + size,
1667                                               end - ptr - size);
1668                 item_size -= size;
1669                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1670                 BUG_ON(ret);
1671         }
1672         btrfs_mark_buffer_dirty(leaf);
1673         return 0;
1674 }
1675
1676 static noinline_for_stack
1677 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1678                                  struct btrfs_root *root,
1679                                  struct btrfs_path *path,
1680                                  u64 bytenr, u64 num_bytes, u64 parent,
1681                                  u64 root_objectid, u64 owner,
1682                                  u64 offset, int refs_to_add,
1683                                  struct btrfs_delayed_extent_op *extent_op)
1684 {
1685         struct btrfs_extent_inline_ref *iref;
1686         int ret;
1687
1688         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1689                                            bytenr, num_bytes, parent,
1690                                            root_objectid, owner, offset, 1);
1691         if (ret == 0) {
1692                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1693                 ret = update_inline_extent_backref(trans, root, path, iref,
1694                                                    refs_to_add, extent_op);
1695         } else if (ret == -ENOENT) {
1696                 ret = setup_inline_extent_backref(trans, root, path, iref,
1697                                                   parent, root_objectid,
1698                                                   owner, offset, refs_to_add,
1699                                                   extent_op);
1700         }
1701         return ret;
1702 }
1703
1704 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1705                                  struct btrfs_root *root,
1706                                  struct btrfs_path *path,
1707                                  u64 bytenr, u64 parent, u64 root_objectid,
1708                                  u64 owner, u64 offset, int refs_to_add)
1709 {
1710         int ret;
1711         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1712                 BUG_ON(refs_to_add != 1);
1713                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1714                                             parent, root_objectid);
1715         } else {
1716                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1717                                              parent, root_objectid,
1718                                              owner, offset, refs_to_add);
1719         }
1720         return ret;
1721 }
1722
1723 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1724                                  struct btrfs_root *root,
1725                                  struct btrfs_path *path,
1726                                  struct btrfs_extent_inline_ref *iref,
1727                                  int refs_to_drop, int is_data)
1728 {
1729         int ret;
1730
1731         BUG_ON(!is_data && refs_to_drop != 1);
1732         if (iref) {
1733                 ret = update_inline_extent_backref(trans, root, path, iref,
1734                                                    -refs_to_drop, NULL);
1735         } else if (is_data) {
1736                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1737         } else {
1738                 ret = btrfs_del_item(trans, root, path);
1739         }
1740         return ret;
1741 }
1742
1743 static void btrfs_issue_discard(struct block_device *bdev,
1744                                 u64 start, u64 len)
1745 {
1746         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL,
1747                         BLKDEV_IFL_WAIT | BLKDEV_IFL_BARRIER);
1748 }
1749
1750 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1751                                 u64 num_bytes)
1752 {
1753         int ret;
1754         u64 map_length = num_bytes;
1755         struct btrfs_multi_bio *multi = NULL;
1756
1757         if (!btrfs_test_opt(root, DISCARD))
1758                 return 0;
1759
1760         /* Tell the block device(s) that the sectors can be discarded */
1761         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1762                               bytenr, &map_length, &multi, 0);
1763         if (!ret) {
1764                 struct btrfs_bio_stripe *stripe = multi->stripes;
1765                 int i;
1766
1767                 if (map_length > num_bytes)
1768                         map_length = num_bytes;
1769
1770                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1771                         btrfs_issue_discard(stripe->dev->bdev,
1772                                             stripe->physical,
1773                                             map_length);
1774                 }
1775                 kfree(multi);
1776         }
1777
1778         return ret;
1779 }
1780
1781 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1782                          struct btrfs_root *root,
1783                          u64 bytenr, u64 num_bytes, u64 parent,
1784                          u64 root_objectid, u64 owner, u64 offset)
1785 {
1786         int ret;
1787         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1788                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1789
1790         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1791                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1792                                         parent, root_objectid, (int)owner,
1793                                         BTRFS_ADD_DELAYED_REF, NULL);
1794         } else {
1795                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1796                                         parent, root_objectid, owner, offset,
1797                                         BTRFS_ADD_DELAYED_REF, NULL);
1798         }
1799         return ret;
1800 }
1801
1802 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1803                                   struct btrfs_root *root,
1804                                   u64 bytenr, u64 num_bytes,
1805                                   u64 parent, u64 root_objectid,
1806                                   u64 owner, u64 offset, int refs_to_add,
1807                                   struct btrfs_delayed_extent_op *extent_op)
1808 {
1809         struct btrfs_path *path;
1810         struct extent_buffer *leaf;
1811         struct btrfs_extent_item *item;
1812         u64 refs;
1813         int ret;
1814         int err = 0;
1815
1816         path = btrfs_alloc_path();
1817         if (!path)
1818                 return -ENOMEM;
1819
1820         path->reada = 1;
1821         path->leave_spinning = 1;
1822         /* this will setup the path even if it fails to insert the back ref */
1823         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1824                                            path, bytenr, num_bytes, parent,
1825                                            root_objectid, owner, offset,
1826                                            refs_to_add, extent_op);
1827         if (ret == 0)
1828                 goto out;
1829
1830         if (ret != -EAGAIN) {
1831                 err = ret;
1832                 goto out;
1833         }
1834
1835         leaf = path->nodes[0];
1836         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1837         refs = btrfs_extent_refs(leaf, item);
1838         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1839         if (extent_op)
1840                 __run_delayed_extent_op(extent_op, leaf, item);
1841
1842         btrfs_mark_buffer_dirty(leaf);
1843         btrfs_release_path(root->fs_info->extent_root, path);
1844
1845         path->reada = 1;
1846         path->leave_spinning = 1;
1847
1848         /* now insert the actual backref */
1849         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1850                                     path, bytenr, parent, root_objectid,
1851                                     owner, offset, refs_to_add);
1852         BUG_ON(ret);
1853 out:
1854         btrfs_free_path(path);
1855         return err;
1856 }
1857
1858 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1859                                 struct btrfs_root *root,
1860                                 struct btrfs_delayed_ref_node *node,
1861                                 struct btrfs_delayed_extent_op *extent_op,
1862                                 int insert_reserved)
1863 {
1864         int ret = 0;
1865         struct btrfs_delayed_data_ref *ref;
1866         struct btrfs_key ins;
1867         u64 parent = 0;
1868         u64 ref_root = 0;
1869         u64 flags = 0;
1870
1871         ins.objectid = node->bytenr;
1872         ins.offset = node->num_bytes;
1873         ins.type = BTRFS_EXTENT_ITEM_KEY;
1874
1875         ref = btrfs_delayed_node_to_data_ref(node);
1876         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1877                 parent = ref->parent;
1878         else
1879                 ref_root = ref->root;
1880
1881         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1882                 if (extent_op) {
1883                         BUG_ON(extent_op->update_key);
1884                         flags |= extent_op->flags_to_set;
1885                 }
1886                 ret = alloc_reserved_file_extent(trans, root,
1887                                                  parent, ref_root, flags,
1888                                                  ref->objectid, ref->offset,
1889                                                  &ins, node->ref_mod);
1890         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1891                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1892                                              node->num_bytes, parent,
1893                                              ref_root, ref->objectid,
1894                                              ref->offset, node->ref_mod,
1895                                              extent_op);
1896         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1897                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1898                                           node->num_bytes, parent,
1899                                           ref_root, ref->objectid,
1900                                           ref->offset, node->ref_mod,
1901                                           extent_op);
1902         } else {
1903                 BUG();
1904         }
1905         return ret;
1906 }
1907
1908 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1909                                     struct extent_buffer *leaf,
1910                                     struct btrfs_extent_item *ei)
1911 {
1912         u64 flags = btrfs_extent_flags(leaf, ei);
1913         if (extent_op->update_flags) {
1914                 flags |= extent_op->flags_to_set;
1915                 btrfs_set_extent_flags(leaf, ei, flags);
1916         }
1917
1918         if (extent_op->update_key) {
1919                 struct btrfs_tree_block_info *bi;
1920                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1921                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1922                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1923         }
1924 }
1925
1926 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1927                                  struct btrfs_root *root,
1928                                  struct btrfs_delayed_ref_node *node,
1929                                  struct btrfs_delayed_extent_op *extent_op)
1930 {
1931         struct btrfs_key key;
1932         struct btrfs_path *path;
1933         struct btrfs_extent_item *ei;
1934         struct extent_buffer *leaf;
1935         u32 item_size;
1936         int ret;
1937         int err = 0;
1938
1939         path = btrfs_alloc_path();
1940         if (!path)
1941                 return -ENOMEM;
1942
1943         key.objectid = node->bytenr;
1944         key.type = BTRFS_EXTENT_ITEM_KEY;
1945         key.offset = node->num_bytes;
1946
1947         path->reada = 1;
1948         path->leave_spinning = 1;
1949         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1950                                 path, 0, 1);
1951         if (ret < 0) {
1952                 err = ret;
1953                 goto out;
1954         }
1955         if (ret > 0) {
1956                 err = -EIO;
1957                 goto out;
1958         }
1959
1960         leaf = path->nodes[0];
1961         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1962 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1963         if (item_size < sizeof(*ei)) {
1964                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1965                                              path, (u64)-1, 0);
1966                 if (ret < 0) {
1967                         err = ret;
1968                         goto out;
1969                 }
1970                 leaf = path->nodes[0];
1971                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1972         }
1973 #endif
1974         BUG_ON(item_size < sizeof(*ei));
1975         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1976         __run_delayed_extent_op(extent_op, leaf, ei);
1977
1978         btrfs_mark_buffer_dirty(leaf);
1979 out:
1980         btrfs_free_path(path);
1981         return err;
1982 }
1983
1984 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1985                                 struct btrfs_root *root,
1986                                 struct btrfs_delayed_ref_node *node,
1987                                 struct btrfs_delayed_extent_op *extent_op,
1988                                 int insert_reserved)
1989 {
1990         int ret = 0;
1991         struct btrfs_delayed_tree_ref *ref;
1992         struct btrfs_key ins;
1993         u64 parent = 0;
1994         u64 ref_root = 0;
1995
1996         ins.objectid = node->bytenr;
1997         ins.offset = node->num_bytes;
1998         ins.type = BTRFS_EXTENT_ITEM_KEY;
1999
2000         ref = btrfs_delayed_node_to_tree_ref(node);
2001         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2002                 parent = ref->parent;
2003         else
2004                 ref_root = ref->root;
2005
2006         BUG_ON(node->ref_mod != 1);
2007         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2008                 BUG_ON(!extent_op || !extent_op->update_flags ||
2009                        !extent_op->update_key);
2010                 ret = alloc_reserved_tree_block(trans, root,
2011                                                 parent, ref_root,
2012                                                 extent_op->flags_to_set,
2013                                                 &extent_op->key,
2014                                                 ref->level, &ins);
2015         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2016                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2017                                              node->num_bytes, parent, ref_root,
2018                                              ref->level, 0, 1, extent_op);
2019         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2020                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2021                                           node->num_bytes, parent, ref_root,
2022                                           ref->level, 0, 1, extent_op);
2023         } else {
2024                 BUG();
2025         }
2026         return ret;
2027 }
2028
2029 /* helper function to actually process a single delayed ref entry */
2030 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2031                                struct btrfs_root *root,
2032                                struct btrfs_delayed_ref_node *node,
2033                                struct btrfs_delayed_extent_op *extent_op,
2034                                int insert_reserved)
2035 {
2036         int ret;
2037         if (btrfs_delayed_ref_is_head(node)) {
2038                 struct btrfs_delayed_ref_head *head;
2039                 /*
2040                  * we've hit the end of the chain and we were supposed
2041                  * to insert this extent into the tree.  But, it got
2042                  * deleted before we ever needed to insert it, so all
2043                  * we have to do is clean up the accounting
2044                  */
2045                 BUG_ON(extent_op);
2046                 head = btrfs_delayed_node_to_head(node);
2047                 if (insert_reserved) {
2048                         btrfs_pin_extent(root, node->bytenr,
2049                                          node->num_bytes, 1);
2050                         if (head->is_data) {
2051                                 ret = btrfs_del_csums(trans, root,
2052                                                       node->bytenr,
2053                                                       node->num_bytes);
2054                                 BUG_ON(ret);
2055                         }
2056                 }
2057                 mutex_unlock(&head->mutex);
2058                 return 0;
2059         }
2060
2061         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2062             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2063                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2064                                            insert_reserved);
2065         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2066                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2067                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2068                                            insert_reserved);
2069         else
2070                 BUG();
2071         return ret;
2072 }
2073
2074 static noinline struct btrfs_delayed_ref_node *
2075 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2076 {
2077         struct rb_node *node;
2078         struct btrfs_delayed_ref_node *ref;
2079         int action = BTRFS_ADD_DELAYED_REF;
2080 again:
2081         /*
2082          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2083          * this prevents ref count from going down to zero when
2084          * there still are pending delayed ref.
2085          */
2086         node = rb_prev(&head->node.rb_node);
2087         while (1) {
2088                 if (!node)
2089                         break;
2090                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2091                                 rb_node);
2092                 if (ref->bytenr != head->node.bytenr)
2093                         break;
2094                 if (ref->action == action)
2095                         return ref;
2096                 node = rb_prev(node);
2097         }
2098         if (action == BTRFS_ADD_DELAYED_REF) {
2099                 action = BTRFS_DROP_DELAYED_REF;
2100                 goto again;
2101         }
2102         return NULL;
2103 }
2104
2105 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2106                                        struct btrfs_root *root,
2107                                        struct list_head *cluster)
2108 {
2109         struct btrfs_delayed_ref_root *delayed_refs;
2110         struct btrfs_delayed_ref_node *ref;
2111         struct btrfs_delayed_ref_head *locked_ref = NULL;
2112         struct btrfs_delayed_extent_op *extent_op;
2113         int ret;
2114         int count = 0;
2115         int must_insert_reserved = 0;
2116
2117         delayed_refs = &trans->transaction->delayed_refs;
2118         while (1) {
2119                 if (!locked_ref) {
2120                         /* pick a new head ref from the cluster list */
2121                         if (list_empty(cluster))
2122                                 break;
2123
2124                         locked_ref = list_entry(cluster->next,
2125                                      struct btrfs_delayed_ref_head, cluster);
2126
2127                         /* grab the lock that says we are going to process
2128                          * all the refs for this head */
2129                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2130
2131                         /*
2132                          * we may have dropped the spin lock to get the head
2133                          * mutex lock, and that might have given someone else
2134                          * time to free the head.  If that's true, it has been
2135                          * removed from our list and we can move on.
2136                          */
2137                         if (ret == -EAGAIN) {
2138                                 locked_ref = NULL;
2139                                 count++;
2140                                 continue;
2141                         }
2142                 }
2143
2144                 /*
2145                  * record the must insert reserved flag before we
2146                  * drop the spin lock.
2147                  */
2148                 must_insert_reserved = locked_ref->must_insert_reserved;
2149                 locked_ref->must_insert_reserved = 0;
2150
2151                 extent_op = locked_ref->extent_op;
2152                 locked_ref->extent_op = NULL;
2153
2154                 /*
2155                  * locked_ref is the head node, so we have to go one
2156                  * node back for any delayed ref updates
2157                  */
2158                 ref = select_delayed_ref(locked_ref);
2159                 if (!ref) {
2160                         /* All delayed refs have been processed, Go ahead
2161                          * and send the head node to run_one_delayed_ref,
2162                          * so that any accounting fixes can happen
2163                          */
2164                         ref = &locked_ref->node;
2165
2166                         if (extent_op && must_insert_reserved) {
2167                                 kfree(extent_op);
2168                                 extent_op = NULL;
2169                         }
2170
2171                         if (extent_op) {
2172                                 spin_unlock(&delayed_refs->lock);
2173
2174                                 ret = run_delayed_extent_op(trans, root,
2175                                                             ref, extent_op);
2176                                 BUG_ON(ret);
2177                                 kfree(extent_op);
2178
2179                                 cond_resched();
2180                                 spin_lock(&delayed_refs->lock);
2181                                 continue;
2182                         }
2183
2184                         list_del_init(&locked_ref->cluster);
2185                         locked_ref = NULL;
2186                 }
2187
2188                 ref->in_tree = 0;
2189                 rb_erase(&ref->rb_node, &delayed_refs->root);
2190                 delayed_refs->num_entries--;
2191
2192                 spin_unlock(&delayed_refs->lock);
2193
2194                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2195                                           must_insert_reserved);
2196                 BUG_ON(ret);
2197
2198                 btrfs_put_delayed_ref(ref);
2199                 kfree(extent_op);
2200                 count++;
2201
2202                 cond_resched();
2203                 spin_lock(&delayed_refs->lock);
2204         }
2205         return count;
2206 }
2207
2208 /*
2209  * this starts processing the delayed reference count updates and
2210  * extent insertions we have queued up so far.  count can be
2211  * 0, which means to process everything in the tree at the start
2212  * of the run (but not newly added entries), or it can be some target
2213  * number you'd like to process.
2214  */
2215 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2216                            struct btrfs_root *root, unsigned long count)
2217 {
2218         struct rb_node *node;
2219         struct btrfs_delayed_ref_root *delayed_refs;
2220         struct btrfs_delayed_ref_node *ref;
2221         struct list_head cluster;
2222         int ret;
2223         int run_all = count == (unsigned long)-1;
2224         int run_most = 0;
2225
2226         if (root == root->fs_info->extent_root)
2227                 root = root->fs_info->tree_root;
2228
2229         delayed_refs = &trans->transaction->delayed_refs;
2230         INIT_LIST_HEAD(&cluster);
2231 again:
2232         spin_lock(&delayed_refs->lock);
2233         if (count == 0) {
2234                 count = delayed_refs->num_entries * 2;
2235                 run_most = 1;
2236         }
2237         while (1) {
2238                 if (!(run_all || run_most) &&
2239                     delayed_refs->num_heads_ready < 64)
2240                         break;
2241
2242                 /*
2243                  * go find something we can process in the rbtree.  We start at
2244                  * the beginning of the tree, and then build a cluster
2245                  * of refs to process starting at the first one we are able to
2246                  * lock
2247                  */
2248                 ret = btrfs_find_ref_cluster(trans, &cluster,
2249                                              delayed_refs->run_delayed_start);
2250                 if (ret)
2251                         break;
2252
2253                 ret = run_clustered_refs(trans, root, &cluster);
2254                 BUG_ON(ret < 0);
2255
2256                 count -= min_t(unsigned long, ret, count);
2257
2258                 if (count == 0)
2259                         break;
2260         }
2261
2262         if (run_all) {
2263                 node = rb_first(&delayed_refs->root);
2264                 if (!node)
2265                         goto out;
2266                 count = (unsigned long)-1;
2267
2268                 while (node) {
2269                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2270                                        rb_node);
2271                         if (btrfs_delayed_ref_is_head(ref)) {
2272                                 struct btrfs_delayed_ref_head *head;
2273
2274                                 head = btrfs_delayed_node_to_head(ref);
2275                                 atomic_inc(&ref->refs);
2276
2277                                 spin_unlock(&delayed_refs->lock);
2278                                 mutex_lock(&head->mutex);
2279                                 mutex_unlock(&head->mutex);
2280
2281                                 btrfs_put_delayed_ref(ref);
2282                                 cond_resched();
2283                                 goto again;
2284                         }
2285                         node = rb_next(node);
2286                 }
2287                 spin_unlock(&delayed_refs->lock);
2288                 schedule_timeout(1);
2289                 goto again;
2290         }
2291 out:
2292         spin_unlock(&delayed_refs->lock);
2293         return 0;
2294 }
2295
2296 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2297                                 struct btrfs_root *root,
2298                                 u64 bytenr, u64 num_bytes, u64 flags,
2299                                 int is_data)
2300 {
2301         struct btrfs_delayed_extent_op *extent_op;
2302         int ret;
2303
2304         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2305         if (!extent_op)
2306                 return -ENOMEM;
2307
2308         extent_op->flags_to_set = flags;
2309         extent_op->update_flags = 1;
2310         extent_op->update_key = 0;
2311         extent_op->is_data = is_data ? 1 : 0;
2312
2313         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2314         if (ret)
2315                 kfree(extent_op);
2316         return ret;
2317 }
2318
2319 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2320                                       struct btrfs_root *root,
2321                                       struct btrfs_path *path,
2322                                       u64 objectid, u64 offset, u64 bytenr)
2323 {
2324         struct btrfs_delayed_ref_head *head;
2325         struct btrfs_delayed_ref_node *ref;
2326         struct btrfs_delayed_data_ref *data_ref;
2327         struct btrfs_delayed_ref_root *delayed_refs;
2328         struct rb_node *node;
2329         int ret = 0;
2330
2331         ret = -ENOENT;
2332         delayed_refs = &trans->transaction->delayed_refs;
2333         spin_lock(&delayed_refs->lock);
2334         head = btrfs_find_delayed_ref_head(trans, bytenr);
2335         if (!head)
2336                 goto out;
2337
2338         if (!mutex_trylock(&head->mutex)) {
2339                 atomic_inc(&head->node.refs);
2340                 spin_unlock(&delayed_refs->lock);
2341
2342                 btrfs_release_path(root->fs_info->extent_root, path);
2343
2344                 mutex_lock(&head->mutex);
2345                 mutex_unlock(&head->mutex);
2346                 btrfs_put_delayed_ref(&head->node);
2347                 return -EAGAIN;
2348         }
2349
2350         node = rb_prev(&head->node.rb_node);
2351         if (!node)
2352                 goto out_unlock;
2353
2354         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2355
2356         if (ref->bytenr != bytenr)
2357                 goto out_unlock;
2358
2359         ret = 1;
2360         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2361                 goto out_unlock;
2362
2363         data_ref = btrfs_delayed_node_to_data_ref(ref);
2364
2365         node = rb_prev(node);
2366         if (node) {
2367                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2368                 if (ref->bytenr == bytenr)
2369                         goto out_unlock;
2370         }
2371
2372         if (data_ref->root != root->root_key.objectid ||
2373             data_ref->objectid != objectid || data_ref->offset != offset)
2374                 goto out_unlock;
2375
2376         ret = 0;
2377 out_unlock:
2378         mutex_unlock(&head->mutex);
2379 out:
2380         spin_unlock(&delayed_refs->lock);
2381         return ret;
2382 }
2383
2384 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2385                                         struct btrfs_root *root,
2386                                         struct btrfs_path *path,
2387                                         u64 objectid, u64 offset, u64 bytenr)
2388 {
2389         struct btrfs_root *extent_root = root->fs_info->extent_root;
2390         struct extent_buffer *leaf;
2391         struct btrfs_extent_data_ref *ref;
2392         struct btrfs_extent_inline_ref *iref;
2393         struct btrfs_extent_item *ei;
2394         struct btrfs_key key;
2395         u32 item_size;
2396         int ret;
2397
2398         key.objectid = bytenr;
2399         key.offset = (u64)-1;
2400         key.type = BTRFS_EXTENT_ITEM_KEY;
2401
2402         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2403         if (ret < 0)
2404                 goto out;
2405         BUG_ON(ret == 0);
2406
2407         ret = -ENOENT;
2408         if (path->slots[0] == 0)
2409                 goto out;
2410
2411         path->slots[0]--;
2412         leaf = path->nodes[0];
2413         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2414
2415         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2416                 goto out;
2417
2418         ret = 1;
2419         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2420 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2421         if (item_size < sizeof(*ei)) {
2422                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2423                 goto out;
2424         }
2425 #endif
2426         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2427
2428         if (item_size != sizeof(*ei) +
2429             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2430                 goto out;
2431
2432         if (btrfs_extent_generation(leaf, ei) <=
2433             btrfs_root_last_snapshot(&root->root_item))
2434                 goto out;
2435
2436         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2437         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2438             BTRFS_EXTENT_DATA_REF_KEY)
2439                 goto out;
2440
2441         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2442         if (btrfs_extent_refs(leaf, ei) !=
2443             btrfs_extent_data_ref_count(leaf, ref) ||
2444             btrfs_extent_data_ref_root(leaf, ref) !=
2445             root->root_key.objectid ||
2446             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2447             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2448                 goto out;
2449
2450         ret = 0;
2451 out:
2452         return ret;
2453 }
2454
2455 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2456                           struct btrfs_root *root,
2457                           u64 objectid, u64 offset, u64 bytenr)
2458 {
2459         struct btrfs_path *path;
2460         int ret;
2461         int ret2;
2462
2463         path = btrfs_alloc_path();
2464         if (!path)
2465                 return -ENOENT;
2466
2467         do {
2468                 ret = check_committed_ref(trans, root, path, objectid,
2469                                           offset, bytenr);
2470                 if (ret && ret != -ENOENT)
2471                         goto out;
2472
2473                 ret2 = check_delayed_ref(trans, root, path, objectid,
2474                                          offset, bytenr);
2475         } while (ret2 == -EAGAIN);
2476
2477         if (ret2 && ret2 != -ENOENT) {
2478                 ret = ret2;
2479                 goto out;
2480         }
2481
2482         if (ret != -ENOENT || ret2 != -ENOENT)
2483                 ret = 0;
2484 out:
2485         btrfs_free_path(path);
2486         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2487                 WARN_ON(ret > 0);
2488         return ret;
2489 }
2490
2491 #if 0
2492 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2493                     struct extent_buffer *buf, u32 nr_extents)
2494 {
2495         struct btrfs_key key;
2496         struct btrfs_file_extent_item *fi;
2497         u64 root_gen;
2498         u32 nritems;
2499         int i;
2500         int level;
2501         int ret = 0;
2502         int shared = 0;
2503
2504         if (!root->ref_cows)
2505                 return 0;
2506
2507         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2508                 shared = 0;
2509                 root_gen = root->root_key.offset;
2510         } else {
2511                 shared = 1;
2512                 root_gen = trans->transid - 1;
2513         }
2514
2515         level = btrfs_header_level(buf);
2516         nritems = btrfs_header_nritems(buf);
2517
2518         if (level == 0) {
2519                 struct btrfs_leaf_ref *ref;
2520                 struct btrfs_extent_info *info;
2521
2522                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2523                 if (!ref) {
2524                         ret = -ENOMEM;
2525                         goto out;
2526                 }
2527
2528                 ref->root_gen = root_gen;
2529                 ref->bytenr = buf->start;
2530                 ref->owner = btrfs_header_owner(buf);
2531                 ref->generation = btrfs_header_generation(buf);
2532                 ref->nritems = nr_extents;
2533                 info = ref->extents;
2534
2535                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2536                         u64 disk_bytenr;
2537                         btrfs_item_key_to_cpu(buf, &key, i);
2538                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2539                                 continue;
2540                         fi = btrfs_item_ptr(buf, i,
2541                                             struct btrfs_file_extent_item);
2542                         if (btrfs_file_extent_type(buf, fi) ==
2543                             BTRFS_FILE_EXTENT_INLINE)
2544                                 continue;
2545                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2546                         if (disk_bytenr == 0)
2547                                 continue;
2548
2549                         info->bytenr = disk_bytenr;
2550                         info->num_bytes =
2551                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2552                         info->objectid = key.objectid;
2553                         info->offset = key.offset;
2554                         info++;
2555                 }
2556
2557                 ret = btrfs_add_leaf_ref(root, ref, shared);
2558                 if (ret == -EEXIST && shared) {
2559                         struct btrfs_leaf_ref *old;
2560                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2561                         BUG_ON(!old);
2562                         btrfs_remove_leaf_ref(root, old);
2563                         btrfs_free_leaf_ref(root, old);
2564                         ret = btrfs_add_leaf_ref(root, ref, shared);
2565                 }
2566                 WARN_ON(ret);
2567                 btrfs_free_leaf_ref(root, ref);
2568         }
2569 out:
2570         return ret;
2571 }
2572
2573 /* when a block goes through cow, we update the reference counts of
2574  * everything that block points to.  The internal pointers of the block
2575  * can be in just about any order, and it is likely to have clusters of
2576  * things that are close together and clusters of things that are not.
2577  *
2578  * To help reduce the seeks that come with updating all of these reference
2579  * counts, sort them by byte number before actual updates are done.
2580  *
2581  * struct refsort is used to match byte number to slot in the btree block.
2582  * we sort based on the byte number and then use the slot to actually
2583  * find the item.
2584  *
2585  * struct refsort is smaller than strcut btrfs_item and smaller than
2586  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2587  * for a btree block, there's no way for a kmalloc of refsorts for a
2588  * single node to be bigger than a page.
2589  */
2590 struct refsort {
2591         u64 bytenr;
2592         u32 slot;
2593 };
2594
2595 /*
2596  * for passing into sort()
2597  */
2598 static int refsort_cmp(const void *a_void, const void *b_void)
2599 {
2600         const struct refsort *a = a_void;
2601         const struct refsort *b = b_void;
2602
2603         if (a->bytenr < b->bytenr)
2604                 return -1;
2605         if (a->bytenr > b->bytenr)
2606                 return 1;
2607         return 0;
2608 }
2609 #endif
2610
2611 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2612                            struct btrfs_root *root,
2613                            struct extent_buffer *buf,
2614                            int full_backref, int inc)
2615 {
2616         u64 bytenr;
2617         u64 num_bytes;
2618         u64 parent;
2619         u64 ref_root;
2620         u32 nritems;
2621         struct btrfs_key key;
2622         struct btrfs_file_extent_item *fi;
2623         int i;
2624         int level;
2625         int ret = 0;
2626         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2627                             u64, u64, u64, u64, u64, u64);
2628
2629         ref_root = btrfs_header_owner(buf);
2630         nritems = btrfs_header_nritems(buf);
2631         level = btrfs_header_level(buf);
2632
2633         if (!root->ref_cows && level == 0)
2634                 return 0;
2635
2636         if (inc)
2637                 process_func = btrfs_inc_extent_ref;
2638         else
2639                 process_func = btrfs_free_extent;
2640
2641         if (full_backref)
2642                 parent = buf->start;
2643         else
2644                 parent = 0;
2645
2646         for (i = 0; i < nritems; i++) {
2647                 if (level == 0) {
2648                         btrfs_item_key_to_cpu(buf, &key, i);
2649                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2650                                 continue;
2651                         fi = btrfs_item_ptr(buf, i,
2652                                             struct btrfs_file_extent_item);
2653                         if (btrfs_file_extent_type(buf, fi) ==
2654                             BTRFS_FILE_EXTENT_INLINE)
2655                                 continue;
2656                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2657                         if (bytenr == 0)
2658                                 continue;
2659
2660                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2661                         key.offset -= btrfs_file_extent_offset(buf, fi);
2662                         ret = process_func(trans, root, bytenr, num_bytes,
2663                                            parent, ref_root, key.objectid,
2664                                            key.offset);
2665                         if (ret)
2666                                 goto fail;
2667                 } else {
2668                         bytenr = btrfs_node_blockptr(buf, i);
2669                         num_bytes = btrfs_level_size(root, level - 1);
2670                         ret = process_func(trans, root, bytenr, num_bytes,
2671                                            parent, ref_root, level - 1, 0);
2672                         if (ret)
2673                                 goto fail;
2674                 }
2675         }
2676         return 0;
2677 fail:
2678         BUG();
2679         return ret;
2680 }
2681
2682 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2683                   struct extent_buffer *buf, int full_backref)
2684 {
2685         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2686 }
2687
2688 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2689                   struct extent_buffer *buf, int full_backref)
2690 {
2691         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2692 }
2693
2694 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2695                                  struct btrfs_root *root,
2696                                  struct btrfs_path *path,
2697                                  struct btrfs_block_group_cache *cache)
2698 {
2699         int ret;
2700         struct btrfs_root *extent_root = root->fs_info->extent_root;
2701         unsigned long bi;
2702         struct extent_buffer *leaf;
2703
2704         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2705         if (ret < 0)
2706                 goto fail;
2707         BUG_ON(ret);
2708
2709         leaf = path->nodes[0];
2710         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2711         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2712         btrfs_mark_buffer_dirty(leaf);
2713         btrfs_release_path(extent_root, path);
2714 fail:
2715         if (ret)
2716                 return ret;
2717         return 0;
2718
2719 }
2720
2721 static struct btrfs_block_group_cache *
2722 next_block_group(struct btrfs_root *root,
2723                  struct btrfs_block_group_cache *cache)
2724 {
2725         struct rb_node *node;
2726         spin_lock(&root->fs_info->block_group_cache_lock);
2727         node = rb_next(&cache->cache_node);
2728         btrfs_put_block_group(cache);
2729         if (node) {
2730                 cache = rb_entry(node, struct btrfs_block_group_cache,
2731                                  cache_node);
2732                 btrfs_get_block_group(cache);
2733         } else
2734                 cache = NULL;
2735         spin_unlock(&root->fs_info->block_group_cache_lock);
2736         return cache;
2737 }
2738
2739 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2740                             struct btrfs_trans_handle *trans,
2741                             struct btrfs_path *path)
2742 {
2743         struct btrfs_root *root = block_group->fs_info->tree_root;
2744         struct inode *inode = NULL;
2745         u64 alloc_hint = 0;
2746         int dcs = BTRFS_DC_ERROR;
2747         int num_pages = 0;
2748         int retries = 0;
2749         int ret = 0;
2750
2751         /*
2752          * If this block group is smaller than 100 megs don't bother caching the
2753          * block group.
2754          */
2755         if (block_group->key.offset < (100 * 1024 * 1024)) {
2756                 spin_lock(&block_group->lock);
2757                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2758                 spin_unlock(&block_group->lock);
2759                 return 0;
2760         }
2761
2762 again:
2763         inode = lookup_free_space_inode(root, block_group, path);
2764         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2765                 ret = PTR_ERR(inode);
2766                 btrfs_release_path(root, path);
2767                 goto out;
2768         }
2769
2770         if (IS_ERR(inode)) {
2771                 BUG_ON(retries);
2772                 retries++;
2773
2774                 if (block_group->ro)
2775                         goto out_free;
2776
2777                 ret = create_free_space_inode(root, trans, block_group, path);
2778                 if (ret)
2779                         goto out_free;
2780                 goto again;
2781         }
2782
2783         /*
2784          * We want to set the generation to 0, that way if anything goes wrong
2785          * from here on out we know not to trust this cache when we load up next
2786          * time.
2787          */
2788         BTRFS_I(inode)->generation = 0;
2789         ret = btrfs_update_inode(trans, root, inode);
2790         WARN_ON(ret);
2791
2792         if (i_size_read(inode) > 0) {
2793                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2794                                                       inode);
2795                 if (ret)
2796                         goto out_put;
2797         }
2798
2799         spin_lock(&block_group->lock);
2800         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2801                 /* We're not cached, don't bother trying to write stuff out */
2802                 dcs = BTRFS_DC_WRITTEN;
2803                 spin_unlock(&block_group->lock);
2804                 goto out_put;
2805         }
2806         spin_unlock(&block_group->lock);
2807
2808         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2809         if (!num_pages)
2810                 num_pages = 1;
2811
2812         /*
2813          * Just to make absolutely sure we have enough space, we're going to
2814          * preallocate 12 pages worth of space for each block group.  In
2815          * practice we ought to use at most 8, but we need extra space so we can
2816          * add our header and have a terminator between the extents and the
2817          * bitmaps.
2818          */
2819         num_pages *= 16;
2820         num_pages *= PAGE_CACHE_SIZE;
2821
2822         ret = btrfs_check_data_free_space(inode, num_pages);
2823         if (ret)
2824                 goto out_put;
2825
2826         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2827                                               num_pages, num_pages,
2828                                               &alloc_hint);
2829         if (!ret)
2830                 dcs = BTRFS_DC_SETUP;
2831         btrfs_free_reserved_data_space(inode, num_pages);
2832 out_put:
2833         iput(inode);
2834 out_free:
2835         btrfs_release_path(root, path);
2836 out:
2837         spin_lock(&block_group->lock);
2838         block_group->disk_cache_state = dcs;
2839         spin_unlock(&block_group->lock);
2840
2841         return ret;
2842 }
2843
2844 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2845                                    struct btrfs_root *root)
2846 {
2847         struct btrfs_block_group_cache *cache;
2848         int err = 0;
2849         struct btrfs_path *path;
2850         u64 last = 0;
2851
2852         path = btrfs_alloc_path();
2853         if (!path)
2854                 return -ENOMEM;
2855
2856 again:
2857         while (1) {
2858                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2859                 while (cache) {
2860                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2861                                 break;
2862                         cache = next_block_group(root, cache);
2863                 }
2864                 if (!cache) {
2865                         if (last == 0)
2866                                 break;
2867                         last = 0;
2868                         continue;
2869                 }
2870                 err = cache_save_setup(cache, trans, path);
2871                 last = cache->key.objectid + cache->key.offset;
2872                 btrfs_put_block_group(cache);
2873         }
2874
2875         while (1) {
2876                 if (last == 0) {
2877                         err = btrfs_run_delayed_refs(trans, root,
2878                                                      (unsigned long)-1);
2879                         BUG_ON(err);
2880                 }
2881
2882                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2883                 while (cache) {
2884                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2885                                 btrfs_put_block_group(cache);
2886                                 goto again;
2887                         }
2888
2889                         if (cache->dirty)
2890                                 break;
2891                         cache = next_block_group(root, cache);
2892                 }
2893                 if (!cache) {
2894                         if (last == 0)
2895                                 break;
2896                         last = 0;
2897                         continue;
2898                 }
2899
2900                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2901                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2902                 cache->dirty = 0;
2903                 last = cache->key.objectid + cache->key.offset;
2904
2905                 err = write_one_cache_group(trans, root, path, cache);
2906                 BUG_ON(err);
2907                 btrfs_put_block_group(cache);
2908         }
2909
2910         while (1) {
2911                 /*
2912                  * I don't think this is needed since we're just marking our
2913                  * preallocated extent as written, but just in case it can't
2914                  * hurt.
2915                  */
2916                 if (last == 0) {
2917                         err = btrfs_run_delayed_refs(trans, root,
2918                                                      (unsigned long)-1);
2919                         BUG_ON(err);
2920                 }
2921
2922                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2923                 while (cache) {
2924                         /*
2925                          * Really this shouldn't happen, but it could if we
2926                          * couldn't write the entire preallocated extent and
2927                          * splitting the extent resulted in a new block.
2928                          */
2929                         if (cache->dirty) {
2930                                 btrfs_put_block_group(cache);
2931                                 goto again;
2932                         }
2933                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2934                                 break;
2935                         cache = next_block_group(root, cache);
2936                 }
2937                 if (!cache) {
2938                         if (last == 0)
2939                                 break;
2940                         last = 0;
2941                         continue;
2942                 }
2943
2944                 btrfs_write_out_cache(root, trans, cache, path);
2945
2946                 /*
2947                  * If we didn't have an error then the cache state is still
2948                  * NEED_WRITE, so we can set it to WRITTEN.
2949                  */
2950                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2951                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2952                 last = cache->key.objectid + cache->key.offset;
2953                 btrfs_put_block_group(cache);
2954         }
2955
2956         btrfs_free_path(path);
2957         return 0;
2958 }
2959
2960 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2961 {
2962         struct btrfs_block_group_cache *block_group;
2963         int readonly = 0;
2964
2965         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2966         if (!block_group || block_group->ro)
2967                 readonly = 1;
2968         if (block_group)
2969                 btrfs_put_block_group(block_group);
2970         return readonly;
2971 }
2972
2973 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2974                              u64 total_bytes, u64 bytes_used,
2975                              struct btrfs_space_info **space_info)
2976 {
2977         struct btrfs_space_info *found;
2978         int i;
2979         int factor;
2980
2981         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2982                      BTRFS_BLOCK_GROUP_RAID10))
2983                 factor = 2;
2984         else
2985                 factor = 1;
2986
2987         found = __find_space_info(info, flags);
2988         if (found) {
2989                 spin_lock(&found->lock);
2990                 found->total_bytes += total_bytes;
2991                 found->disk_total += total_bytes * factor;
2992                 found->bytes_used += bytes_used;
2993                 found->disk_used += bytes_used * factor;
2994                 found->full = 0;
2995                 spin_unlock(&found->lock);
2996                 *space_info = found;
2997                 return 0;
2998         }
2999         found = kzalloc(sizeof(*found), GFP_NOFS);
3000         if (!found)
3001                 return -ENOMEM;
3002
3003         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3004                 INIT_LIST_HEAD(&found->block_groups[i]);
3005         init_rwsem(&found->groups_sem);
3006         spin_lock_init(&found->lock);
3007         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3008                                 BTRFS_BLOCK_GROUP_SYSTEM |
3009                                 BTRFS_BLOCK_GROUP_METADATA);
3010         found->total_bytes = total_bytes;
3011         found->disk_total = total_bytes * factor;
3012         found->bytes_used = bytes_used;
3013         found->disk_used = bytes_used * factor;
3014         found->bytes_pinned = 0;
3015         found->bytes_reserved = 0;
3016         found->bytes_readonly = 0;
3017         found->bytes_may_use = 0;
3018         found->full = 0;
3019         found->force_alloc = 0;
3020         *space_info = found;
3021         list_add_rcu(&found->list, &info->space_info);
3022         atomic_set(&found->caching_threads, 0);
3023         return 0;
3024 }
3025
3026 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3027 {
3028         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3029                                    BTRFS_BLOCK_GROUP_RAID1 |
3030                                    BTRFS_BLOCK_GROUP_RAID10 |
3031                                    BTRFS_BLOCK_GROUP_DUP);
3032         if (extra_flags) {
3033                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3034                         fs_info->avail_data_alloc_bits |= extra_flags;
3035                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3036                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3037                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3038                         fs_info->avail_system_alloc_bits |= extra_flags;
3039         }
3040 }
3041
3042 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3043 {
3044         /*
3045          * we add in the count of missing devices because we want
3046          * to make sure that any RAID levels on a degraded FS
3047          * continue to be honored.
3048          */
3049         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3050                 root->fs_info->fs_devices->missing_devices;
3051
3052         if (num_devices == 1)
3053                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3054         if (num_devices < 4)
3055                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3056
3057         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3058             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3059                       BTRFS_BLOCK_GROUP_RAID10))) {
3060                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3061         }
3062
3063         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3064             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3065                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3066         }
3067
3068         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3069             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3070              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3071              (flags & BTRFS_BLOCK_GROUP_DUP)))
3072                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3073         return flags;
3074 }
3075
3076 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3077 {
3078         if (flags & BTRFS_BLOCK_GROUP_DATA)
3079                 flags |= root->fs_info->avail_data_alloc_bits &
3080                          root->fs_info->data_alloc_profile;
3081         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3082                 flags |= root->fs_info->avail_system_alloc_bits &
3083                          root->fs_info->system_alloc_profile;
3084         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3085                 flags |= root->fs_info->avail_metadata_alloc_bits &
3086                          root->fs_info->metadata_alloc_profile;
3087         return btrfs_reduce_alloc_profile(root, flags);
3088 }
3089
3090 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3091 {
3092         u64 flags;
3093
3094         if (data)
3095                 flags = BTRFS_BLOCK_GROUP_DATA;
3096         else if (root == root->fs_info->chunk_root)
3097                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3098         else
3099                 flags = BTRFS_BLOCK_GROUP_METADATA;
3100
3101         return get_alloc_profile(root, flags);
3102 }
3103
3104 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3105 {
3106         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3107                                                        BTRFS_BLOCK_GROUP_DATA);
3108 }
3109
3110 /*
3111  * This will check the space that the inode allocates from to make sure we have
3112  * enough space for bytes.
3113  */
3114 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3115 {
3116         struct btrfs_space_info *data_sinfo;
3117         struct btrfs_root *root = BTRFS_I(inode)->root;
3118         u64 used;
3119         int ret = 0, committed = 0, alloc_chunk = 1;
3120
3121         /* make sure bytes are sectorsize aligned */
3122         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3123
3124         if (root == root->fs_info->tree_root) {
3125                 alloc_chunk = 0;
3126                 committed = 1;
3127         }
3128
3129         data_sinfo = BTRFS_I(inode)->space_info;
3130         if (!data_sinfo)
3131                 goto alloc;
3132
3133 again:
3134         /* make sure we have enough space to handle the data first */
3135         spin_lock(&data_sinfo->lock);
3136         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3137                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3138                 data_sinfo->bytes_may_use;
3139
3140         if (used + bytes > data_sinfo->total_bytes) {
3141                 struct btrfs_trans_handle *trans;
3142
3143                 /*
3144                  * if we don't have enough free bytes in this space then we need
3145                  * to alloc a new chunk.
3146                  */
3147                 if (!data_sinfo->full && alloc_chunk) {
3148                         u64 alloc_target;
3149
3150                         data_sinfo->force_alloc = 1;
3151                         spin_unlock(&data_sinfo->lock);
3152 alloc:
3153                         alloc_target = btrfs_get_alloc_profile(root, 1);
3154                         trans = btrfs_join_transaction(root, 1);
3155                         if (IS_ERR(trans))
3156                                 return PTR_ERR(trans);
3157
3158                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3159                                              bytes + 2 * 1024 * 1024,
3160                                              alloc_target, 0);
3161                         btrfs_end_transaction(trans, root);
3162                         if (ret < 0) {
3163                                 if (ret != -ENOSPC)
3164                                         return ret;
3165                                 else
3166                                         goto commit_trans;
3167                         }
3168
3169                         if (!data_sinfo) {
3170                                 btrfs_set_inode_space_info(root, inode);
3171                                 data_sinfo = BTRFS_I(inode)->space_info;
3172                         }
3173                         goto again;
3174                 }
3175                 spin_unlock(&data_sinfo->lock);
3176
3177                 /* commit the current transaction and try again */
3178 commit_trans:
3179                 if (!committed && !root->fs_info->open_ioctl_trans) {
3180                         committed = 1;
3181                         trans = btrfs_join_transaction(root, 1);
3182                         if (IS_ERR(trans))
3183                                 return PTR_ERR(trans);
3184                         ret = btrfs_commit_transaction(trans, root);
3185                         if (ret)
3186                                 return ret;
3187                         goto again;
3188                 }
3189
3190 #if 0 /* I hope we never need this code again, just in case */
3191                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3192                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3193                        "%llu bytes_readonly, %llu may use %llu total\n",
3194                        (unsigned long long)bytes,
3195                        (unsigned long long)data_sinfo->bytes_used,
3196                        (unsigned long long)data_sinfo->bytes_reserved,
3197                        (unsigned long long)data_sinfo->bytes_pinned,
3198                        (unsigned long long)data_sinfo->bytes_readonly,
3199                        (unsigned long long)data_sinfo->bytes_may_use,
3200                        (unsigned long long)data_sinfo->total_bytes);
3201 #endif
3202                 return -ENOSPC;
3203         }
3204         data_sinfo->bytes_may_use += bytes;
3205         BTRFS_I(inode)->reserved_bytes += bytes;
3206         spin_unlock(&data_sinfo->lock);
3207
3208         return 0;
3209 }
3210
3211 /*
3212  * called when we are clearing an delalloc extent from the
3213  * inode's io_tree or there was an error for whatever reason
3214  * after calling btrfs_check_data_free_space
3215  */
3216 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3217 {
3218         struct btrfs_root *root = BTRFS_I(inode)->root;
3219         struct btrfs_space_info *data_sinfo;
3220
3221         /* make sure bytes are sectorsize aligned */
3222         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3223
3224         data_sinfo = BTRFS_I(inode)->space_info;
3225         spin_lock(&data_sinfo->lock);
3226         data_sinfo->bytes_may_use -= bytes;
3227         BTRFS_I(inode)->reserved_bytes -= bytes;
3228         spin_unlock(&data_sinfo->lock);
3229 }
3230
3231 static void force_metadata_allocation(struct btrfs_fs_info *info)
3232 {
3233         struct list_head *head = &info->space_info;
3234         struct btrfs_space_info *found;
3235
3236         rcu_read_lock();
3237         list_for_each_entry_rcu(found, head, list) {
3238                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3239                         found->force_alloc = 1;
3240         }
3241         rcu_read_unlock();
3242 }
3243
3244 static int should_alloc_chunk(struct btrfs_root *root,
3245                               struct btrfs_space_info *sinfo, u64 alloc_bytes)
3246 {
3247         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3248         u64 thresh;
3249
3250         if (sinfo->bytes_used + sinfo->bytes_reserved +
3251             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3252                 return 0;
3253
3254         if (sinfo->bytes_used + sinfo->bytes_reserved +
3255             alloc_bytes < div_factor(num_bytes, 8))
3256                 return 0;
3257
3258         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3259         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3260
3261         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3262                 return 0;
3263
3264         return 1;
3265 }
3266
3267 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3268                           struct btrfs_root *extent_root, u64 alloc_bytes,
3269                           u64 flags, int force)
3270 {
3271         struct btrfs_space_info *space_info;
3272         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3273         int ret = 0;
3274
3275         mutex_lock(&fs_info->chunk_mutex);
3276
3277         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3278
3279         space_info = __find_space_info(extent_root->fs_info, flags);
3280         if (!space_info) {
3281                 ret = update_space_info(extent_root->fs_info, flags,
3282                                         0, 0, &space_info);
3283                 BUG_ON(ret);
3284         }
3285         BUG_ON(!space_info);
3286
3287         spin_lock(&space_info->lock);
3288         if (space_info->force_alloc)
3289                 force = 1;
3290         if (space_info->full) {
3291                 spin_unlock(&space_info->lock);
3292                 goto out;
3293         }
3294
3295         if (!force && !should_alloc_chunk(extent_root, space_info,
3296                                           alloc_bytes)) {
3297                 spin_unlock(&space_info->lock);
3298                 goto out;
3299         }
3300         spin_unlock(&space_info->lock);
3301
3302         /*
3303          * If we have mixed data/metadata chunks we want to make sure we keep
3304          * allocating mixed chunks instead of individual chunks.
3305          */
3306         if (btrfs_mixed_space_info(space_info))
3307                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3308
3309         /*
3310          * if we're doing a data chunk, go ahead and make sure that
3311          * we keep a reasonable number of metadata chunks allocated in the
3312          * FS as well.
3313          */
3314         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3315                 fs_info->data_chunk_allocations++;
3316                 if (!(fs_info->data_chunk_allocations %
3317                       fs_info->metadata_ratio))
3318                         force_metadata_allocation(fs_info);
3319         }
3320
3321         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3322         spin_lock(&space_info->lock);
3323         if (ret)
3324                 space_info->full = 1;
3325         else
3326                 ret = 1;
3327         space_info->force_alloc = 0;
3328         spin_unlock(&space_info->lock);
3329 out:
3330         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3331         return ret;
3332 }
3333
3334 /*
3335  * shrink metadata reservation for delalloc
3336  */
3337 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3338                            struct btrfs_root *root, u64 to_reclaim, int sync)
3339 {
3340         struct btrfs_block_rsv *block_rsv;
3341         struct btrfs_space_info *space_info;
3342         u64 reserved;
3343         u64 max_reclaim;
3344         u64 reclaimed = 0;
3345         long time_left;
3346         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3347         int loops = 0;
3348         unsigned long progress;
3349
3350         block_rsv = &root->fs_info->delalloc_block_rsv;
3351         space_info = block_rsv->space_info;
3352
3353         smp_mb();
3354         reserved = space_info->bytes_reserved;
3355         progress = space_info->reservation_progress;
3356
3357         if (reserved == 0)
3358                 return 0;
3359
3360         max_reclaim = min(reserved, to_reclaim);
3361
3362         while (loops < 1024) {
3363                 /* have the flusher threads jump in and do some IO */
3364                 smp_mb();
3365                 nr_pages = min_t(unsigned long, nr_pages,
3366                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3367                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3368
3369                 spin_lock(&space_info->lock);
3370                 if (reserved > space_info->bytes_reserved)
3371                         reclaimed += reserved - space_info->bytes_reserved;
3372                 reserved = space_info->bytes_reserved;
3373                 spin_unlock(&space_info->lock);
3374
3375                 loops++;
3376
3377                 if (reserved == 0 || reclaimed >= max_reclaim)
3378                         break;
3379
3380                 if (trans && trans->transaction->blocked)
3381                         return -EAGAIN;
3382
3383                 time_left = schedule_timeout_interruptible(1);
3384
3385                 /* We were interrupted, exit */
3386                 if (time_left)
3387                         break;
3388
3389                 /* we've kicked the IO a few times, if anything has been freed,
3390                  * exit.  There is no sense in looping here for a long time
3391                  * when we really need to commit the transaction, or there are
3392                  * just too many writers without enough free space
3393                  */
3394
3395                 if (loops > 3) {
3396                         smp_mb();
3397                         if (progress != space_info->reservation_progress)
3398                                 break;
3399                 }
3400
3401         }
3402         return reclaimed >= to_reclaim;
3403 }
3404
3405 /*
3406  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3407  * idea is if this is the first call (retries == 0) then we will add to our
3408  * reserved count if we can't make the allocation in order to hold our place
3409  * while we go and try and free up space.  That way for retries > 1 we don't try
3410  * and add space, we just check to see if the amount of unused space is >= the
3411  * total space, meaning that our reservation is valid.
3412  *
3413  * However if we don't intend to retry this reservation, pass -1 as retries so
3414  * that it short circuits this logic.
3415  */
3416 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3417                                   struct btrfs_root *root,
3418                                   struct btrfs_block_rsv *block_rsv,
3419                                   u64 orig_bytes, int flush)
3420 {
3421         struct btrfs_space_info *space_info = block_rsv->space_info;
3422         u64 unused;
3423         u64 num_bytes = orig_bytes;
3424         int retries = 0;
3425         int ret = 0;
3426         bool reserved = false;
3427         bool committed = false;
3428
3429 again:
3430         ret = -ENOSPC;
3431         if (reserved)
3432                 num_bytes = 0;
3433
3434         spin_lock(&space_info->lock);
3435         unused = space_info->bytes_used + space_info->bytes_reserved +
3436                  space_info->bytes_pinned + space_info->bytes_readonly +
3437                  space_info->bytes_may_use;
3438
3439         /*
3440          * The idea here is that we've not already over-reserved the block group
3441          * then we can go ahead and save our reservation first and then start
3442          * flushing if we need to.  Otherwise if we've already overcommitted
3443          * lets start flushing stuff first and then come back and try to make
3444          * our reservation.
3445          */
3446         if (unused <= space_info->total_bytes) {
3447                 unused = space_info->total_bytes - unused;
3448                 if (unused >= num_bytes) {
3449                         if (!reserved)
3450                                 space_info->bytes_reserved += orig_bytes;
3451                         ret = 0;
3452                 } else {
3453                         /*
3454                          * Ok set num_bytes to orig_bytes since we aren't
3455                          * overocmmitted, this way we only try and reclaim what
3456                          * we need.
3457                          */
3458                         num_bytes = orig_bytes;
3459                 }
3460         } else {
3461                 /*
3462                  * Ok we're over committed, set num_bytes to the overcommitted
3463                  * amount plus the amount of bytes that we need for this
3464                  * reservation.
3465                  */
3466                 num_bytes = unused - space_info->total_bytes +
3467                         (orig_bytes * (retries + 1));
3468         }
3469
3470         /*
3471          * Couldn't make our reservation, save our place so while we're trying
3472          * to reclaim space we can actually use it instead of somebody else
3473          * stealing it from us.
3474          */
3475         if (ret && !reserved) {
3476                 space_info->bytes_reserved += orig_bytes;
3477                 reserved = true;
3478         }
3479
3480         spin_unlock(&space_info->lock);
3481
3482         if (!ret)
3483                 return 0;
3484
3485         if (!flush)
3486                 goto out;
3487
3488         /*
3489          * We do synchronous shrinking since we don't actually unreserve
3490          * metadata until after the IO is completed.
3491          */
3492         ret = shrink_delalloc(trans, root, num_bytes, 1);
3493         if (ret > 0)
3494                 return 0;
3495         else if (ret < 0)
3496                 goto out;
3497
3498         /*
3499          * So if we were overcommitted it's possible that somebody else flushed
3500          * out enough space and we simply didn't have enough space to reclaim,
3501          * so go back around and try again.
3502          */
3503         if (retries < 2) {
3504                 retries++;
3505                 goto again;
3506         }
3507
3508         spin_lock(&space_info->lock);
3509         /*
3510          * Not enough space to be reclaimed, don't bother committing the
3511          * transaction.
3512          */
3513         if (space_info->bytes_pinned < orig_bytes)
3514                 ret = -ENOSPC;
3515         spin_unlock(&space_info->lock);
3516         if (ret)
3517                 goto out;
3518
3519         ret = -EAGAIN;
3520         if (trans || committed)
3521                 goto out;
3522
3523         ret = -ENOSPC;
3524         trans = btrfs_join_transaction(root, 1);
3525         if (IS_ERR(trans))
3526                 goto out;
3527         ret = btrfs_commit_transaction(trans, root);
3528         if (!ret) {
3529                 trans = NULL;
3530                 committed = true;
3531                 goto again;
3532         }
3533
3534 out:
3535         if (reserved) {
3536                 spin_lock(&space_info->lock);
3537                 space_info->bytes_reserved -= orig_bytes;
3538                 spin_unlock(&space_info->lock);
3539         }
3540
3541         return ret;
3542 }
3543
3544 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3545                                              struct btrfs_root *root)
3546 {
3547         struct btrfs_block_rsv *block_rsv;
3548         if (root->ref_cows)
3549                 block_rsv = trans->block_rsv;
3550         else
3551                 block_rsv = root->block_rsv;
3552
3553         if (!block_rsv)
3554                 block_rsv = &root->fs_info->empty_block_rsv;
3555
3556         return block_rsv;
3557 }
3558
3559 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3560                                u64 num_bytes)
3561 {
3562         int ret = -ENOSPC;
3563         spin_lock(&block_rsv->lock);
3564         if (block_rsv->reserved >= num_bytes) {
3565                 block_rsv->reserved -= num_bytes;
3566                 if (block_rsv->reserved < block_rsv->size)
3567                         block_rsv->full = 0;
3568                 ret = 0;
3569         }
3570         spin_unlock(&block_rsv->lock);
3571         return ret;
3572 }
3573
3574 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3575                                 u64 num_bytes, int update_size)
3576 {
3577         spin_lock(&block_rsv->lock);
3578         block_rsv->reserved += num_bytes;
3579         if (update_size)
3580                 block_rsv->size += num_bytes;
3581         else if (block_rsv->reserved >= block_rsv->size)
3582                 block_rsv->full = 1;
3583         spin_unlock(&block_rsv->lock);
3584 }
3585
3586 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3587                              struct btrfs_block_rsv *dest, u64 num_bytes)
3588 {
3589         struct btrfs_space_info *space_info = block_rsv->space_info;
3590
3591         spin_lock(&block_rsv->lock);
3592         if (num_bytes == (u64)-1)
3593                 num_bytes = block_rsv->size;
3594         block_rsv->size -= num_bytes;
3595         if (block_rsv->reserved >= block_rsv->size) {
3596                 num_bytes = block_rsv->reserved - block_rsv->size;
3597                 block_rsv->reserved = block_rsv->size;
3598                 block_rsv->full = 1;
3599         } else {
3600                 num_bytes = 0;
3601         }
3602         spin_unlock(&block_rsv->lock);
3603
3604         if (num_bytes > 0) {
3605                 if (dest) {
3606                         spin_lock(&dest->lock);
3607                         if (!dest->full) {
3608                                 u64 bytes_to_add;
3609
3610                                 bytes_to_add = dest->size - dest->reserved;
3611                                 bytes_to_add = min(num_bytes, bytes_to_add);
3612                                 dest->reserved += bytes_to_add;
3613                                 if (dest->reserved >= dest->size)
3614                                         dest->full = 1;
3615                                 num_bytes -= bytes_to_add;
3616                         }
3617                         spin_unlock(&dest->lock);
3618                 }
3619                 if (num_bytes) {
3620                         spin_lock(&space_info->lock);
3621                         space_info->bytes_reserved -= num_bytes;
3622                         space_info->reservation_progress++;
3623                         spin_unlock(&space_info->lock);
3624                 }
3625         }
3626 }
3627
3628 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3629                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3630 {
3631         int ret;
3632
3633         ret = block_rsv_use_bytes(src, num_bytes);
3634         if (ret)
3635                 return ret;
3636
3637         block_rsv_add_bytes(dst, num_bytes, 1);
3638         return 0;
3639 }
3640
3641 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3642 {
3643         memset(rsv, 0, sizeof(*rsv));
3644         spin_lock_init(&rsv->lock);
3645         atomic_set(&rsv->usage, 1);
3646         rsv->priority = 6;
3647         INIT_LIST_HEAD(&rsv->list);
3648 }
3649
3650 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3651 {
3652         struct btrfs_block_rsv *block_rsv;
3653         struct btrfs_fs_info *fs_info = root->fs_info;
3654
3655         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3656         if (!block_rsv)
3657                 return NULL;
3658
3659         btrfs_init_block_rsv(block_rsv);
3660         block_rsv->space_info = __find_space_info(fs_info,
3661                                                   BTRFS_BLOCK_GROUP_METADATA);
3662         return block_rsv;
3663 }
3664
3665 void btrfs_free_block_rsv(struct btrfs_root *root,
3666                           struct btrfs_block_rsv *rsv)
3667 {
3668         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3669                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3670                 if (!rsv->durable)
3671                         kfree(rsv);
3672         }
3673 }
3674
3675 /*
3676  * make the block_rsv struct be able to capture freed space.
3677  * the captured space will re-add to the the block_rsv struct
3678  * after transaction commit
3679  */
3680 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3681                                  struct btrfs_block_rsv *block_rsv)
3682 {
3683         block_rsv->durable = 1;
3684         mutex_lock(&fs_info->durable_block_rsv_mutex);
3685         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3686         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3687 }
3688
3689 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3690                         struct btrfs_root *root,
3691                         struct btrfs_block_rsv *block_rsv,
3692                         u64 num_bytes)
3693 {
3694         int ret;
3695
3696         if (num_bytes == 0)
3697                 return 0;
3698
3699         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3700         if (!ret) {
3701                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3702                 return 0;
3703         }
3704
3705         return ret;
3706 }
3707
3708 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3709                           struct btrfs_root *root,
3710                           struct btrfs_block_rsv *block_rsv,
3711                           u64 min_reserved, int min_factor)
3712 {
3713         u64 num_bytes = 0;
3714         int commit_trans = 0;
3715         int ret = -ENOSPC;
3716
3717         if (!block_rsv)
3718                 return 0;
3719
3720         spin_lock(&block_rsv->lock);
3721         if (min_factor > 0)
3722                 num_bytes = div_factor(block_rsv->size, min_factor);
3723         if (min_reserved > num_bytes)
3724                 num_bytes = min_reserved;
3725
3726         if (block_rsv->reserved >= num_bytes) {
3727                 ret = 0;
3728         } else {
3729                 num_bytes -= block_rsv->reserved;
3730                 if (block_rsv->durable &&
3731                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3732                         commit_trans = 1;
3733         }
3734         spin_unlock(&block_rsv->lock);
3735         if (!ret)
3736                 return 0;
3737
3738         if (block_rsv->refill_used) {
3739                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3740                                              num_bytes, 0);
3741                 if (!ret) {
3742                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3743                         return 0;
3744                 }
3745         }
3746
3747         if (commit_trans) {
3748                 if (trans)
3749                         return -EAGAIN;
3750
3751                 trans = btrfs_join_transaction(root, 1);
3752                 BUG_ON(IS_ERR(trans));
3753                 ret = btrfs_commit_transaction(trans, root);
3754                 return 0;
3755         }
3756
3757         return -ENOSPC;
3758 }
3759
3760 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3761                             struct btrfs_block_rsv *dst_rsv,
3762                             u64 num_bytes)
3763 {
3764         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3765 }
3766
3767 void btrfs_block_rsv_release(struct btrfs_root *root,
3768                              struct btrfs_block_rsv *block_rsv,
3769                              u64 num_bytes)
3770 {
3771         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3772         if (global_rsv->full || global_rsv == block_rsv ||
3773             block_rsv->space_info != global_rsv->space_info)
3774                 global_rsv = NULL;
3775         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3776 }
3777
3778 /*
3779  * helper to calculate size of global block reservation.
3780  * the desired value is sum of space used by extent tree,
3781  * checksum tree and root tree
3782  */
3783 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3784 {
3785         struct btrfs_space_info *sinfo;
3786         u64 num_bytes;
3787         u64 meta_used;
3788         u64 data_used;
3789         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3790 #if 0
3791         /*
3792          * per tree used space accounting can be inaccuracy, so we
3793          * can't rely on it.
3794          */
3795         spin_lock(&fs_info->extent_root->accounting_lock);
3796         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3797         spin_unlock(&fs_info->extent_root->accounting_lock);
3798
3799         spin_lock(&fs_info->csum_root->accounting_lock);
3800         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3801         spin_unlock(&fs_info->csum_root->accounting_lock);
3802
3803         spin_lock(&fs_info->tree_root->accounting_lock);
3804         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3805         spin_unlock(&fs_info->tree_root->accounting_lock);
3806 #endif
3807         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3808         spin_lock(&sinfo->lock);
3809         data_used = sinfo->bytes_used;
3810         spin_unlock(&sinfo->lock);
3811
3812         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3813         spin_lock(&sinfo->lock);
3814         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3815                 data_used = 0;
3816         meta_used = sinfo->bytes_used;
3817         spin_unlock(&sinfo->lock);
3818
3819         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3820                     csum_size * 2;
3821         num_bytes += div64_u64(data_used + meta_used, 50);
3822
3823         if (num_bytes * 3 > meta_used)
3824                 num_bytes = div64_u64(meta_used, 3);
3825
3826         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3827 }
3828
3829 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3830 {
3831         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3832         struct btrfs_space_info *sinfo = block_rsv->space_info;
3833         u64 num_bytes;
3834
3835         num_bytes = calc_global_metadata_size(fs_info);
3836
3837         spin_lock(&block_rsv->lock);
3838         spin_lock(&sinfo->lock);
3839
3840         block_rsv->size = num_bytes;
3841
3842         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3843                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3844                     sinfo->bytes_may_use;
3845
3846         if (sinfo->total_bytes > num_bytes) {
3847                 num_bytes = sinfo->total_bytes - num_bytes;
3848                 block_rsv->reserved += num_bytes;
3849                 sinfo->bytes_reserved += num_bytes;
3850         }
3851
3852         if (block_rsv->reserved >= block_rsv->size) {
3853                 num_bytes = block_rsv->reserved - block_rsv->size;
3854                 sinfo->bytes_reserved -= num_bytes;
3855                 sinfo->reservation_progress++;
3856                 block_rsv->reserved = block_rsv->size;
3857                 block_rsv->full = 1;
3858         }
3859 #if 0
3860         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3861                 block_rsv->size, block_rsv->reserved);
3862 #endif
3863         spin_unlock(&sinfo->lock);
3864         spin_unlock(&block_rsv->lock);
3865 }
3866
3867 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3868 {
3869         struct btrfs_space_info *space_info;
3870
3871         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3872         fs_info->chunk_block_rsv.space_info = space_info;
3873         fs_info->chunk_block_rsv.priority = 10;
3874
3875         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3876         fs_info->global_block_rsv.space_info = space_info;
3877         fs_info->global_block_rsv.priority = 10;
3878         fs_info->global_block_rsv.refill_used = 1;
3879         fs_info->delalloc_block_rsv.space_info = space_info;
3880         fs_info->trans_block_rsv.space_info = space_info;
3881         fs_info->empty_block_rsv.space_info = space_info;
3882         fs_info->empty_block_rsv.priority = 10;
3883
3884         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3885         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3886         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3887         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3888         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3889
3890         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3891
3892         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3893
3894         update_global_block_rsv(fs_info);
3895 }
3896
3897 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3898 {
3899         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3900         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3901         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3902         WARN_ON(fs_info->trans_block_rsv.size > 0);
3903         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3904         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3905         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3906 }
3907
3908 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3909 {
3910         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3911                 3 * num_items;
3912 }
3913
3914 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3915                                  struct btrfs_root *root,
3916                                  int num_items)
3917 {
3918         u64 num_bytes;
3919         int ret;
3920
3921         if (num_items == 0 || root->fs_info->chunk_root == root)
3922                 return 0;
3923
3924         num_bytes = calc_trans_metadata_size(root, num_items);
3925         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3926                                   num_bytes);
3927         if (!ret) {
3928                 trans->bytes_reserved += num_bytes;
3929                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3930         }
3931         return ret;
3932 }
3933
3934 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3935                                   struct btrfs_root *root)
3936 {
3937         if (!trans->bytes_reserved)
3938                 return;
3939
3940         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3941         btrfs_block_rsv_release(root, trans->block_rsv,
3942                                 trans->bytes_reserved);
3943         trans->bytes_reserved = 0;
3944 }
3945
3946 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3947                                   struct inode *inode)
3948 {
3949         struct btrfs_root *root = BTRFS_I(inode)->root;
3950         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3951         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3952
3953         /*
3954          * one for deleting orphan item, one for updating inode and
3955          * two for calling btrfs_truncate_inode_items.
3956          *
3957          * btrfs_truncate_inode_items is a delete operation, it frees
3958          * more space than it uses in most cases. So two units of
3959          * metadata space should be enough for calling it many times.
3960          * If all of the metadata space is used, we can commit
3961          * transaction and use space it freed.
3962          */
3963         u64 num_bytes = calc_trans_metadata_size(root, 4);
3964         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3965 }
3966
3967 void btrfs_orphan_release_metadata(struct inode *inode)
3968 {
3969         struct btrfs_root *root = BTRFS_I(inode)->root;
3970         u64 num_bytes = calc_trans_metadata_size(root, 4);
3971         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3972 }
3973
3974 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3975                                 struct btrfs_pending_snapshot *pending)
3976 {
3977         struct btrfs_root *root = pending->root;
3978         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3979         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3980         /*
3981          * two for root back/forward refs, two for directory entries
3982          * and one for root of the snapshot.
3983          */
3984         u64 num_bytes = calc_trans_metadata_size(root, 5);
3985         dst_rsv->space_info = src_rsv->space_info;
3986         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3987 }
3988
3989 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3990 {
3991         return num_bytes >>= 3;
3992 }
3993
3994 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3995 {
3996         struct btrfs_root *root = BTRFS_I(inode)->root;
3997         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3998         u64 to_reserve;
3999         int nr_extents;
4000         int ret;
4001
4002         if (btrfs_transaction_in_commit(root->fs_info))
4003                 schedule_timeout(1);
4004
4005         num_bytes = ALIGN(num_bytes, root->sectorsize);
4006
4007         spin_lock(&BTRFS_I(inode)->accounting_lock);
4008         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4009         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
4010                 nr_extents -= BTRFS_I(inode)->reserved_extents;
4011                 to_reserve = calc_trans_metadata_size(root, nr_extents);
4012         } else {
4013                 nr_extents = 0;
4014                 to_reserve = 0;
4015         }
4016         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4017         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4018         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4019         if (ret)
4020                 return ret;
4021
4022         spin_lock(&BTRFS_I(inode)->accounting_lock);
4023         BTRFS_I(inode)->reserved_extents += nr_extents;
4024         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4025         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4026
4027         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4028
4029         if (block_rsv->size > 512 * 1024 * 1024)
4030                 shrink_delalloc(NULL, root, to_reserve, 0);
4031
4032         return 0;
4033 }
4034
4035 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4036 {
4037         struct btrfs_root *root = BTRFS_I(inode)->root;
4038         u64 to_free;
4039         int nr_extents;
4040
4041         num_bytes = ALIGN(num_bytes, root->sectorsize);
4042         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4043         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4044
4045         spin_lock(&BTRFS_I(inode)->accounting_lock);
4046         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4047         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4048                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4049                 BTRFS_I(inode)->reserved_extents -= nr_extents;
4050         } else {
4051                 nr_extents = 0;
4052         }
4053         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4054
4055         to_free = calc_csum_metadata_size(inode, num_bytes);
4056         if (nr_extents > 0)
4057                 to_free += calc_trans_metadata_size(root, nr_extents);
4058
4059         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4060                                 to_free);
4061 }
4062
4063 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4064 {
4065         int ret;
4066
4067         ret = btrfs_check_data_free_space(inode, num_bytes);
4068         if (ret)
4069                 return ret;
4070
4071         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4072         if (ret) {
4073                 btrfs_free_reserved_data_space(inode, num_bytes);
4074                 return ret;
4075         }
4076
4077         return 0;
4078 }
4079
4080 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4081 {
4082         btrfs_delalloc_release_metadata(inode, num_bytes);
4083         btrfs_free_reserved_data_space(inode, num_bytes);
4084 }
4085
4086 static int update_block_group(struct btrfs_trans_handle *trans,
4087                               struct btrfs_root *root,
4088                               u64 bytenr, u64 num_bytes, int alloc)
4089 {
4090         struct btrfs_block_group_cache *cache = NULL;
4091         struct btrfs_fs_info *info = root->fs_info;
4092         u64 total = num_bytes;
4093         u64 old_val;
4094         u64 byte_in_group;
4095         int factor;
4096
4097         /* block accounting for super block */
4098         spin_lock(&info->delalloc_lock);
4099         old_val = btrfs_super_bytes_used(&info->super_copy);
4100         if (alloc)
4101                 old_val += num_bytes;
4102         else
4103                 old_val -= num_bytes;
4104         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4105         spin_unlock(&info->delalloc_lock);
4106
4107         while (total) {
4108                 cache = btrfs_lookup_block_group(info, bytenr);
4109                 if (!cache)
4110                         return -1;
4111                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4112                                     BTRFS_BLOCK_GROUP_RAID1 |
4113                                     BTRFS_BLOCK_GROUP_RAID10))
4114                         factor = 2;
4115                 else
4116                         factor = 1;
4117                 /*
4118                  * If this block group has free space cache written out, we
4119                  * need to make sure to load it if we are removing space.  This
4120                  * is because we need the unpinning stage to actually add the
4121                  * space back to the block group, otherwise we will leak space.
4122                  */
4123                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4124                         cache_block_group(cache, trans, NULL, 1);
4125
4126                 byte_in_group = bytenr - cache->key.objectid;
4127                 WARN_ON(byte_in_group > cache->key.offset);
4128
4129                 spin_lock(&cache->space_info->lock);
4130                 spin_lock(&cache->lock);
4131
4132                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4133                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4134                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4135
4136                 cache->dirty = 1;
4137                 old_val = btrfs_block_group_used(&cache->item);
4138                 num_bytes = min(total, cache->key.offset - byte_in_group);
4139                 if (alloc) {
4140                         old_val += num_bytes;
4141                         btrfs_set_block_group_used(&cache->item, old_val);
4142                         cache->reserved -= num_bytes;
4143                         cache->space_info->bytes_reserved -= num_bytes;
4144                         cache->space_info->reservation_progress++;
4145                         cache->space_info->bytes_used += num_bytes;
4146                         cache->space_info->disk_used += num_bytes * factor;
4147                         spin_unlock(&cache->lock);
4148                         spin_unlock(&cache->space_info->lock);
4149                 } else {
4150                         old_val -= num_bytes;
4151                         btrfs_set_block_group_used(&cache->item, old_val);
4152                         cache->pinned += num_bytes;
4153                         cache->space_info->bytes_pinned += num_bytes;
4154                         cache->space_info->bytes_used -= num_bytes;
4155                         cache->space_info->disk_used -= num_bytes * factor;
4156                         spin_unlock(&cache->lock);
4157                         spin_unlock(&cache->space_info->lock);
4158
4159                         set_extent_dirty(info->pinned_extents,
4160                                          bytenr, bytenr + num_bytes - 1,
4161                                          GFP_NOFS | __GFP_NOFAIL);
4162                 }
4163                 btrfs_put_block_group(cache);
4164                 total -= num_bytes;
4165                 bytenr += num_bytes;
4166         }
4167         return 0;
4168 }
4169
4170 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4171 {
4172         struct btrfs_block_group_cache *cache;
4173         u64 bytenr;
4174
4175         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4176         if (!cache)
4177                 return 0;
4178
4179         bytenr = cache->key.objectid;
4180         btrfs_put_block_group(cache);
4181
4182         return bytenr;
4183 }
4184
4185 static int pin_down_extent(struct btrfs_root *root,
4186                            struct btrfs_block_group_cache *cache,
4187                            u64 bytenr, u64 num_bytes, int reserved)
4188 {
4189         spin_lock(&cache->space_info->lock);
4190         spin_lock(&cache->lock);
4191         cache->pinned += num_bytes;
4192         cache->space_info->bytes_pinned += num_bytes;
4193         if (reserved) {
4194                 cache->reserved -= num_bytes;
4195                 cache->space_info->bytes_reserved -= num_bytes;
4196                 cache->space_info->reservation_progress++;
4197         }
4198         spin_unlock(&cache->lock);
4199         spin_unlock(&cache->space_info->lock);
4200
4201         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4202                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4203         return 0;
4204 }
4205
4206 /*
4207  * this function must be called within transaction
4208  */
4209 int btrfs_pin_extent(struct btrfs_root *root,
4210                      u64 bytenr, u64 num_bytes, int reserved)
4211 {
4212         struct btrfs_block_group_cache *cache;
4213
4214         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4215         BUG_ON(!cache);
4216
4217         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4218
4219         btrfs_put_block_group(cache);
4220         return 0;
4221 }
4222
4223 /*
4224  * update size of reserved extents. this function may return -EAGAIN
4225  * if 'reserve' is true or 'sinfo' is false.
4226  */
4227 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4228                                  u64 num_bytes, int reserve, int sinfo)
4229 {
4230         int ret = 0;
4231         if (sinfo) {
4232                 struct btrfs_space_info *space_info = cache->space_info;
4233                 spin_lock(&space_info->lock);
4234                 spin_lock(&cache->lock);
4235                 if (reserve) {
4236                         if (cache->ro) {
4237                                 ret = -EAGAIN;
4238                         } else {
4239                                 cache->reserved += num_bytes;
4240                                 space_info->bytes_reserved += num_bytes;
4241                         }
4242                 } else {
4243                         if (cache->ro)
4244                                 space_info->bytes_readonly += num_bytes;
4245                         cache->reserved -= num_bytes;
4246                         space_info->bytes_reserved -= num_bytes;
4247                         space_info->reservation_progress++;
4248                 }
4249                 spin_unlock(&cache->lock);
4250                 spin_unlock(&space_info->lock);
4251         } else {
4252                 spin_lock(&cache->lock);
4253                 if (cache->ro) {
4254                         ret = -EAGAIN;
4255                 } else {
4256                         if (reserve)
4257                                 cache->reserved += num_bytes;
4258                         else
4259                                 cache->reserved -= num_bytes;
4260                 }
4261                 spin_unlock(&cache->lock);
4262         }
4263         return ret;
4264 }
4265
4266 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4267                                 struct btrfs_root *root)
4268 {
4269         struct btrfs_fs_info *fs_info = root->fs_info;
4270         struct btrfs_caching_control *next;
4271         struct btrfs_caching_control *caching_ctl;
4272         struct btrfs_block_group_cache *cache;
4273
4274         down_write(&fs_info->extent_commit_sem);
4275
4276         list_for_each_entry_safe(caching_ctl, next,
4277                                  &fs_info->caching_block_groups, list) {
4278                 cache = caching_ctl->block_group;
4279                 if (block_group_cache_done(cache)) {
4280                         cache->last_byte_to_unpin = (u64)-1;
4281                         list_del_init(&caching_ctl->list);
4282                         put_caching_control(caching_ctl);
4283                 } else {
4284                         cache->last_byte_to_unpin = caching_ctl->progress;
4285                 }
4286         }
4287
4288         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4289                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4290         else
4291                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4292
4293         up_write(&fs_info->extent_commit_sem);
4294
4295         update_global_block_rsv(fs_info);
4296         return 0;
4297 }
4298
4299 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4300 {
4301         struct btrfs_fs_info *fs_info = root->fs_info;
4302         struct btrfs_block_group_cache *cache = NULL;
4303         u64 len;
4304
4305         while (start <= end) {
4306                 if (!cache ||
4307                     start >= cache->key.objectid + cache->key.offset) {
4308                         if (cache)
4309                                 btrfs_put_block_group(cache);
4310                         cache = btrfs_lookup_block_group(fs_info, start);
4311                         BUG_ON(!cache);
4312                 }
4313
4314                 len = cache->key.objectid + cache->key.offset - start;
4315                 len = min(len, end + 1 - start);
4316
4317                 if (start < cache->last_byte_to_unpin) {
4318                         len = min(len, cache->last_byte_to_unpin - start);
4319                         btrfs_add_free_space(cache, start, len);
4320                 }
4321
4322                 start += len;
4323
4324                 spin_lock(&cache->space_info->lock);
4325                 spin_lock(&cache->lock);
4326                 cache->pinned -= len;
4327                 cache->space_info->bytes_pinned -= len;
4328                 if (cache->ro) {
4329                         cache->space_info->bytes_readonly += len;
4330                 } else if (cache->reserved_pinned > 0) {
4331                         len = min(len, cache->reserved_pinned);
4332                         cache->reserved_pinned -= len;
4333                         cache->space_info->bytes_reserved += len;
4334                 }
4335                 spin_unlock(&cache->lock);
4336                 spin_unlock(&cache->space_info->lock);
4337         }
4338
4339         if (cache)
4340                 btrfs_put_block_group(cache);
4341         return 0;
4342 }
4343
4344 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4345                                struct btrfs_root *root)
4346 {
4347         struct btrfs_fs_info *fs_info = root->fs_info;
4348         struct extent_io_tree *unpin;
4349         struct btrfs_block_rsv *block_rsv;
4350         struct btrfs_block_rsv *next_rsv;
4351         u64 start;
4352         u64 end;
4353         int idx;
4354         int ret;
4355
4356         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4357                 unpin = &fs_info->freed_extents[1];
4358         else
4359                 unpin = &fs_info->freed_extents[0];
4360
4361         while (1) {
4362                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4363                                             EXTENT_DIRTY);
4364                 if (ret)
4365                         break;
4366
4367                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4368
4369                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4370                 unpin_extent_range(root, start, end);
4371                 cond_resched();
4372         }
4373
4374         mutex_lock(&fs_info->durable_block_rsv_mutex);
4375         list_for_each_entry_safe(block_rsv, next_rsv,
4376                                  &fs_info->durable_block_rsv_list, list) {
4377
4378                 idx = trans->transid & 0x1;
4379                 if (block_rsv->freed[idx] > 0) {
4380                         block_rsv_add_bytes(block_rsv,
4381                                             block_rsv->freed[idx], 0);
4382                         block_rsv->freed[idx] = 0;
4383                 }
4384                 if (atomic_read(&block_rsv->usage) == 0) {
4385                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4386
4387                         if (block_rsv->freed[0] == 0 &&
4388                             block_rsv->freed[1] == 0) {
4389                                 list_del_init(&block_rsv->list);
4390                                 kfree(block_rsv);
4391                         }
4392                 } else {
4393                         btrfs_block_rsv_release(root, block_rsv, 0);
4394                 }
4395         }
4396         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4397
4398         return 0;
4399 }
4400
4401 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4402                                 struct btrfs_root *root,
4403                                 u64 bytenr, u64 num_bytes, u64 parent,
4404                                 u64 root_objectid, u64 owner_objectid,
4405                                 u64 owner_offset, int refs_to_drop,
4406                                 struct btrfs_delayed_extent_op *extent_op)
4407 {
4408         struct btrfs_key key;
4409         struct btrfs_path *path;
4410         struct btrfs_fs_info *info = root->fs_info;
4411         struct btrfs_root *extent_root = info->extent_root;
4412         struct extent_buffer *leaf;
4413         struct btrfs_extent_item *ei;
4414         struct btrfs_extent_inline_ref *iref;
4415         int ret;
4416         int is_data;
4417         int extent_slot = 0;
4418         int found_extent = 0;
4419         int num_to_del = 1;
4420         u32 item_size;
4421         u64 refs;
4422
4423         path = btrfs_alloc_path();
4424         if (!path)
4425                 return -ENOMEM;
4426
4427         path->reada = 1;
4428         path->leave_spinning = 1;
4429
4430         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4431         BUG_ON(!is_data && refs_to_drop != 1);
4432
4433         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4434                                     bytenr, num_bytes, parent,
4435                                     root_objectid, owner_objectid,
4436                                     owner_offset);
4437         if (ret == 0) {
4438                 extent_slot = path->slots[0];
4439                 while (extent_slot >= 0) {
4440                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4441                                               extent_slot);
4442                         if (key.objectid != bytenr)
4443                                 break;
4444                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4445                             key.offset == num_bytes) {
4446                                 found_extent = 1;
4447                                 break;
4448                         }
4449                         if (path->slots[0] - extent_slot > 5)
4450                                 break;
4451                         extent_slot--;
4452                 }
4453 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4454                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4455                 if (found_extent && item_size < sizeof(*ei))
4456                         found_extent = 0;
4457 #endif
4458                 if (!found_extent) {
4459                         BUG_ON(iref);
4460                         ret = remove_extent_backref(trans, extent_root, path,
4461                                                     NULL, refs_to_drop,
4462                                                     is_data);
4463                         BUG_ON(ret);
4464                         btrfs_release_path(extent_root, path);
4465                         path->leave_spinning = 1;
4466
4467                         key.objectid = bytenr;
4468                         key.type = BTRFS_EXTENT_ITEM_KEY;
4469                         key.offset = num_bytes;
4470
4471                         ret = btrfs_search_slot(trans, extent_root,
4472                                                 &key, path, -1, 1);
4473                         if (ret) {
4474                                 printk(KERN_ERR "umm, got %d back from search"
4475                                        ", was looking for %llu\n", ret,
4476                                        (unsigned long long)bytenr);
4477                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4478                         }
4479                         BUG_ON(ret);
4480                         extent_slot = path->slots[0];
4481                 }
4482         } else {
4483                 btrfs_print_leaf(extent_root, path->nodes[0]);
4484                 WARN_ON(1);
4485                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4486                        "parent %llu root %llu  owner %llu offset %llu\n",
4487                        (unsigned long long)bytenr,
4488                        (unsigned long long)parent,
4489                        (unsigned long long)root_objectid,
4490                        (unsigned long long)owner_objectid,
4491                        (unsigned long long)owner_offset);
4492         }
4493
4494         leaf = path->nodes[0];
4495         item_size = btrfs_item_size_nr(leaf, extent_slot);
4496 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4497         if (item_size < sizeof(*ei)) {
4498                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4499                 ret = convert_extent_item_v0(trans, extent_root, path,
4500                                              owner_objectid, 0);
4501                 BUG_ON(ret < 0);
4502
4503                 btrfs_release_path(extent_root, path);
4504                 path->leave_spinning = 1;
4505
4506                 key.objectid = bytenr;
4507                 key.type = BTRFS_EXTENT_ITEM_KEY;
4508                 key.offset = num_bytes;
4509
4510                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4511                                         -1, 1);
4512                 if (ret) {
4513                         printk(KERN_ERR "umm, got %d back from search"
4514                                ", was looking for %llu\n", ret,
4515                                (unsigned long long)bytenr);
4516                         btrfs_print_leaf(extent_root, path->nodes[0]);
4517                 }
4518                 BUG_ON(ret);
4519                 extent_slot = path->slots[0];
4520                 leaf = path->nodes[0];
4521                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4522         }
4523 #endif
4524         BUG_ON(item_size < sizeof(*ei));
4525         ei = btrfs_item_ptr(leaf, extent_slot,
4526                             struct btrfs_extent_item);
4527         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4528                 struct btrfs_tree_block_info *bi;
4529                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4530                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4531                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4532         }
4533
4534         refs = btrfs_extent_refs(leaf, ei);
4535         BUG_ON(refs < refs_to_drop);
4536         refs -= refs_to_drop;
4537
4538         if (refs > 0) {
4539                 if (extent_op)
4540                         __run_delayed_extent_op(extent_op, leaf, ei);
4541                 /*
4542                  * In the case of inline back ref, reference count will
4543                  * be updated by remove_extent_backref
4544                  */
4545                 if (iref) {
4546                         BUG_ON(!found_extent);
4547                 } else {
4548                         btrfs_set_extent_refs(leaf, ei, refs);
4549                         btrfs_mark_buffer_dirty(leaf);
4550                 }
4551                 if (found_extent) {
4552                         ret = remove_extent_backref(trans, extent_root, path,
4553                                                     iref, refs_to_drop,
4554                                                     is_data);
4555                         BUG_ON(ret);
4556                 }
4557         } else {
4558                 if (found_extent) {
4559                         BUG_ON(is_data && refs_to_drop !=
4560                                extent_data_ref_count(root, path, iref));
4561                         if (iref) {
4562                                 BUG_ON(path->slots[0] != extent_slot);
4563                         } else {
4564                                 BUG_ON(path->slots[0] != extent_slot + 1);
4565                                 path->slots[0] = extent_slot;
4566                                 num_to_del = 2;
4567                         }
4568                 }
4569
4570                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4571                                       num_to_del);
4572                 BUG_ON(ret);
4573                 btrfs_release_path(extent_root, path);
4574
4575                 if (is_data) {
4576                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4577                         BUG_ON(ret);
4578                 } else {
4579                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4580                              bytenr >> PAGE_CACHE_SHIFT,
4581                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4582                 }
4583
4584                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4585                 BUG_ON(ret);
4586         }
4587         btrfs_free_path(path);
4588         return ret;
4589 }
4590
4591 /*
4592  * when we free an block, it is possible (and likely) that we free the last
4593  * delayed ref for that extent as well.  This searches the delayed ref tree for
4594  * a given extent, and if there are no other delayed refs to be processed, it
4595  * removes it from the tree.
4596  */
4597 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4598                                       struct btrfs_root *root, u64 bytenr)
4599 {
4600         struct btrfs_delayed_ref_head *head;
4601         struct btrfs_delayed_ref_root *delayed_refs;
4602         struct btrfs_delayed_ref_node *ref;
4603         struct rb_node *node;
4604         int ret = 0;
4605
4606         delayed_refs = &trans->transaction->delayed_refs;
4607         spin_lock(&delayed_refs->lock);
4608         head = btrfs_find_delayed_ref_head(trans, bytenr);
4609         if (!head)
4610                 goto out;
4611
4612         node = rb_prev(&head->node.rb_node);
4613         if (!node)
4614                 goto out;
4615
4616         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4617
4618         /* there are still entries for this ref, we can't drop it */
4619         if (ref->bytenr == bytenr)
4620                 goto out;
4621
4622         if (head->extent_op) {
4623                 if (!head->must_insert_reserved)
4624                         goto out;
4625                 kfree(head->extent_op);
4626                 head->extent_op = NULL;
4627         }
4628
4629         /*
4630          * waiting for the lock here would deadlock.  If someone else has it
4631          * locked they are already in the process of dropping it anyway
4632          */
4633         if (!mutex_trylock(&head->mutex))
4634                 goto out;
4635
4636         /*
4637          * at this point we have a head with no other entries.  Go
4638          * ahead and process it.
4639          */
4640         head->node.in_tree = 0;
4641         rb_erase(&head->node.rb_node, &delayed_refs->root);
4642
4643         delayed_refs->num_entries--;
4644
4645         /*
4646          * we don't take a ref on the node because we're removing it from the
4647          * tree, so we just steal the ref the tree was holding.
4648          */
4649         delayed_refs->num_heads--;
4650         if (list_empty(&head->cluster))
4651                 delayed_refs->num_heads_ready--;
4652
4653         list_del_init(&head->cluster);
4654         spin_unlock(&delayed_refs->lock);
4655
4656         BUG_ON(head->extent_op);
4657         if (head->must_insert_reserved)
4658                 ret = 1;
4659
4660         mutex_unlock(&head->mutex);
4661         btrfs_put_delayed_ref(&head->node);
4662         return ret;
4663 out:
4664         spin_unlock(&delayed_refs->lock);
4665         return 0;
4666 }
4667
4668 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4669                            struct btrfs_root *root,
4670                            struct extent_buffer *buf,
4671                            u64 parent, int last_ref)
4672 {
4673         struct btrfs_block_rsv *block_rsv;
4674         struct btrfs_block_group_cache *cache = NULL;
4675         int ret;
4676
4677         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4678                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4679                                                 parent, root->root_key.objectid,
4680                                                 btrfs_header_level(buf),
4681                                                 BTRFS_DROP_DELAYED_REF, NULL);
4682                 BUG_ON(ret);
4683         }
4684
4685         if (!last_ref)
4686                 return;
4687
4688         block_rsv = get_block_rsv(trans, root);
4689         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4690         if (block_rsv->space_info != cache->space_info)
4691                 goto out;
4692
4693         if (btrfs_header_generation(buf) == trans->transid) {
4694                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4695                         ret = check_ref_cleanup(trans, root, buf->start);
4696                         if (!ret)
4697                                 goto pin;
4698                 }
4699
4700                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4701                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4702                         goto pin;
4703                 }
4704
4705                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4706
4707                 btrfs_add_free_space(cache, buf->start, buf->len);
4708                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4709                 if (ret == -EAGAIN) {
4710                         /* block group became read-only */
4711                         update_reserved_bytes(cache, buf->len, 0, 1);
4712                         goto out;
4713                 }
4714
4715                 ret = 1;
4716                 spin_lock(&block_rsv->lock);
4717                 if (block_rsv->reserved < block_rsv->size) {
4718                         block_rsv->reserved += buf->len;
4719                         ret = 0;
4720                 }
4721                 spin_unlock(&block_rsv->lock);
4722
4723                 if (ret) {
4724                         spin_lock(&cache->space_info->lock);
4725                         cache->space_info->bytes_reserved -= buf->len;
4726                         cache->space_info->reservation_progress++;
4727                         spin_unlock(&cache->space_info->lock);
4728                 }
4729                 goto out;
4730         }
4731 pin:
4732         if (block_rsv->durable && !cache->ro) {
4733                 ret = 0;
4734                 spin_lock(&cache->lock);
4735                 if (!cache->ro) {
4736                         cache->reserved_pinned += buf->len;
4737                         ret = 1;
4738                 }
4739                 spin_unlock(&cache->lock);
4740
4741                 if (ret) {
4742                         spin_lock(&block_rsv->lock);
4743                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4744                         spin_unlock(&block_rsv->lock);
4745                 }
4746         }
4747 out:
4748         btrfs_put_block_group(cache);
4749 }
4750
4751 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4752                       struct btrfs_root *root,
4753                       u64 bytenr, u64 num_bytes, u64 parent,
4754                       u64 root_objectid, u64 owner, u64 offset)
4755 {
4756         int ret;
4757
4758         /*
4759          * tree log blocks never actually go into the extent allocation
4760          * tree, just update pinning info and exit early.
4761          */
4762         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4763                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4764                 /* unlocks the pinned mutex */
4765                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4766                 ret = 0;
4767         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4768                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4769                                         parent, root_objectid, (int)owner,
4770                                         BTRFS_DROP_DELAYED_REF, NULL);
4771                 BUG_ON(ret);
4772         } else {
4773                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4774                                         parent, root_objectid, owner,
4775                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4776                 BUG_ON(ret);
4777         }
4778         return ret;
4779 }
4780
4781 static u64 stripe_align(struct btrfs_root *root, u64 val)
4782 {
4783         u64 mask = ((u64)root->stripesize - 1);
4784         u64 ret = (val + mask) & ~mask;
4785         return ret;
4786 }
4787
4788 /*
4789  * when we wait for progress in the block group caching, its because
4790  * our allocation attempt failed at least once.  So, we must sleep
4791  * and let some progress happen before we try again.
4792  *
4793  * This function will sleep at least once waiting for new free space to
4794  * show up, and then it will check the block group free space numbers
4795  * for our min num_bytes.  Another option is to have it go ahead
4796  * and look in the rbtree for a free extent of a given size, but this
4797  * is a good start.
4798  */
4799 static noinline int
4800 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4801                                 u64 num_bytes)
4802 {
4803         struct btrfs_caching_control *caching_ctl;
4804         DEFINE_WAIT(wait);
4805
4806         caching_ctl = get_caching_control(cache);
4807         if (!caching_ctl)
4808                 return 0;
4809
4810         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4811                    (cache->free_space >= num_bytes));
4812
4813         put_caching_control(caching_ctl);
4814         return 0;
4815 }
4816
4817 static noinline int
4818 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4819 {
4820         struct btrfs_caching_control *caching_ctl;
4821         DEFINE_WAIT(wait);
4822
4823         caching_ctl = get_caching_control(cache);
4824         if (!caching_ctl)
4825                 return 0;
4826
4827         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4828
4829         put_caching_control(caching_ctl);
4830         return 0;
4831 }
4832
4833 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4834 {
4835         int index;
4836         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4837                 index = 0;
4838         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4839                 index = 1;
4840         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4841                 index = 2;
4842         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4843                 index = 3;
4844         else
4845                 index = 4;
4846         return index;
4847 }
4848
4849 enum btrfs_loop_type {
4850         LOOP_FIND_IDEAL = 0,
4851         LOOP_CACHING_NOWAIT = 1,
4852         LOOP_CACHING_WAIT = 2,
4853         LOOP_ALLOC_CHUNK = 3,
4854         LOOP_NO_EMPTY_SIZE = 4,
4855 };
4856
4857 /*
4858  * walks the btree of allocated extents and find a hole of a given size.
4859  * The key ins is changed to record the hole:
4860  * ins->objectid == block start
4861  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4862  * ins->offset == number of blocks
4863  * Any available blocks before search_start are skipped.
4864  */
4865 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4866                                      struct btrfs_root *orig_root,
4867                                      u64 num_bytes, u64 empty_size,
4868                                      u64 search_start, u64 search_end,
4869                                      u64 hint_byte, struct btrfs_key *ins,
4870                                      int data)
4871 {
4872         int ret = 0;
4873         struct btrfs_root *root = orig_root->fs_info->extent_root;
4874         struct btrfs_free_cluster *last_ptr = NULL;
4875         struct btrfs_block_group_cache *block_group = NULL;
4876         int empty_cluster = 2 * 1024 * 1024;
4877         int allowed_chunk_alloc = 0;
4878         int done_chunk_alloc = 0;
4879         struct btrfs_space_info *space_info;
4880         int last_ptr_loop = 0;
4881         int loop = 0;
4882         int index = 0;
4883         bool found_uncached_bg = false;
4884         bool failed_cluster_refill = false;
4885         bool failed_alloc = false;
4886         bool use_cluster = true;
4887         u64 ideal_cache_percent = 0;
4888         u64 ideal_cache_offset = 0;
4889
4890         WARN_ON(num_bytes < root->sectorsize);
4891         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4892         ins->objectid = 0;
4893         ins->offset = 0;
4894
4895         space_info = __find_space_info(root->fs_info, data);
4896         if (!space_info) {
4897                 printk(KERN_ERR "No space info for %d\n", data);
4898                 return -ENOSPC;
4899         }
4900
4901         /*
4902          * If the space info is for both data and metadata it means we have a
4903          * small filesystem and we can't use the clustering stuff.
4904          */
4905         if (btrfs_mixed_space_info(space_info))
4906                 use_cluster = false;
4907
4908         if (orig_root->ref_cows || empty_size)
4909                 allowed_chunk_alloc = 1;
4910
4911         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4912                 last_ptr = &root->fs_info->meta_alloc_cluster;
4913                 if (!btrfs_test_opt(root, SSD))
4914                         empty_cluster = 64 * 1024;
4915         }
4916
4917         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4918             btrfs_test_opt(root, SSD)) {
4919                 last_ptr = &root->fs_info->data_alloc_cluster;
4920         }
4921
4922         if (last_ptr) {
4923                 spin_lock(&last_ptr->lock);
4924                 if (last_ptr->block_group)
4925                         hint_byte = last_ptr->window_start;
4926                 spin_unlock(&last_ptr->lock);
4927         }
4928
4929         search_start = max(search_start, first_logical_byte(root, 0));
4930         search_start = max(search_start, hint_byte);
4931
4932         if (!last_ptr)
4933                 empty_cluster = 0;
4934
4935         if (search_start == hint_byte) {
4936 ideal_cache:
4937                 block_group = btrfs_lookup_block_group(root->fs_info,
4938                                                        search_start);
4939                 /*
4940                  * we don't want to use the block group if it doesn't match our
4941                  * allocation bits, or if its not cached.
4942                  *
4943                  * However if we are re-searching with an ideal block group
4944                  * picked out then we don't care that the block group is cached.
4945                  */
4946                 if (block_group && block_group_bits(block_group, data) &&
4947                     (block_group->cached != BTRFS_CACHE_NO ||
4948                      search_start == ideal_cache_offset)) {
4949                         down_read(&space_info->groups_sem);
4950                         if (list_empty(&block_group->list) ||
4951                             block_group->ro) {
4952                                 /*
4953                                  * someone is removing this block group,
4954                                  * we can't jump into the have_block_group
4955                                  * target because our list pointers are not
4956                                  * valid
4957                                  */
4958                                 btrfs_put_block_group(block_group);
4959                                 up_read(&space_info->groups_sem);
4960                         } else {
4961                                 index = get_block_group_index(block_group);
4962                                 goto have_block_group;
4963                         }
4964                 } else if (block_group) {
4965                         btrfs_put_block_group(block_group);
4966                 }
4967         }
4968 search:
4969         down_read(&space_info->groups_sem);
4970         list_for_each_entry(block_group, &space_info->block_groups[index],
4971                             list) {
4972                 u64 offset;
4973                 int cached;
4974
4975                 btrfs_get_block_group(block_group);
4976                 search_start = block_group->key.objectid;
4977
4978                 /*
4979                  * this can happen if we end up cycling through all the
4980                  * raid types, but we want to make sure we only allocate
4981                  * for the proper type.
4982                  */
4983                 if (!block_group_bits(block_group, data)) {
4984                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
4985                                 BTRFS_BLOCK_GROUP_RAID1 |
4986                                 BTRFS_BLOCK_GROUP_RAID10;
4987
4988                         /*
4989                          * if they asked for extra copies and this block group
4990                          * doesn't provide them, bail.  This does allow us to
4991                          * fill raid0 from raid1.
4992                          */
4993                         if ((data & extra) && !(block_group->flags & extra))
4994                                 goto loop;
4995                 }
4996
4997 have_block_group:
4998                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4999                         u64 free_percent;
5000
5001                         ret = cache_block_group(block_group, trans,
5002                                                 orig_root, 1);
5003                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5004                                 goto have_block_group;
5005
5006                         free_percent = btrfs_block_group_used(&block_group->item);
5007                         free_percent *= 100;
5008                         free_percent = div64_u64(free_percent,
5009                                                  block_group->key.offset);
5010                         free_percent = 100 - free_percent;
5011                         if (free_percent > ideal_cache_percent &&
5012                             likely(!block_group->ro)) {
5013                                 ideal_cache_offset = block_group->key.objectid;
5014                                 ideal_cache_percent = free_percent;
5015                         }
5016
5017                         /*
5018                          * We only want to start kthread caching if we are at
5019                          * the point where we will wait for caching to make
5020                          * progress, or if our ideal search is over and we've
5021                          * found somebody to start caching.
5022                          */
5023                         if (loop > LOOP_CACHING_NOWAIT ||
5024                             (loop > LOOP_FIND_IDEAL &&
5025                              atomic_read(&space_info->caching_threads) < 2)) {
5026                                 ret = cache_block_group(block_group, trans,
5027                                                         orig_root, 0);
5028                                 BUG_ON(ret);
5029                         }
5030                         found_uncached_bg = true;
5031
5032                         /*
5033                          * If loop is set for cached only, try the next block
5034                          * group.
5035                          */
5036                         if (loop == LOOP_FIND_IDEAL)
5037                                 goto loop;
5038                 }
5039
5040                 cached = block_group_cache_done(block_group);
5041                 if (unlikely(!cached))
5042                         found_uncached_bg = true;
5043
5044                 if (unlikely(block_group->ro))
5045                         goto loop;
5046
5047                 /*
5048                  * Ok we want to try and use the cluster allocator, so lets look
5049                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5050                  * have tried the cluster allocator plenty of times at this
5051                  * point and not have found anything, so we are likely way too
5052                  * fragmented for the clustering stuff to find anything, so lets
5053                  * just skip it and let the allocator find whatever block it can
5054                  * find
5055                  */
5056                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5057                         /*
5058                          * the refill lock keeps out other
5059                          * people trying to start a new cluster
5060                          */
5061                         spin_lock(&last_ptr->refill_lock);
5062                         if (last_ptr->block_group &&
5063                             (last_ptr->block_group->ro ||
5064                             !block_group_bits(last_ptr->block_group, data))) {
5065                                 offset = 0;
5066                                 goto refill_cluster;
5067                         }
5068
5069                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5070                                                  num_bytes, search_start);
5071                         if (offset) {
5072                                 /* we have a block, we're done */
5073                                 spin_unlock(&last_ptr->refill_lock);
5074                                 goto checks;
5075                         }
5076
5077                         spin_lock(&last_ptr->lock);
5078                         /*
5079                          * whoops, this cluster doesn't actually point to
5080                          * this block group.  Get a ref on the block
5081                          * group is does point to and try again
5082                          */
5083                         if (!last_ptr_loop && last_ptr->block_group &&
5084                             last_ptr->block_group != block_group) {
5085
5086                                 btrfs_put_block_group(block_group);
5087                                 block_group = last_ptr->block_group;
5088                                 btrfs_get_block_group(block_group);
5089                                 spin_unlock(&last_ptr->lock);
5090                                 spin_unlock(&last_ptr->refill_lock);
5091
5092                                 last_ptr_loop = 1;
5093                                 search_start = block_group->key.objectid;
5094                                 /*
5095                                  * we know this block group is properly
5096                                  * in the list because
5097                                  * btrfs_remove_block_group, drops the
5098                                  * cluster before it removes the block
5099                                  * group from the list
5100                                  */
5101                                 goto have_block_group;
5102                         }
5103                         spin_unlock(&last_ptr->lock);
5104 refill_cluster:
5105                         /*
5106                          * this cluster didn't work out, free it and
5107                          * start over
5108                          */
5109                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5110
5111                         last_ptr_loop = 0;
5112
5113                         /* allocate a cluster in this block group */
5114                         ret = btrfs_find_space_cluster(trans, root,
5115                                                block_group, last_ptr,
5116                                                offset, num_bytes,
5117                                                empty_cluster + empty_size);
5118                         if (ret == 0) {
5119                                 /*
5120                                  * now pull our allocation out of this
5121                                  * cluster
5122                                  */
5123                                 offset = btrfs_alloc_from_cluster(block_group,
5124                                                   last_ptr, num_bytes,
5125                                                   search_start);
5126                                 if (offset) {
5127                                         /* we found one, proceed */
5128                                         spin_unlock(&last_ptr->refill_lock);
5129                                         goto checks;
5130                                 }
5131                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5132                                    && !failed_cluster_refill) {
5133                                 spin_unlock(&last_ptr->refill_lock);
5134
5135                                 failed_cluster_refill = true;
5136                                 wait_block_group_cache_progress(block_group,
5137                                        num_bytes + empty_cluster + empty_size);
5138                                 goto have_block_group;
5139                         }
5140
5141                         /*
5142                          * at this point we either didn't find a cluster
5143                          * or we weren't able to allocate a block from our
5144                          * cluster.  Free the cluster we've been trying
5145                          * to use, and go to the next block group
5146                          */
5147                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5148                         spin_unlock(&last_ptr->refill_lock);
5149                         goto loop;
5150                 }
5151
5152                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5153                                                     num_bytes, empty_size);
5154                 /*
5155                  * If we didn't find a chunk, and we haven't failed on this
5156                  * block group before, and this block group is in the middle of
5157                  * caching and we are ok with waiting, then go ahead and wait
5158                  * for progress to be made, and set failed_alloc to true.
5159                  *
5160                  * If failed_alloc is true then we've already waited on this
5161                  * block group once and should move on to the next block group.
5162                  */
5163                 if (!offset && !failed_alloc && !cached &&
5164                     loop > LOOP_CACHING_NOWAIT) {
5165                         wait_block_group_cache_progress(block_group,
5166                                                 num_bytes + empty_size);
5167                         failed_alloc = true;
5168                         goto have_block_group;
5169                 } else if (!offset) {
5170                         goto loop;
5171                 }
5172 checks:
5173                 search_start = stripe_align(root, offset);
5174                 /* move on to the next group */
5175                 if (search_start + num_bytes >= search_end) {
5176                         btrfs_add_free_space(block_group, offset, num_bytes);
5177                         goto loop;
5178                 }
5179
5180                 /* move on to the next group */
5181                 if (search_start + num_bytes >
5182                     block_group->key.objectid + block_group->key.offset) {
5183                         btrfs_add_free_space(block_group, offset, num_bytes);
5184                         goto loop;
5185                 }
5186
5187                 ins->objectid = search_start;
5188                 ins->offset = num_bytes;
5189
5190                 if (offset < search_start)
5191                         btrfs_add_free_space(block_group, offset,
5192                                              search_start - offset);
5193                 BUG_ON(offset > search_start);
5194
5195                 ret = update_reserved_bytes(block_group, num_bytes, 1,
5196                                             (data & BTRFS_BLOCK_GROUP_DATA));
5197                 if (ret == -EAGAIN) {
5198                         btrfs_add_free_space(block_group, offset, num_bytes);
5199                         goto loop;
5200                 }
5201
5202                 /* we are all good, lets return */
5203                 ins->objectid = search_start;
5204                 ins->offset = num_bytes;
5205
5206                 if (offset < search_start)
5207                         btrfs_add_free_space(block_group, offset,
5208                                              search_start - offset);
5209                 BUG_ON(offset > search_start);
5210                 break;
5211 loop:
5212                 failed_cluster_refill = false;
5213                 failed_alloc = false;
5214                 BUG_ON(index != get_block_group_index(block_group));
5215                 btrfs_put_block_group(block_group);
5216         }
5217         up_read(&space_info->groups_sem);
5218
5219         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5220                 goto search;
5221
5222         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5223          *                      for them to make caching progress.  Also
5224          *                      determine the best possible bg to cache
5225          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5226          *                      caching kthreads as we move along
5227          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5228          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5229          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5230          *                      again
5231          */
5232         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5233             (found_uncached_bg || empty_size || empty_cluster ||
5234              allowed_chunk_alloc)) {
5235                 index = 0;
5236                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5237                         found_uncached_bg = false;
5238                         loop++;
5239                         if (!ideal_cache_percent &&
5240                             atomic_read(&space_info->caching_threads))
5241                                 goto search;
5242
5243                         /*
5244                          * 1 of the following 2 things have happened so far
5245                          *
5246                          * 1) We found an ideal block group for caching that
5247                          * is mostly full and will cache quickly, so we might
5248                          * as well wait for it.
5249                          *
5250                          * 2) We searched for cached only and we didn't find
5251                          * anything, and we didn't start any caching kthreads
5252                          * either, so chances are we will loop through and
5253                          * start a couple caching kthreads, and then come back
5254                          * around and just wait for them.  This will be slower
5255                          * because we will have 2 caching kthreads reading at
5256                          * the same time when we could have just started one
5257                          * and waited for it to get far enough to give us an
5258                          * allocation, so go ahead and go to the wait caching
5259                          * loop.
5260                          */
5261                         loop = LOOP_CACHING_WAIT;
5262                         search_start = ideal_cache_offset;
5263                         ideal_cache_percent = 0;
5264                         goto ideal_cache;
5265                 } else if (loop == LOOP_FIND_IDEAL) {
5266                         /*
5267                          * Didn't find a uncached bg, wait on anything we find
5268                          * next.
5269                          */
5270                         loop = LOOP_CACHING_WAIT;
5271                         goto search;
5272                 }
5273
5274                 if (loop < LOOP_CACHING_WAIT) {
5275                         loop++;
5276                         goto search;
5277                 }
5278
5279                 if (loop == LOOP_ALLOC_CHUNK) {
5280                         empty_size = 0;
5281                         empty_cluster = 0;
5282                 }
5283
5284                 if (allowed_chunk_alloc) {
5285                         ret = do_chunk_alloc(trans, root, num_bytes +
5286                                              2 * 1024 * 1024, data, 1);
5287                         allowed_chunk_alloc = 0;
5288                         done_chunk_alloc = 1;
5289                 } else if (!done_chunk_alloc) {
5290                         space_info->force_alloc = 1;
5291                 }
5292
5293                 if (loop < LOOP_NO_EMPTY_SIZE) {
5294                         loop++;
5295                         goto search;
5296                 }
5297                 ret = -ENOSPC;
5298         } else if (!ins->objectid) {
5299                 ret = -ENOSPC;
5300         }
5301
5302         /* we found what we needed */
5303         if (ins->objectid) {
5304                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5305                         trans->block_group = block_group->key.objectid;
5306
5307                 btrfs_put_block_group(block_group);
5308                 ret = 0;
5309         }
5310
5311         return ret;
5312 }
5313
5314 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5315                             int dump_block_groups)
5316 {
5317         struct btrfs_block_group_cache *cache;
5318         int index = 0;
5319
5320         spin_lock(&info->lock);
5321         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5322                (unsigned long long)(info->total_bytes - info->bytes_used -
5323                                     info->bytes_pinned - info->bytes_reserved -
5324                                     info->bytes_readonly),
5325                (info->full) ? "" : "not ");
5326         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5327                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5328                (unsigned long long)info->total_bytes,
5329                (unsigned long long)info->bytes_used,
5330                (unsigned long long)info->bytes_pinned,
5331                (unsigned long long)info->bytes_reserved,
5332                (unsigned long long)info->bytes_may_use,
5333                (unsigned long long)info->bytes_readonly);
5334         spin_unlock(&info->lock);
5335
5336         if (!dump_block_groups)
5337                 return;
5338
5339         down_read(&info->groups_sem);
5340 again:
5341         list_for_each_entry(cache, &info->block_groups[index], list) {
5342                 spin_lock(&cache->lock);
5343                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5344                        "%llu pinned %llu reserved\n",
5345                        (unsigned long long)cache->key.objectid,
5346                        (unsigned long long)cache->key.offset,
5347                        (unsigned long long)btrfs_block_group_used(&cache->item),
5348                        (unsigned long long)cache->pinned,
5349                        (unsigned long long)cache->reserved);
5350                 btrfs_dump_free_space(cache, bytes);
5351                 spin_unlock(&cache->lock);
5352         }
5353         if (++index < BTRFS_NR_RAID_TYPES)
5354                 goto again;
5355         up_read(&info->groups_sem);
5356 }
5357
5358 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5359                          struct btrfs_root *root,
5360                          u64 num_bytes, u64 min_alloc_size,
5361                          u64 empty_size, u64 hint_byte,
5362                          u64 search_end, struct btrfs_key *ins,
5363                          u64 data)
5364 {
5365         int ret;
5366         u64 search_start = 0;
5367
5368         data = btrfs_get_alloc_profile(root, data);
5369 again:
5370         /*
5371          * the only place that sets empty_size is btrfs_realloc_node, which
5372          * is not called recursively on allocations
5373          */
5374         if (empty_size || root->ref_cows)
5375                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5376                                      num_bytes + 2 * 1024 * 1024, data, 0);
5377
5378         WARN_ON(num_bytes < root->sectorsize);
5379         ret = find_free_extent(trans, root, num_bytes, empty_size,
5380                                search_start, search_end, hint_byte,
5381                                ins, data);
5382
5383         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5384                 num_bytes = num_bytes >> 1;
5385                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5386                 num_bytes = max(num_bytes, min_alloc_size);
5387                 do_chunk_alloc(trans, root->fs_info->extent_root,
5388                                num_bytes, data, 1);
5389                 goto again;
5390         }
5391         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5392                 struct btrfs_space_info *sinfo;
5393
5394                 sinfo = __find_space_info(root->fs_info, data);
5395                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5396                        "wanted %llu\n", (unsigned long long)data,
5397                        (unsigned long long)num_bytes);
5398                 dump_space_info(sinfo, num_bytes, 1);
5399         }
5400
5401         return ret;
5402 }
5403
5404 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5405 {
5406         struct btrfs_block_group_cache *cache;
5407         int ret = 0;
5408
5409         cache = btrfs_lookup_block_group(root->fs_info, start);
5410         if (!cache) {
5411                 printk(KERN_ERR "Unable to find block group for %llu\n",
5412                        (unsigned long long)start);
5413                 return -ENOSPC;
5414         }
5415
5416         ret = btrfs_discard_extent(root, start, len);
5417
5418         btrfs_add_free_space(cache, start, len);
5419         update_reserved_bytes(cache, len, 0, 1);
5420         btrfs_put_block_group(cache);
5421
5422         return ret;
5423 }
5424
5425 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5426                                       struct btrfs_root *root,
5427                                       u64 parent, u64 root_objectid,
5428                                       u64 flags, u64 owner, u64 offset,
5429                                       struct btrfs_key *ins, int ref_mod)
5430 {
5431         int ret;
5432         struct btrfs_fs_info *fs_info = root->fs_info;
5433         struct btrfs_extent_item *extent_item;
5434         struct btrfs_extent_inline_ref *iref;
5435         struct btrfs_path *path;
5436         struct extent_buffer *leaf;
5437         int type;
5438         u32 size;
5439
5440         if (parent > 0)
5441                 type = BTRFS_SHARED_DATA_REF_KEY;
5442         else
5443                 type = BTRFS_EXTENT_DATA_REF_KEY;
5444
5445         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5446
5447         path = btrfs_alloc_path();
5448         BUG_ON(!path);
5449
5450         path->leave_spinning = 1;
5451         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5452                                       ins, size);
5453         BUG_ON(ret);
5454
5455         leaf = path->nodes[0];
5456         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5457                                      struct btrfs_extent_item);
5458         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5459         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5460         btrfs_set_extent_flags(leaf, extent_item,
5461                                flags | BTRFS_EXTENT_FLAG_DATA);
5462
5463         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5464         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5465         if (parent > 0) {
5466                 struct btrfs_shared_data_ref *ref;
5467                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5468                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5469                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5470         } else {
5471                 struct btrfs_extent_data_ref *ref;
5472                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5473                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5474                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5475                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5476                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5477         }
5478
5479         btrfs_mark_buffer_dirty(path->nodes[0]);
5480         btrfs_free_path(path);
5481
5482         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5483         if (ret) {
5484                 printk(KERN_ERR "btrfs update block group failed for %llu "
5485                        "%llu\n", (unsigned long long)ins->objectid,
5486                        (unsigned long long)ins->offset);
5487                 BUG();
5488         }
5489         return ret;
5490 }
5491
5492 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5493                                      struct btrfs_root *root,
5494                                      u64 parent, u64 root_objectid,
5495                                      u64 flags, struct btrfs_disk_key *key,
5496                                      int level, struct btrfs_key *ins)
5497 {
5498         int ret;
5499         struct btrfs_fs_info *fs_info = root->fs_info;
5500         struct btrfs_extent_item *extent_item;
5501         struct btrfs_tree_block_info *block_info;
5502         struct btrfs_extent_inline_ref *iref;
5503         struct btrfs_path *path;
5504         struct extent_buffer *leaf;
5505         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5506
5507         path = btrfs_alloc_path();
5508         BUG_ON(!path);
5509
5510         path->leave_spinning = 1;
5511         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5512                                       ins, size);
5513         BUG_ON(ret);
5514
5515         leaf = path->nodes[0];
5516         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5517                                      struct btrfs_extent_item);
5518         btrfs_set_extent_refs(leaf, extent_item, 1);
5519         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5520         btrfs_set_extent_flags(leaf, extent_item,
5521                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5522         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5523
5524         btrfs_set_tree_block_key(leaf, block_info, key);
5525         btrfs_set_tree_block_level(leaf, block_info, level);
5526
5527         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5528         if (parent > 0) {
5529                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5530                 btrfs_set_extent_inline_ref_type(leaf, iref,
5531                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5532                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5533         } else {
5534                 btrfs_set_extent_inline_ref_type(leaf, iref,
5535                                                  BTRFS_TREE_BLOCK_REF_KEY);
5536                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5537         }
5538
5539         btrfs_mark_buffer_dirty(leaf);
5540         btrfs_free_path(path);
5541
5542         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5543         if (ret) {
5544                 printk(KERN_ERR "btrfs update block group failed for %llu "
5545                        "%llu\n", (unsigned long long)ins->objectid,
5546                        (unsigned long long)ins->offset);
5547                 BUG();
5548         }
5549         return ret;
5550 }
5551
5552 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5553                                      struct btrfs_root *root,
5554                                      u64 root_objectid, u64 owner,
5555                                      u64 offset, struct btrfs_key *ins)
5556 {
5557         int ret;
5558
5559         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5560
5561         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5562                                          0, root_objectid, owner, offset,
5563                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5564         return ret;
5565 }
5566
5567 /*
5568  * this is used by the tree logging recovery code.  It records that
5569  * an extent has been allocated and makes sure to clear the free
5570  * space cache bits as well
5571  */
5572 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5573                                    struct btrfs_root *root,
5574                                    u64 root_objectid, u64 owner, u64 offset,
5575                                    struct btrfs_key *ins)
5576 {
5577         int ret;
5578         struct btrfs_block_group_cache *block_group;
5579         struct btrfs_caching_control *caching_ctl;
5580         u64 start = ins->objectid;
5581         u64 num_bytes = ins->offset;
5582
5583         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5584         cache_block_group(block_group, trans, NULL, 0);
5585         caching_ctl = get_caching_control(block_group);
5586
5587         if (!caching_ctl) {
5588                 BUG_ON(!block_group_cache_done(block_group));
5589                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5590                 BUG_ON(ret);
5591         } else {
5592                 mutex_lock(&caching_ctl->mutex);
5593
5594                 if (start >= caching_ctl->progress) {
5595                         ret = add_excluded_extent(root, start, num_bytes);
5596                         BUG_ON(ret);
5597                 } else if (start + num_bytes <= caching_ctl->progress) {
5598                         ret = btrfs_remove_free_space(block_group,
5599                                                       start, num_bytes);
5600                         BUG_ON(ret);
5601                 } else {
5602                         num_bytes = caching_ctl->progress - start;
5603                         ret = btrfs_remove_free_space(block_group,
5604                                                       start, num_bytes);
5605                         BUG_ON(ret);
5606
5607                         start = caching_ctl->progress;
5608                         num_bytes = ins->objectid + ins->offset -
5609                                     caching_ctl->progress;
5610                         ret = add_excluded_extent(root, start, num_bytes);
5611                         BUG_ON(ret);
5612                 }
5613
5614                 mutex_unlock(&caching_ctl->mutex);
5615                 put_caching_control(caching_ctl);
5616         }
5617
5618         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5619         BUG_ON(ret);
5620         btrfs_put_block_group(block_group);
5621         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5622                                          0, owner, offset, ins, 1);
5623         return ret;
5624 }
5625
5626 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5627                                             struct btrfs_root *root,
5628                                             u64 bytenr, u32 blocksize,
5629                                             int level)
5630 {
5631         struct extent_buffer *buf;
5632
5633         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5634         if (!buf)
5635                 return ERR_PTR(-ENOMEM);
5636         btrfs_set_header_generation(buf, trans->transid);
5637         btrfs_set_buffer_lockdep_class(buf, level);
5638         btrfs_tree_lock(buf);
5639         clean_tree_block(trans, root, buf);
5640
5641         btrfs_set_lock_blocking(buf);
5642         btrfs_set_buffer_uptodate(buf);
5643
5644         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5645                 /*
5646                  * we allow two log transactions at a time, use different
5647                  * EXENT bit to differentiate dirty pages.
5648                  */
5649                 if (root->log_transid % 2 == 0)
5650                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5651                                         buf->start + buf->len - 1, GFP_NOFS);
5652                 else
5653                         set_extent_new(&root->dirty_log_pages, buf->start,
5654                                         buf->start + buf->len - 1, GFP_NOFS);
5655         } else {
5656                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5657                          buf->start + buf->len - 1, GFP_NOFS);
5658         }
5659         trans->blocks_used++;
5660         /* this returns a buffer locked for blocking */
5661         return buf;
5662 }
5663
5664 static struct btrfs_block_rsv *
5665 use_block_rsv(struct btrfs_trans_handle *trans,
5666               struct btrfs_root *root, u32 blocksize)
5667 {
5668         struct btrfs_block_rsv *block_rsv;
5669         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5670         int ret;
5671
5672         block_rsv = get_block_rsv(trans, root);
5673
5674         if (block_rsv->size == 0) {
5675                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5676                                              blocksize, 0);
5677                 /*
5678                  * If we couldn't reserve metadata bytes try and use some from
5679                  * the global reserve.
5680                  */
5681                 if (ret && block_rsv != global_rsv) {
5682                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5683                         if (!ret)
5684                                 return global_rsv;
5685                         return ERR_PTR(ret);
5686                 } else if (ret) {
5687                         return ERR_PTR(ret);
5688                 }
5689                 return block_rsv;
5690         }
5691
5692         ret = block_rsv_use_bytes(block_rsv, blocksize);
5693         if (!ret)
5694                 return block_rsv;
5695         if (ret) {
5696                 WARN_ON(1);
5697                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5698                                              0);
5699                 if (!ret) {
5700                         spin_lock(&block_rsv->lock);
5701                         block_rsv->size += blocksize;
5702                         spin_unlock(&block_rsv->lock);
5703                         return block_rsv;
5704                 } else if (ret && block_rsv != global_rsv) {
5705                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5706                         if (!ret)
5707                                 return global_rsv;
5708                 }
5709         }
5710
5711         return ERR_PTR(-ENOSPC);
5712 }
5713
5714 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5715 {
5716         block_rsv_add_bytes(block_rsv, blocksize, 0);
5717         block_rsv_release_bytes(block_rsv, NULL, 0);
5718 }
5719
5720 /*
5721  * finds a free extent and does all the dirty work required for allocation
5722  * returns the key for the extent through ins, and a tree buffer for
5723  * the first block of the extent through buf.
5724  *
5725  * returns the tree buffer or NULL.
5726  */
5727 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5728                                         struct btrfs_root *root, u32 blocksize,
5729                                         u64 parent, u64 root_objectid,
5730                                         struct btrfs_disk_key *key, int level,
5731                                         u64 hint, u64 empty_size)
5732 {
5733         struct btrfs_key ins;
5734         struct btrfs_block_rsv *block_rsv;
5735         struct extent_buffer *buf;
5736         u64 flags = 0;
5737         int ret;
5738
5739
5740         block_rsv = use_block_rsv(trans, root, blocksize);
5741         if (IS_ERR(block_rsv))
5742                 return ERR_CAST(block_rsv);
5743
5744         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5745                                    empty_size, hint, (u64)-1, &ins, 0);
5746         if (ret) {
5747                 unuse_block_rsv(block_rsv, blocksize);
5748                 return ERR_PTR(ret);
5749         }
5750
5751         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5752                                     blocksize, level);
5753         BUG_ON(IS_ERR(buf));
5754
5755         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5756                 if (parent == 0)
5757                         parent = ins.objectid;
5758                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5759         } else
5760                 BUG_ON(parent > 0);
5761
5762         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5763                 struct btrfs_delayed_extent_op *extent_op;
5764                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5765                 BUG_ON(!extent_op);
5766                 if (key)
5767                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5768                 else
5769                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5770                 extent_op->flags_to_set = flags;
5771                 extent_op->update_key = 1;
5772                 extent_op->update_flags = 1;
5773                 extent_op->is_data = 0;
5774
5775                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5776                                         ins.offset, parent, root_objectid,
5777                                         level, BTRFS_ADD_DELAYED_EXTENT,
5778                                         extent_op);
5779                 BUG_ON(ret);
5780         }
5781         return buf;
5782 }
5783
5784 struct walk_control {
5785         u64 refs[BTRFS_MAX_LEVEL];
5786         u64 flags[BTRFS_MAX_LEVEL];
5787         struct btrfs_key update_progress;
5788         int stage;
5789         int level;
5790         int shared_level;
5791         int update_ref;
5792         int keep_locks;
5793         int reada_slot;
5794         int reada_count;
5795 };
5796
5797 #define DROP_REFERENCE  1
5798 #define UPDATE_BACKREF  2
5799
5800 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5801                                      struct btrfs_root *root,
5802                                      struct walk_control *wc,
5803                                      struct btrfs_path *path)
5804 {
5805         u64 bytenr;
5806         u64 generation;
5807         u64 refs;
5808         u64 flags;
5809         u32 nritems;
5810         u32 blocksize;
5811         struct btrfs_key key;
5812         struct extent_buffer *eb;
5813         int ret;
5814         int slot;
5815         int nread = 0;
5816
5817         if (path->slots[wc->level] < wc->reada_slot) {
5818                 wc->reada_count = wc->reada_count * 2 / 3;
5819                 wc->reada_count = max(wc->reada_count, 2);
5820         } else {
5821                 wc->reada_count = wc->reada_count * 3 / 2;
5822                 wc->reada_count = min_t(int, wc->reada_count,
5823                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5824         }
5825
5826         eb = path->nodes[wc->level];
5827         nritems = btrfs_header_nritems(eb);
5828         blocksize = btrfs_level_size(root, wc->level - 1);
5829
5830         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5831                 if (nread >= wc->reada_count)
5832                         break;
5833
5834                 cond_resched();
5835                 bytenr = btrfs_node_blockptr(eb, slot);
5836                 generation = btrfs_node_ptr_generation(eb, slot);
5837
5838                 if (slot == path->slots[wc->level])
5839                         goto reada;
5840
5841                 if (wc->stage == UPDATE_BACKREF &&
5842                     generation <= root->root_key.offset)
5843                         continue;
5844
5845                 /* We don't lock the tree block, it's OK to be racy here */
5846                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5847                                                &refs, &flags);
5848                 BUG_ON(ret);
5849                 BUG_ON(refs == 0);
5850
5851                 if (wc->stage == DROP_REFERENCE) {
5852                         if (refs == 1)
5853                                 goto reada;
5854
5855                         if (wc->level == 1 &&
5856                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5857                                 continue;
5858                         if (!wc->update_ref ||
5859                             generation <= root->root_key.offset)
5860                                 continue;
5861                         btrfs_node_key_to_cpu(eb, &key, slot);
5862                         ret = btrfs_comp_cpu_keys(&key,
5863                                                   &wc->update_progress);
5864                         if (ret < 0)
5865                                 continue;
5866                 } else {
5867                         if (wc->level == 1 &&
5868                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5869                                 continue;
5870                 }
5871 reada:
5872                 ret = readahead_tree_block(root, bytenr, blocksize,
5873                                            generation);
5874                 if (ret)
5875                         break;
5876                 nread++;
5877         }
5878         wc->reada_slot = slot;
5879 }
5880
5881 /*
5882  * hepler to process tree block while walking down the tree.
5883  *
5884  * when wc->stage == UPDATE_BACKREF, this function updates
5885  * back refs for pointers in the block.
5886  *
5887  * NOTE: return value 1 means we should stop walking down.
5888  */
5889 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5890                                    struct btrfs_root *root,
5891                                    struct btrfs_path *path,
5892                                    struct walk_control *wc, int lookup_info)
5893 {
5894         int level = wc->level;
5895         struct extent_buffer *eb = path->nodes[level];
5896         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5897         int ret;
5898
5899         if (wc->stage == UPDATE_BACKREF &&
5900             btrfs_header_owner(eb) != root->root_key.objectid)
5901                 return 1;
5902
5903         /*
5904          * when reference count of tree block is 1, it won't increase
5905          * again. once full backref flag is set, we never clear it.
5906          */
5907         if (lookup_info &&
5908             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5909              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5910                 BUG_ON(!path->locks[level]);
5911                 ret = btrfs_lookup_extent_info(trans, root,
5912                                                eb->start, eb->len,
5913                                                &wc->refs[level],
5914                                                &wc->flags[level]);
5915                 BUG_ON(ret);
5916                 BUG_ON(wc->refs[level] == 0);
5917         }
5918
5919         if (wc->stage == DROP_REFERENCE) {
5920                 if (wc->refs[level] > 1)
5921                         return 1;
5922
5923                 if (path->locks[level] && !wc->keep_locks) {
5924                         btrfs_tree_unlock(eb);
5925                         path->locks[level] = 0;
5926                 }
5927                 return 0;
5928         }
5929
5930         /* wc->stage == UPDATE_BACKREF */
5931         if (!(wc->flags[level] & flag)) {
5932                 BUG_ON(!path->locks[level]);
5933                 ret = btrfs_inc_ref(trans, root, eb, 1);
5934                 BUG_ON(ret);
5935                 ret = btrfs_dec_ref(trans, root, eb, 0);
5936                 BUG_ON(ret);
5937                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5938                                                   eb->len, flag, 0);
5939                 BUG_ON(ret);
5940                 wc->flags[level] |= flag;
5941         }
5942
5943         /*
5944          * the block is shared by multiple trees, so it's not good to
5945          * keep the tree lock
5946          */
5947         if (path->locks[level] && level > 0) {
5948                 btrfs_tree_unlock(eb);
5949                 path->locks[level] = 0;
5950         }
5951         return 0;
5952 }
5953
5954 /*
5955  * hepler to process tree block pointer.
5956  *
5957  * when wc->stage == DROP_REFERENCE, this function checks
5958  * reference count of the block pointed to. if the block
5959  * is shared and we need update back refs for the subtree
5960  * rooted at the block, this function changes wc->stage to
5961  * UPDATE_BACKREF. if the block is shared and there is no
5962  * need to update back, this function drops the reference
5963  * to the block.
5964  *
5965  * NOTE: return value 1 means we should stop walking down.
5966  */
5967 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5968                                  struct btrfs_root *root,
5969                                  struct btrfs_path *path,
5970                                  struct walk_control *wc, int *lookup_info)
5971 {
5972         u64 bytenr;
5973         u64 generation;
5974         u64 parent;
5975         u32 blocksize;
5976         struct btrfs_key key;
5977         struct extent_buffer *next;
5978         int level = wc->level;
5979         int reada = 0;
5980         int ret = 0;
5981
5982         generation = btrfs_node_ptr_generation(path->nodes[level],
5983                                                path->slots[level]);
5984         /*
5985          * if the lower level block was created before the snapshot
5986          * was created, we know there is no need to update back refs
5987          * for the subtree
5988          */
5989         if (wc->stage == UPDATE_BACKREF &&
5990             generation <= root->root_key.offset) {
5991                 *lookup_info = 1;
5992                 return 1;
5993         }
5994
5995         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5996         blocksize = btrfs_level_size(root, level - 1);
5997
5998         next = btrfs_find_tree_block(root, bytenr, blocksize);
5999         if (!next) {
6000                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6001                 if (!next)
6002                         return -ENOMEM;
6003                 reada = 1;
6004         }
6005         btrfs_tree_lock(next);
6006         btrfs_set_lock_blocking(next);
6007
6008         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6009                                        &wc->refs[level - 1],
6010                                        &wc->flags[level - 1]);
6011         BUG_ON(ret);
6012         BUG_ON(wc->refs[level - 1] == 0);
6013         *lookup_info = 0;
6014
6015         if (wc->stage == DROP_REFERENCE) {
6016                 if (wc->refs[level - 1] > 1) {
6017                         if (level == 1 &&
6018                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6019                                 goto skip;
6020
6021                         if (!wc->update_ref ||
6022                             generation <= root->root_key.offset)
6023                                 goto skip;
6024
6025                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6026                                               path->slots[level]);
6027                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6028                         if (ret < 0)
6029                                 goto skip;
6030
6031                         wc->stage = UPDATE_BACKREF;
6032                         wc->shared_level = level - 1;
6033                 }
6034         } else {
6035                 if (level == 1 &&
6036                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6037                         goto skip;
6038         }
6039
6040         if (!btrfs_buffer_uptodate(next, generation)) {
6041                 btrfs_tree_unlock(next);
6042                 free_extent_buffer(next);
6043                 next = NULL;
6044                 *lookup_info = 1;
6045         }
6046
6047         if (!next) {
6048                 if (reada && level == 1)
6049                         reada_walk_down(trans, root, wc, path);
6050                 next = read_tree_block(root, bytenr, blocksize, generation);
6051                 btrfs_tree_lock(next);
6052                 btrfs_set_lock_blocking(next);
6053         }
6054
6055         level--;
6056         BUG_ON(level != btrfs_header_level(next));
6057         path->nodes[level] = next;
6058         path->slots[level] = 0;
6059         path->locks[level] = 1;
6060         wc->level = level;
6061         if (wc->level == 1)
6062                 wc->reada_slot = 0;
6063         return 0;
6064 skip:
6065         wc->refs[level - 1] = 0;
6066         wc->flags[level - 1] = 0;
6067         if (wc->stage == DROP_REFERENCE) {
6068                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6069                         parent = path->nodes[level]->start;
6070                 } else {
6071                         BUG_ON(root->root_key.objectid !=
6072                                btrfs_header_owner(path->nodes[level]));
6073                         parent = 0;
6074                 }
6075
6076                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6077                                         root->root_key.objectid, level - 1, 0);
6078                 BUG_ON(ret);
6079         }
6080         btrfs_tree_unlock(next);
6081         free_extent_buffer(next);
6082         *lookup_info = 1;
6083         return 1;
6084 }
6085
6086 /*
6087  * hepler to process tree block while walking up the tree.
6088  *
6089  * when wc->stage == DROP_REFERENCE, this function drops
6090  * reference count on the block.
6091  *
6092  * when wc->stage == UPDATE_BACKREF, this function changes
6093  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6094  * to UPDATE_BACKREF previously while processing the block.
6095  *
6096  * NOTE: return value 1 means we should stop walking up.
6097  */
6098 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6099                                  struct btrfs_root *root,
6100                                  struct btrfs_path *path,
6101                                  struct walk_control *wc)
6102 {
6103         int ret;
6104         int level = wc->level;
6105         struct extent_buffer *eb = path->nodes[level];
6106         u64 parent = 0;
6107
6108         if (wc->stage == UPDATE_BACKREF) {
6109                 BUG_ON(wc->shared_level < level);
6110                 if (level < wc->shared_level)
6111                         goto out;
6112
6113                 ret = find_next_key(path, level + 1, &wc->update_progress);
6114                 if (ret > 0)
6115                         wc->update_ref = 0;
6116
6117                 wc->stage = DROP_REFERENCE;
6118                 wc->shared_level = -1;
6119                 path->slots[level] = 0;
6120
6121                 /*
6122                  * check reference count again if the block isn't locked.
6123                  * we should start walking down the tree again if reference
6124                  * count is one.
6125                  */
6126                 if (!path->locks[level]) {
6127                         BUG_ON(level == 0);
6128                         btrfs_tree_lock(eb);
6129                         btrfs_set_lock_blocking(eb);
6130                         path->locks[level] = 1;
6131
6132                         ret = btrfs_lookup_extent_info(trans, root,
6133                                                        eb->start, eb->len,
6134                                                        &wc->refs[level],
6135                                                        &wc->flags[level]);
6136                         BUG_ON(ret);
6137                         BUG_ON(wc->refs[level] == 0);
6138                         if (wc->refs[level] == 1) {
6139                                 btrfs_tree_unlock(eb);
6140                                 path->locks[level] = 0;
6141                                 return 1;
6142                         }
6143                 }
6144         }
6145
6146         /* wc->stage == DROP_REFERENCE */
6147         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6148
6149         if (wc->refs[level] == 1) {
6150                 if (level == 0) {
6151                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6152                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6153                         else
6154                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6155                         BUG_ON(ret);
6156                 }
6157                 /* make block locked assertion in clean_tree_block happy */
6158                 if (!path->locks[level] &&
6159                     btrfs_header_generation(eb) == trans->transid) {
6160                         btrfs_tree_lock(eb);
6161                         btrfs_set_lock_blocking(eb);
6162                         path->locks[level] = 1;
6163                 }
6164                 clean_tree_block(trans, root, eb);
6165         }
6166
6167         if (eb == root->node) {
6168                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6169                         parent = eb->start;
6170                 else
6171                         BUG_ON(root->root_key.objectid !=
6172                                btrfs_header_owner(eb));
6173         } else {
6174                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6175                         parent = path->nodes[level + 1]->start;
6176                 else
6177                         BUG_ON(root->root_key.objectid !=
6178                                btrfs_header_owner(path->nodes[level + 1]));
6179         }
6180
6181         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6182 out:
6183         wc->refs[level] = 0;
6184         wc->flags[level] = 0;
6185         return 0;
6186 }
6187
6188 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6189                                    struct btrfs_root *root,
6190                                    struct btrfs_path *path,
6191                                    struct walk_control *wc)
6192 {
6193         int level = wc->level;
6194         int lookup_info = 1;
6195         int ret;
6196
6197         while (level >= 0) {
6198                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6199                 if (ret > 0)
6200                         break;
6201
6202                 if (level == 0)
6203                         break;
6204
6205                 if (path->slots[level] >=
6206                     btrfs_header_nritems(path->nodes[level]))
6207                         break;
6208
6209                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6210                 if (ret > 0) {
6211                         path->slots[level]++;
6212                         continue;
6213                 } else if (ret < 0)
6214                         return ret;
6215                 level = wc->level;
6216         }
6217         return 0;
6218 }
6219
6220 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6221                                  struct btrfs_root *root,
6222                                  struct btrfs_path *path,
6223                                  struct walk_control *wc, int max_level)
6224 {
6225         int level = wc->level;
6226         int ret;
6227
6228         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6229         while (level < max_level && path->nodes[level]) {
6230                 wc->level = level;
6231                 if (path->slots[level] + 1 <
6232                     btrfs_header_nritems(path->nodes[level])) {
6233                         path->slots[level]++;
6234                         return 0;
6235                 } else {
6236                         ret = walk_up_proc(trans, root, path, wc);
6237                         if (ret > 0)
6238                                 return 0;
6239
6240                         if (path->locks[level]) {
6241                                 btrfs_tree_unlock(path->nodes[level]);
6242                                 path->locks[level] = 0;
6243                         }
6244                         free_extent_buffer(path->nodes[level]);
6245                         path->nodes[level] = NULL;
6246                         level++;
6247                 }
6248         }
6249         return 1;
6250 }
6251
6252 /*
6253  * drop a subvolume tree.
6254  *
6255  * this function traverses the tree freeing any blocks that only
6256  * referenced by the tree.
6257  *
6258  * when a shared tree block is found. this function decreases its
6259  * reference count by one. if update_ref is true, this function
6260  * also make sure backrefs for the shared block and all lower level
6261  * blocks are properly updated.
6262  */
6263 int btrfs_drop_snapshot(struct btrfs_root *root,
6264                         struct btrfs_block_rsv *block_rsv, int update_ref)
6265 {
6266         struct btrfs_path *path;
6267         struct btrfs_trans_handle *trans;
6268         struct btrfs_root *tree_root = root->fs_info->tree_root;
6269         struct btrfs_root_item *root_item = &root->root_item;
6270         struct walk_control *wc;
6271         struct btrfs_key key;
6272         int err = 0;
6273         int ret;
6274         int level;
6275
6276         path = btrfs_alloc_path();
6277         BUG_ON(!path);
6278
6279         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6280         BUG_ON(!wc);
6281
6282         trans = btrfs_start_transaction(tree_root, 0);
6283         BUG_ON(IS_ERR(trans));
6284
6285         if (block_rsv)
6286                 trans->block_rsv = block_rsv;
6287
6288         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6289                 level = btrfs_header_level(root->node);
6290                 path->nodes[level] = btrfs_lock_root_node(root);
6291                 btrfs_set_lock_blocking(path->nodes[level]);
6292                 path->slots[level] = 0;
6293                 path->locks[level] = 1;
6294                 memset(&wc->update_progress, 0,
6295                        sizeof(wc->update_progress));
6296         } else {
6297                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6298                 memcpy(&wc->update_progress, &key,
6299                        sizeof(wc->update_progress));
6300
6301                 level = root_item->drop_level;
6302                 BUG_ON(level == 0);
6303                 path->lowest_level = level;
6304                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6305                 path->lowest_level = 0;
6306                 if (ret < 0) {
6307                         err = ret;
6308                         goto out;
6309                 }
6310                 WARN_ON(ret > 0);
6311
6312                 /*
6313                  * unlock our path, this is safe because only this
6314                  * function is allowed to delete this snapshot
6315                  */
6316                 btrfs_unlock_up_safe(path, 0);
6317
6318                 level = btrfs_header_level(root->node);
6319                 while (1) {
6320                         btrfs_tree_lock(path->nodes[level]);
6321                         btrfs_set_lock_blocking(path->nodes[level]);
6322
6323                         ret = btrfs_lookup_extent_info(trans, root,
6324                                                 path->nodes[level]->start,
6325                                                 path->nodes[level]->len,
6326                                                 &wc->refs[level],
6327                                                 &wc->flags[level]);
6328                         BUG_ON(ret);
6329                         BUG_ON(wc->refs[level] == 0);
6330
6331                         if (level == root_item->drop_level)
6332                                 break;
6333
6334                         btrfs_tree_unlock(path->nodes[level]);
6335                         WARN_ON(wc->refs[level] != 1);
6336                         level--;
6337                 }
6338         }
6339
6340         wc->level = level;
6341         wc->shared_level = -1;
6342         wc->stage = DROP_REFERENCE;
6343         wc->update_ref = update_ref;
6344         wc->keep_locks = 0;
6345         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6346
6347         while (1) {
6348                 ret = walk_down_tree(trans, root, path, wc);
6349                 if (ret < 0) {
6350                         err = ret;
6351                         break;
6352                 }
6353
6354                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6355                 if (ret < 0) {
6356                         err = ret;
6357                         break;
6358                 }
6359
6360                 if (ret > 0) {
6361                         BUG_ON(wc->stage != DROP_REFERENCE);
6362                         break;
6363                 }
6364
6365                 if (wc->stage == DROP_REFERENCE) {
6366                         level = wc->level;
6367                         btrfs_node_key(path->nodes[level],
6368                                        &root_item->drop_progress,
6369                                        path->slots[level]);
6370                         root_item->drop_level = level;
6371                 }
6372
6373                 BUG_ON(wc->level == 0);
6374                 if (btrfs_should_end_transaction(trans, tree_root)) {
6375                         ret = btrfs_update_root(trans, tree_root,
6376                                                 &root->root_key,
6377                                                 root_item);
6378                         BUG_ON(ret);
6379
6380                         btrfs_end_transaction_throttle(trans, tree_root);
6381                         trans = btrfs_start_transaction(tree_root, 0);
6382                         BUG_ON(IS_ERR(trans));
6383                         if (block_rsv)
6384                                 trans->block_rsv = block_rsv;
6385                 }
6386         }
6387         btrfs_release_path(root, path);
6388         BUG_ON(err);
6389
6390         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6391         BUG_ON(ret);
6392
6393         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6394                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6395                                            NULL, NULL);
6396                 BUG_ON(ret < 0);
6397                 if (ret > 0) {
6398                         /* if we fail to delete the orphan item this time
6399                          * around, it'll get picked up the next time.
6400                          *
6401                          * The most common failure here is just -ENOENT.
6402                          */
6403                         btrfs_del_orphan_item(trans, tree_root,
6404                                               root->root_key.objectid);
6405                 }
6406         }
6407
6408         if (root->in_radix) {
6409                 btrfs_free_fs_root(tree_root->fs_info, root);
6410         } else {
6411                 free_extent_buffer(root->node);
6412                 free_extent_buffer(root->commit_root);
6413                 kfree(root);
6414         }
6415 out:
6416         btrfs_end_transaction_throttle(trans, tree_root);
6417         kfree(wc);
6418         btrfs_free_path(path);
6419         return err;
6420 }
6421
6422 /*
6423  * drop subtree rooted at tree block 'node'.
6424  *
6425  * NOTE: this function will unlock and release tree block 'node'
6426  */
6427 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6428                         struct btrfs_root *root,
6429                         struct extent_buffer *node,
6430                         struct extent_buffer *parent)
6431 {
6432         struct btrfs_path *path;
6433         struct walk_control *wc;
6434         int level;
6435         int parent_level;
6436         int ret = 0;
6437         int wret;
6438
6439         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6440
6441         path = btrfs_alloc_path();
6442         BUG_ON(!path);
6443
6444         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6445         BUG_ON(!wc);
6446
6447         btrfs_assert_tree_locked(parent);
6448         parent_level = btrfs_header_level(parent);
6449         extent_buffer_get(parent);
6450         path->nodes[parent_level] = parent;
6451         path->slots[parent_level] = btrfs_header_nritems(parent);
6452
6453         btrfs_assert_tree_locked(node);
6454         level = btrfs_header_level(node);
6455         path->nodes[level] = node;
6456         path->slots[level] = 0;
6457         path->locks[level] = 1;
6458
6459         wc->refs[parent_level] = 1;
6460         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6461         wc->level = level;
6462         wc->shared_level = -1;
6463         wc->stage = DROP_REFERENCE;
6464         wc->update_ref = 0;
6465         wc->keep_locks = 1;
6466         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6467
6468         while (1) {
6469                 wret = walk_down_tree(trans, root, path, wc);
6470                 if (wret < 0) {
6471                         ret = wret;
6472                         break;
6473                 }
6474
6475                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6476                 if (wret < 0)
6477                         ret = wret;
6478                 if (wret != 0)
6479                         break;
6480         }
6481
6482         kfree(wc);
6483         btrfs_free_path(path);
6484         return ret;
6485 }
6486
6487 #if 0
6488 static unsigned long calc_ra(unsigned long start, unsigned long last,
6489                              unsigned long nr)
6490 {
6491         return min(last, start + nr - 1);
6492 }
6493
6494 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6495                                          u64 len)
6496 {
6497         u64 page_start;
6498         u64 page_end;
6499         unsigned long first_index;
6500         unsigned long last_index;
6501         unsigned long i;
6502         struct page *page;
6503         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6504         struct file_ra_state *ra;
6505         struct btrfs_ordered_extent *ordered;
6506         unsigned int total_read = 0;
6507         unsigned int total_dirty = 0;
6508         int ret = 0;
6509
6510         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6511         if (!ra)
6512                 return -ENOMEM;
6513
6514         mutex_lock(&inode->i_mutex);
6515         first_index = start >> PAGE_CACHE_SHIFT;
6516         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6517
6518         /* make sure the dirty trick played by the caller work */
6519         ret = invalidate_inode_pages2_range(inode->i_mapping,
6520                                             first_index, last_index);
6521         if (ret)
6522                 goto out_unlock;
6523
6524         file_ra_state_init(ra, inode->i_mapping);
6525
6526         for (i = first_index ; i <= last_index; i++) {
6527                 if (total_read % ra->ra_pages == 0) {
6528                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6529                                        calc_ra(i, last_index, ra->ra_pages));
6530                 }
6531                 total_read++;
6532 again:
6533                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6534                         BUG_ON(1);
6535                 page = grab_cache_page(inode->i_mapping, i);
6536                 if (!page) {
6537                         ret = -ENOMEM;
6538                         goto out_unlock;
6539                 }
6540                 if (!PageUptodate(page)) {
6541                         btrfs_readpage(NULL, page);
6542                         lock_page(page);
6543                         if (!PageUptodate(page)) {
6544                                 unlock_page(page);
6545                                 page_cache_release(page);
6546                                 ret = -EIO;
6547                                 goto out_unlock;
6548                         }
6549                 }
6550                 wait_on_page_writeback(page);
6551
6552                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6553                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6554                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6555
6556                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6557                 if (ordered) {
6558                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6559                         unlock_page(page);
6560                         page_cache_release(page);
6561                         btrfs_start_ordered_extent(inode, ordered, 1);
6562                         btrfs_put_ordered_extent(ordered);
6563                         goto again;
6564                 }
6565                 set_page_extent_mapped(page);
6566
6567                 if (i == first_index)
6568                         set_extent_bits(io_tree, page_start, page_end,
6569                                         EXTENT_BOUNDARY, GFP_NOFS);
6570                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6571
6572                 set_page_dirty(page);
6573                 total_dirty++;
6574
6575                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6576                 unlock_page(page);
6577                 page_cache_release(page);
6578         }
6579
6580 out_unlock:
6581         kfree(ra);
6582         mutex_unlock(&inode->i_mutex);
6583         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6584         return ret;
6585 }
6586
6587 static noinline int relocate_data_extent(struct inode *reloc_inode,
6588                                          struct btrfs_key *extent_key,
6589                                          u64 offset)
6590 {
6591         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6592         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6593         struct extent_map *em;
6594         u64 start = extent_key->objectid - offset;
6595         u64 end = start + extent_key->offset - 1;
6596
6597         em = alloc_extent_map(GFP_NOFS);
6598         BUG_ON(!em);
6599
6600         em->start = start;
6601         em->len = extent_key->offset;
6602         em->block_len = extent_key->offset;
6603         em->block_start = extent_key->objectid;
6604         em->bdev = root->fs_info->fs_devices->latest_bdev;
6605         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6606
6607         /* setup extent map to cheat btrfs_readpage */
6608         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6609         while (1) {
6610                 int ret;
6611                 write_lock(&em_tree->lock);
6612                 ret = add_extent_mapping(em_tree, em);
6613                 write_unlock(&em_tree->lock);
6614                 if (ret != -EEXIST) {
6615                         free_extent_map(em);
6616                         break;
6617                 }
6618                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6619         }
6620         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6621
6622         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6623 }
6624
6625 struct btrfs_ref_path {
6626         u64 extent_start;
6627         u64 nodes[BTRFS_MAX_LEVEL];
6628         u64 root_objectid;
6629         u64 root_generation;
6630         u64 owner_objectid;
6631         u32 num_refs;
6632         int lowest_level;
6633         int current_level;
6634         int shared_level;
6635
6636         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6637         u64 new_nodes[BTRFS_MAX_LEVEL];
6638 };
6639
6640 struct disk_extent {
6641         u64 ram_bytes;
6642         u64 disk_bytenr;
6643         u64 disk_num_bytes;
6644         u64 offset;
6645         u64 num_bytes;
6646         u8 compression;
6647         u8 encryption;
6648         u16 other_encoding;
6649 };
6650
6651 static int is_cowonly_root(u64 root_objectid)
6652 {
6653         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6654             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6655             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6656             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6657             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6658             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6659                 return 1;
6660         return 0;
6661 }
6662
6663 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6664                                     struct btrfs_root *extent_root,
6665                                     struct btrfs_ref_path *ref_path,
6666                                     int first_time)
6667 {
6668         struct extent_buffer *leaf;
6669         struct btrfs_path *path;
6670         struct btrfs_extent_ref *ref;
6671         struct btrfs_key key;
6672         struct btrfs_key found_key;
6673         u64 bytenr;
6674         u32 nritems;
6675         int level;
6676         int ret = 1;
6677
6678         path = btrfs_alloc_path();
6679         if (!path)
6680                 return -ENOMEM;
6681
6682         if (first_time) {
6683                 ref_path->lowest_level = -1;
6684                 ref_path->current_level = -1;
6685                 ref_path->shared_level = -1;
6686                 goto walk_up;
6687         }
6688 walk_down:
6689         level = ref_path->current_level - 1;
6690         while (level >= -1) {
6691                 u64 parent;
6692                 if (level < ref_path->lowest_level)
6693                         break;
6694
6695                 if (level >= 0)
6696                         bytenr = ref_path->nodes[level];
6697                 else
6698                         bytenr = ref_path->extent_start;
6699                 BUG_ON(bytenr == 0);
6700
6701                 parent = ref_path->nodes[level + 1];
6702                 ref_path->nodes[level + 1] = 0;
6703                 ref_path->current_level = level;
6704                 BUG_ON(parent == 0);
6705
6706                 key.objectid = bytenr;
6707                 key.offset = parent + 1;
6708                 key.type = BTRFS_EXTENT_REF_KEY;
6709
6710                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6711                 if (ret < 0)
6712                         goto out;
6713                 BUG_ON(ret == 0);
6714
6715                 leaf = path->nodes[0];
6716                 nritems = btrfs_header_nritems(leaf);
6717                 if (path->slots[0] >= nritems) {
6718                         ret = btrfs_next_leaf(extent_root, path);
6719                         if (ret < 0)
6720                                 goto out;
6721                         if (ret > 0)
6722                                 goto next;
6723                         leaf = path->nodes[0];
6724                 }
6725
6726                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6727                 if (found_key.objectid == bytenr &&
6728                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6729                         if (level < ref_path->shared_level)
6730                                 ref_path->shared_level = level;
6731                         goto found;
6732                 }
6733 next:
6734                 level--;
6735                 btrfs_release_path(extent_root, path);
6736                 cond_resched();
6737         }
6738         /* reached lowest level */
6739         ret = 1;
6740         goto out;
6741 walk_up:
6742         level = ref_path->current_level;
6743         while (level < BTRFS_MAX_LEVEL - 1) {
6744                 u64 ref_objectid;
6745
6746                 if (level >= 0)
6747                         bytenr = ref_path->nodes[level];
6748                 else
6749                         bytenr = ref_path->extent_start;
6750
6751                 BUG_ON(bytenr == 0);
6752
6753                 key.objectid = bytenr;
6754                 key.offset = 0;
6755                 key.type = BTRFS_EXTENT_REF_KEY;
6756
6757                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6758                 if (ret < 0)
6759                         goto out;
6760
6761                 leaf = path->nodes[0];
6762                 nritems = btrfs_header_nritems(leaf);
6763                 if (path->slots[0] >= nritems) {
6764                         ret = btrfs_next_leaf(extent_root, path);
6765                         if (ret < 0)
6766                                 goto out;
6767                         if (ret > 0) {
6768                                 /* the extent was freed by someone */
6769                                 if (ref_path->lowest_level == level)
6770                                         goto out;
6771                                 btrfs_release_path(extent_root, path);
6772                                 goto walk_down;
6773                         }
6774                         leaf = path->nodes[0];
6775                 }
6776
6777                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6778                 if (found_key.objectid != bytenr ||
6779                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6780                         /* the extent was freed by someone */
6781                         if (ref_path->lowest_level == level) {
6782                                 ret = 1;
6783                                 goto out;
6784                         }
6785                         btrfs_release_path(extent_root, path);
6786                         goto walk_down;
6787                 }
6788 found:
6789                 ref = btrfs_item_ptr(leaf, path->slots[0],
6790                                 struct btrfs_extent_ref);
6791                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6792                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6793                         if (first_time) {
6794                                 level = (int)ref_objectid;
6795                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6796                                 ref_path->lowest_level = level;
6797                                 ref_path->current_level = level;
6798                                 ref_path->nodes[level] = bytenr;
6799                         } else {
6800                                 WARN_ON(ref_objectid != level);
6801                         }
6802                 } else {
6803                         WARN_ON(level != -1);
6804                 }
6805                 first_time = 0;
6806
6807                 if (ref_path->lowest_level == level) {
6808                         ref_path->owner_objectid = ref_objectid;
6809                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6810                 }
6811
6812                 /*
6813                  * the block is tree root or the block isn't in reference
6814                  * counted tree.
6815                  */
6816                 if (found_key.objectid == found_key.offset ||
6817                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6818                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6819                         ref_path->root_generation =
6820                                 btrfs_ref_generation(leaf, ref);
6821                         if (level < 0) {
6822                                 /* special reference from the tree log */
6823                                 ref_path->nodes[0] = found_key.offset;
6824                                 ref_path->current_level = 0;
6825                         }
6826                         ret = 0;
6827                         goto out;
6828                 }
6829
6830                 level++;
6831                 BUG_ON(ref_path->nodes[level] != 0);
6832                 ref_path->nodes[level] = found_key.offset;
6833                 ref_path->current_level = level;
6834
6835                 /*
6836                  * the reference was created in the running transaction,
6837                  * no need to continue walking up.
6838                  */
6839                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6840                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6841                         ref_path->root_generation =
6842                                 btrfs_ref_generation(leaf, ref);
6843                         ret = 0;
6844                         goto out;
6845                 }
6846
6847                 btrfs_release_path(extent_root, path);
6848                 cond_resched();
6849         }
6850         /* reached max tree level, but no tree root found. */
6851         BUG();
6852 out:
6853         btrfs_free_path(path);
6854         return ret;
6855 }
6856
6857 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6858                                 struct btrfs_root *extent_root,
6859                                 struct btrfs_ref_path *ref_path,
6860                                 u64 extent_start)
6861 {
6862         memset(ref_path, 0, sizeof(*ref_path));
6863         ref_path->extent_start = extent_start;
6864
6865         return __next_ref_path(trans, extent_root, ref_path, 1);
6866 }
6867
6868 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6869                                struct btrfs_root *extent_root,
6870                                struct btrfs_ref_path *ref_path)
6871 {
6872         return __next_ref_path(trans, extent_root, ref_path, 0);
6873 }
6874
6875 static noinline int get_new_locations(struct inode *reloc_inode,
6876                                       struct btrfs_key *extent_key,
6877                                       u64 offset, int no_fragment,
6878                                       struct disk_extent **extents,
6879                                       int *nr_extents)
6880 {
6881         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6882         struct btrfs_path *path;
6883         struct btrfs_file_extent_item *fi;
6884         struct extent_buffer *leaf;
6885         struct disk_extent *exts = *extents;
6886         struct btrfs_key found_key;
6887         u64 cur_pos;
6888         u64 last_byte;
6889         u32 nritems;
6890         int nr = 0;
6891         int max = *nr_extents;
6892         int ret;
6893
6894         WARN_ON(!no_fragment && *extents);
6895         if (!exts) {
6896                 max = 1;
6897                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6898                 if (!exts)
6899                         return -ENOMEM;
6900         }
6901
6902         path = btrfs_alloc_path();
6903         BUG_ON(!path);
6904
6905         cur_pos = extent_key->objectid - offset;
6906         last_byte = extent_key->objectid + extent_key->offset;
6907         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6908                                        cur_pos, 0);
6909         if (ret < 0)
6910                 goto out;
6911         if (ret > 0) {
6912                 ret = -ENOENT;
6913                 goto out;
6914         }
6915
6916         while (1) {
6917                 leaf = path->nodes[0];
6918                 nritems = btrfs_header_nritems(leaf);
6919                 if (path->slots[0] >= nritems) {
6920                         ret = btrfs_next_leaf(root, path);
6921                         if (ret < 0)
6922                                 goto out;
6923                         if (ret > 0)
6924                                 break;
6925                         leaf = path->nodes[0];
6926                 }
6927
6928                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6929                 if (found_key.offset != cur_pos ||
6930                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6931                     found_key.objectid != reloc_inode->i_ino)
6932                         break;
6933
6934                 fi = btrfs_item_ptr(leaf, path->slots[0],
6935                                     struct btrfs_file_extent_item);
6936                 if (btrfs_file_extent_type(leaf, fi) !=
6937                     BTRFS_FILE_EXTENT_REG ||
6938                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6939                         break;
6940
6941                 if (nr == max) {
6942                         struct disk_extent *old = exts;
6943                         max *= 2;
6944                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6945                         memcpy(exts, old, sizeof(*exts) * nr);
6946                         if (old != *extents)
6947                                 kfree(old);
6948                 }
6949
6950                 exts[nr].disk_bytenr =
6951                         btrfs_file_extent_disk_bytenr(leaf, fi);
6952                 exts[nr].disk_num_bytes =
6953                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6954                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6955                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6956                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6957                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6958                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6959                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6960                                                                            fi);
6961                 BUG_ON(exts[nr].offset > 0);
6962                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6963                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6964
6965                 cur_pos += exts[nr].num_bytes;
6966                 nr++;
6967
6968                 if (cur_pos + offset >= last_byte)
6969                         break;
6970
6971                 if (no_fragment) {
6972                         ret = 1;
6973                         goto out;
6974                 }
6975                 path->slots[0]++;
6976         }
6977
6978         BUG_ON(cur_pos + offset > last_byte);
6979         if (cur_pos + offset < last_byte) {
6980                 ret = -ENOENT;
6981                 goto out;
6982         }
6983         ret = 0;
6984 out:
6985         btrfs_free_path(path);
6986         if (ret) {
6987                 if (exts != *extents)
6988                         kfree(exts);
6989         } else {
6990                 *extents = exts;
6991                 *nr_extents = nr;
6992         }
6993         return ret;
6994 }
6995
6996 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6997                                         struct btrfs_root *root,
6998                                         struct btrfs_path *path,
6999                                         struct btrfs_key *extent_key,
7000                                         struct btrfs_key *leaf_key,
7001                                         struct btrfs_ref_path *ref_path,
7002                                         struct disk_extent *new_extents,
7003                                         int nr_extents)
7004 {
7005         struct extent_buffer *leaf;
7006         struct btrfs_file_extent_item *fi;
7007         struct inode *inode = NULL;
7008         struct btrfs_key key;
7009         u64 lock_start = 0;
7010         u64 lock_end = 0;
7011         u64 num_bytes;
7012         u64 ext_offset;
7013         u64 search_end = (u64)-1;
7014         u32 nritems;
7015         int nr_scaned = 0;
7016         int extent_locked = 0;
7017         int extent_type;
7018         int ret;
7019
7020         memcpy(&key, leaf_key, sizeof(key));
7021         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7022                 if (key.objectid < ref_path->owner_objectid ||
7023                     (key.objectid == ref_path->owner_objectid &&
7024                      key.type < BTRFS_EXTENT_DATA_KEY)) {
7025                         key.objectid = ref_path->owner_objectid;
7026                         key.type = BTRFS_EXTENT_DATA_KEY;
7027                         key.offset = 0;
7028                 }
7029         }
7030
7031         while (1) {
7032                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7033                 if (ret < 0)
7034                         goto out;
7035
7036                 leaf = path->nodes[0];
7037                 nritems = btrfs_header_nritems(leaf);
7038 next:
7039                 if (extent_locked && ret > 0) {
7040                         /*
7041                          * the file extent item was modified by someone
7042                          * before the extent got locked.
7043                          */
7044                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7045                                       lock_end, GFP_NOFS);
7046                         extent_locked = 0;
7047                 }
7048
7049                 if (path->slots[0] >= nritems) {
7050                         if (++nr_scaned > 2)
7051                                 break;
7052
7053                         BUG_ON(extent_locked);
7054                         ret = btrfs_next_leaf(root, path);
7055                         if (ret < 0)
7056                                 goto out;
7057                         if (ret > 0)
7058                                 break;
7059                         leaf = path->nodes[0];
7060                         nritems = btrfs_header_nritems(leaf);
7061                 }
7062
7063                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7064
7065                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7066                         if ((key.objectid > ref_path->owner_objectid) ||
7067                             (key.objectid == ref_path->owner_objectid &&
7068                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7069                             key.offset >= search_end)
7070                                 break;
7071                 }
7072
7073                 if (inode && key.objectid != inode->i_ino) {
7074                         BUG_ON(extent_locked);
7075                         btrfs_release_path(root, path);
7076                         mutex_unlock(&inode->i_mutex);
7077                         iput(inode);
7078                         inode = NULL;
7079                         continue;
7080                 }
7081
7082                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7083                         path->slots[0]++;
7084                         ret = 1;
7085                         goto next;
7086                 }
7087                 fi = btrfs_item_ptr(leaf, path->slots[0],
7088                                     struct btrfs_file_extent_item);
7089                 extent_type = btrfs_file_extent_type(leaf, fi);
7090                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7091                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7092                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7093                      extent_key->objectid)) {
7094                         path->slots[0]++;
7095                         ret = 1;
7096                         goto next;
7097                 }
7098
7099                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7100                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7101
7102                 if (search_end == (u64)-1) {
7103                         search_end = key.offset - ext_offset +
7104                                 btrfs_file_extent_ram_bytes(leaf, fi);
7105                 }
7106
7107                 if (!extent_locked) {
7108                         lock_start = key.offset;
7109                         lock_end = lock_start + num_bytes - 1;
7110                 } else {
7111                         if (lock_start > key.offset ||
7112                             lock_end + 1 < key.offset + num_bytes) {
7113                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7114                                               lock_start, lock_end, GFP_NOFS);
7115                                 extent_locked = 0;
7116                         }
7117                 }
7118
7119                 if (!inode) {
7120                         btrfs_release_path(root, path);
7121
7122                         inode = btrfs_iget_locked(root->fs_info->sb,
7123                                                   key.objectid, root);
7124                         if (inode->i_state & I_NEW) {
7125                                 BTRFS_I(inode)->root = root;
7126                                 BTRFS_I(inode)->location.objectid =
7127                                         key.objectid;
7128                                 BTRFS_I(inode)->location.type =
7129                                         BTRFS_INODE_ITEM_KEY;
7130                                 BTRFS_I(inode)->location.offset = 0;
7131                                 btrfs_read_locked_inode(inode);
7132                                 unlock_new_inode(inode);
7133                         }
7134                         /*
7135                          * some code call btrfs_commit_transaction while
7136                          * holding the i_mutex, so we can't use mutex_lock
7137                          * here.
7138                          */
7139                         if (is_bad_inode(inode) ||
7140                             !mutex_trylock(&inode->i_mutex)) {
7141                                 iput(inode);
7142                                 inode = NULL;
7143                                 key.offset = (u64)-1;
7144                                 goto skip;
7145                         }
7146                 }
7147
7148                 if (!extent_locked) {
7149                         struct btrfs_ordered_extent *ordered;
7150
7151                         btrfs_release_path(root, path);
7152
7153                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7154                                     lock_end, GFP_NOFS);
7155                         ordered = btrfs_lookup_first_ordered_extent(inode,
7156                                                                     lock_end);
7157                         if (ordered &&
7158                             ordered->file_offset <= lock_end &&
7159                             ordered->file_offset + ordered->len > lock_start) {
7160                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7161                                               lock_start, lock_end, GFP_NOFS);
7162                                 btrfs_start_ordered_extent(inode, ordered, 1);
7163                                 btrfs_put_ordered_extent(ordered);
7164                                 key.offset += num_bytes;
7165                                 goto skip;
7166                         }
7167                         if (ordered)
7168                                 btrfs_put_ordered_extent(ordered);
7169
7170                         extent_locked = 1;
7171                         continue;
7172                 }
7173
7174                 if (nr_extents == 1) {
7175                         /* update extent pointer in place */
7176                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7177                                                 new_extents[0].disk_bytenr);
7178                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7179                                                 new_extents[0].disk_num_bytes);
7180                         btrfs_mark_buffer_dirty(leaf);
7181
7182                         btrfs_drop_extent_cache(inode, key.offset,
7183                                                 key.offset + num_bytes - 1, 0);
7184
7185                         ret = btrfs_inc_extent_ref(trans, root,
7186                                                 new_extents[0].disk_bytenr,
7187                                                 new_extents[0].disk_num_bytes,
7188                                                 leaf->start,
7189                                                 root->root_key.objectid,
7190                                                 trans->transid,
7191                                                 key.objectid);
7192                         BUG_ON(ret);
7193
7194                         ret = btrfs_free_extent(trans, root,
7195                                                 extent_key->objectid,
7196                                                 extent_key->offset,
7197                                                 leaf->start,
7198                                                 btrfs_header_owner(leaf),
7199                                                 btrfs_header_generation(leaf),
7200                                                 key.objectid, 0);
7201                         BUG_ON(ret);
7202
7203                         btrfs_release_path(root, path);
7204                         key.offset += num_bytes;
7205                 } else {
7206                         BUG_ON(1);
7207 #if 0
7208                         u64 alloc_hint;
7209                         u64 extent_len;
7210                         int i;
7211                         /*
7212                          * drop old extent pointer at first, then insert the
7213                          * new pointers one bye one
7214                          */
7215                         btrfs_release_path(root, path);
7216                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7217                                                  key.offset + num_bytes,
7218                                                  key.offset, &alloc_hint);
7219                         BUG_ON(ret);
7220
7221                         for (i = 0; i < nr_extents; i++) {
7222                                 if (ext_offset >= new_extents[i].num_bytes) {
7223                                         ext_offset -= new_extents[i].num_bytes;
7224                                         continue;
7225                                 }
7226                                 extent_len = min(new_extents[i].num_bytes -
7227                                                  ext_offset, num_bytes);
7228
7229                                 ret = btrfs_insert_empty_item(trans, root,
7230                                                               path, &key,
7231                                                               sizeof(*fi));
7232                                 BUG_ON(ret);
7233
7234                                 leaf = path->nodes[0];
7235                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7236                                                 struct btrfs_file_extent_item);
7237                                 btrfs_set_file_extent_generation(leaf, fi,
7238                                                         trans->transid);
7239                                 btrfs_set_file_extent_type(leaf, fi,
7240                                                         BTRFS_FILE_EXTENT_REG);
7241                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7242                                                 new_extents[i].disk_bytenr);
7243                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7244                                                 new_extents[i].disk_num_bytes);
7245                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7246                                                 new_extents[i].ram_bytes);
7247
7248                                 btrfs_set_file_extent_compression(leaf, fi,
7249                                                 new_extents[i].compression);
7250                                 btrfs_set_file_extent_encryption(leaf, fi,
7251                                                 new_extents[i].encryption);
7252                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7253                                                 new_extents[i].other_encoding);
7254
7255                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7256                                                         extent_len);
7257                                 ext_offset += new_extents[i].offset;
7258                                 btrfs_set_file_extent_offset(leaf, fi,
7259                                                         ext_offset);
7260                                 btrfs_mark_buffer_dirty(leaf);
7261
7262                                 btrfs_drop_extent_cache(inode, key.offset,
7263                                                 key.offset + extent_len - 1, 0);
7264
7265                                 ret = btrfs_inc_extent_ref(trans, root,
7266                                                 new_extents[i].disk_bytenr,
7267                                                 new_extents[i].disk_num_bytes,
7268                                                 leaf->start,
7269                                                 root->root_key.objectid,
7270                                                 trans->transid, key.objectid);
7271                                 BUG_ON(ret);
7272                                 btrfs_release_path(root, path);
7273
7274                                 inode_add_bytes(inode, extent_len);
7275
7276                                 ext_offset = 0;
7277                                 num_bytes -= extent_len;
7278                                 key.offset += extent_len;
7279
7280                                 if (num_bytes == 0)
7281                                         break;
7282                         }
7283                         BUG_ON(i >= nr_extents);
7284 #endif
7285                 }
7286
7287                 if (extent_locked) {
7288                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7289                                       lock_end, GFP_NOFS);
7290                         extent_locked = 0;
7291                 }
7292 skip:
7293                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7294                     key.offset >= search_end)
7295                         break;
7296
7297                 cond_resched();
7298         }
7299         ret = 0;
7300 out:
7301         btrfs_release_path(root, path);
7302         if (inode) {
7303                 mutex_unlock(&inode->i_mutex);
7304                 if (extent_locked) {
7305                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7306                                       lock_end, GFP_NOFS);
7307                 }
7308                 iput(inode);
7309         }
7310         return ret;
7311 }
7312
7313 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7314                                struct btrfs_root *root,
7315                                struct extent_buffer *buf, u64 orig_start)
7316 {
7317         int level;
7318         int ret;
7319
7320         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7321         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7322
7323         level = btrfs_header_level(buf);
7324         if (level == 0) {
7325                 struct btrfs_leaf_ref *ref;
7326                 struct btrfs_leaf_ref *orig_ref;
7327
7328                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7329                 if (!orig_ref)
7330                         return -ENOENT;
7331
7332                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7333                 if (!ref) {
7334                         btrfs_free_leaf_ref(root, orig_ref);
7335                         return -ENOMEM;
7336                 }
7337
7338                 ref->nritems = orig_ref->nritems;
7339                 memcpy(ref->extents, orig_ref->extents,
7340                         sizeof(ref->extents[0]) * ref->nritems);
7341
7342                 btrfs_free_leaf_ref(root, orig_ref);
7343
7344                 ref->root_gen = trans->transid;
7345                 ref->bytenr = buf->start;
7346                 ref->owner = btrfs_header_owner(buf);
7347                 ref->generation = btrfs_header_generation(buf);
7348
7349                 ret = btrfs_add_leaf_ref(root, ref, 0);
7350                 WARN_ON(ret);
7351                 btrfs_free_leaf_ref(root, ref);
7352         }
7353         return 0;
7354 }
7355
7356 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7357                                         struct extent_buffer *leaf,
7358                                         struct btrfs_block_group_cache *group,
7359                                         struct btrfs_root *target_root)
7360 {
7361         struct btrfs_key key;
7362         struct inode *inode = NULL;
7363         struct btrfs_file_extent_item *fi;
7364         struct extent_state *cached_state = NULL;
7365         u64 num_bytes;
7366         u64 skip_objectid = 0;
7367         u32 nritems;
7368         u32 i;
7369
7370         nritems = btrfs_header_nritems(leaf);
7371         for (i = 0; i < nritems; i++) {
7372                 btrfs_item_key_to_cpu(leaf, &key, i);
7373                 if (key.objectid == skip_objectid ||
7374                     key.type != BTRFS_EXTENT_DATA_KEY)
7375                         continue;
7376                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7377                 if (btrfs_file_extent_type(leaf, fi) ==
7378                     BTRFS_FILE_EXTENT_INLINE)
7379                         continue;
7380                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7381                         continue;
7382                 if (!inode || inode->i_ino != key.objectid) {
7383                         iput(inode);
7384                         inode = btrfs_ilookup(target_root->fs_info->sb,
7385                                               key.objectid, target_root, 1);
7386                 }
7387                 if (!inode) {
7388                         skip_objectid = key.objectid;
7389                         continue;
7390                 }
7391                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7392
7393                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7394                                  key.offset + num_bytes - 1, 0, &cached_state,
7395                                  GFP_NOFS);
7396                 btrfs_drop_extent_cache(inode, key.offset,
7397                                         key.offset + num_bytes - 1, 1);
7398                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7399                                      key.offset + num_bytes - 1, &cached_state,
7400                                      GFP_NOFS);
7401                 cond_resched();
7402         }
7403         iput(inode);
7404         return 0;
7405 }
7406
7407 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7408                                         struct btrfs_root *root,
7409                                         struct extent_buffer *leaf,
7410                                         struct btrfs_block_group_cache *group,
7411                                         struct inode *reloc_inode)
7412 {
7413         struct btrfs_key key;
7414         struct btrfs_key extent_key;
7415         struct btrfs_file_extent_item *fi;
7416         struct btrfs_leaf_ref *ref;
7417         struct disk_extent *new_extent;
7418         u64 bytenr;
7419         u64 num_bytes;
7420         u32 nritems;
7421         u32 i;
7422         int ext_index;
7423         int nr_extent;
7424         int ret;
7425
7426         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7427         BUG_ON(!new_extent);
7428
7429         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7430         BUG_ON(!ref);
7431
7432         ext_index = -1;
7433         nritems = btrfs_header_nritems(leaf);
7434         for (i = 0; i < nritems; i++) {
7435                 btrfs_item_key_to_cpu(leaf, &key, i);
7436                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7437                         continue;
7438                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7439                 if (btrfs_file_extent_type(leaf, fi) ==
7440                     BTRFS_FILE_EXTENT_INLINE)
7441                         continue;
7442                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7443                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7444                 if (bytenr == 0)
7445                         continue;
7446
7447                 ext_index++;
7448                 if (bytenr >= group->key.objectid + group->key.offset ||
7449                     bytenr + num_bytes <= group->key.objectid)
7450                         continue;
7451
7452                 extent_key.objectid = bytenr;
7453                 extent_key.offset = num_bytes;
7454                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7455                 nr_extent = 1;
7456                 ret = get_new_locations(reloc_inode, &extent_key,
7457                                         group->key.objectid, 1,
7458                                         &new_extent, &nr_extent);
7459                 if (ret > 0)
7460                         continue;
7461                 BUG_ON(ret < 0);
7462
7463                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7464                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7465                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7466                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7467
7468                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7469                                                 new_extent->disk_bytenr);
7470                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7471                                                 new_extent->disk_num_bytes);
7472                 btrfs_mark_buffer_dirty(leaf);
7473
7474                 ret = btrfs_inc_extent_ref(trans, root,
7475                                         new_extent->disk_bytenr,
7476                                         new_extent->disk_num_bytes,
7477                                         leaf->start,
7478                                         root->root_key.objectid,
7479                                         trans->transid, key.objectid);
7480                 BUG_ON(ret);
7481
7482                 ret = btrfs_free_extent(trans, root,
7483                                         bytenr, num_bytes, leaf->start,
7484                                         btrfs_header_owner(leaf),
7485                                         btrfs_header_generation(leaf),
7486                                         key.objectid, 0);
7487                 BUG_ON(ret);
7488                 cond_resched();
7489         }
7490         kfree(new_extent);
7491         BUG_ON(ext_index + 1 != ref->nritems);
7492         btrfs_free_leaf_ref(root, ref);
7493         return 0;
7494 }
7495
7496 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7497                           struct btrfs_root *root)
7498 {
7499         struct btrfs_root *reloc_root;
7500         int ret;
7501
7502         if (root->reloc_root) {
7503                 reloc_root = root->reloc_root;
7504                 root->reloc_root = NULL;
7505                 list_add(&reloc_root->dead_list,
7506                          &root->fs_info->dead_reloc_roots);
7507
7508                 btrfs_set_root_bytenr(&reloc_root->root_item,
7509                                       reloc_root->node->start);
7510                 btrfs_set_root_level(&root->root_item,
7511                                      btrfs_header_level(reloc_root->node));
7512                 memset(&reloc_root->root_item.drop_progress, 0,
7513                         sizeof(struct btrfs_disk_key));
7514                 reloc_root->root_item.drop_level = 0;
7515
7516                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7517                                         &reloc_root->root_key,
7518                                         &reloc_root->root_item);
7519                 BUG_ON(ret);
7520         }
7521         return 0;
7522 }
7523
7524 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7525 {
7526         struct btrfs_trans_handle *trans;
7527         struct btrfs_root *reloc_root;
7528         struct btrfs_root *prev_root = NULL;
7529         struct list_head dead_roots;
7530         int ret;
7531         unsigned long nr;
7532
7533         INIT_LIST_HEAD(&dead_roots);
7534         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7535
7536         while (!list_empty(&dead_roots)) {
7537                 reloc_root = list_entry(dead_roots.prev,
7538                                         struct btrfs_root, dead_list);
7539                 list_del_init(&reloc_root->dead_list);
7540
7541                 BUG_ON(reloc_root->commit_root != NULL);
7542                 while (1) {
7543                         trans = btrfs_join_transaction(root, 1);
7544                         BUG_ON(IS_ERR(trans));
7545
7546                         mutex_lock(&root->fs_info->drop_mutex);
7547                         ret = btrfs_drop_snapshot(trans, reloc_root);
7548                         if (ret != -EAGAIN)
7549                                 break;
7550                         mutex_unlock(&root->fs_info->drop_mutex);
7551
7552                         nr = trans->blocks_used;
7553                         ret = btrfs_end_transaction(trans, root);
7554                         BUG_ON(ret);
7555                         btrfs_btree_balance_dirty(root, nr);
7556                 }
7557
7558                 free_extent_buffer(reloc_root->node);
7559
7560                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7561                                      &reloc_root->root_key);
7562                 BUG_ON(ret);
7563                 mutex_unlock(&root->fs_info->drop_mutex);
7564
7565                 nr = trans->blocks_used;
7566                 ret = btrfs_end_transaction(trans, root);
7567                 BUG_ON(ret);
7568                 btrfs_btree_balance_dirty(root, nr);
7569
7570                 kfree(prev_root);
7571                 prev_root = reloc_root;
7572         }
7573         if (prev_root) {
7574                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7575                 kfree(prev_root);
7576         }
7577         return 0;
7578 }
7579
7580 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7581 {
7582         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7583         return 0;
7584 }
7585
7586 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7587 {
7588         struct btrfs_root *reloc_root;
7589         struct btrfs_trans_handle *trans;
7590         struct btrfs_key location;
7591         int found;
7592         int ret;
7593
7594         mutex_lock(&root->fs_info->tree_reloc_mutex);
7595         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7596         BUG_ON(ret);
7597         found = !list_empty(&root->fs_info->dead_reloc_roots);
7598         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7599
7600         if (found) {
7601                 trans = btrfs_start_transaction(root, 1);
7602                 BUG_ON(IS_ERR(trans));
7603                 ret = btrfs_commit_transaction(trans, root);
7604                 BUG_ON(ret);
7605         }
7606
7607         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7608         location.offset = (u64)-1;
7609         location.type = BTRFS_ROOT_ITEM_KEY;
7610
7611         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7612         BUG_ON(!reloc_root);
7613         btrfs_orphan_cleanup(reloc_root);
7614         return 0;
7615 }
7616
7617 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7618                                     struct btrfs_root *root)
7619 {
7620         struct btrfs_root *reloc_root;
7621         struct extent_buffer *eb;
7622         struct btrfs_root_item *root_item;
7623         struct btrfs_key root_key;
7624         int ret;
7625
7626         BUG_ON(!root->ref_cows);
7627         if (root->reloc_root)
7628                 return 0;
7629
7630         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7631         BUG_ON(!root_item);
7632
7633         ret = btrfs_copy_root(trans, root, root->commit_root,
7634                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7635         BUG_ON(ret);
7636
7637         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7638         root_key.offset = root->root_key.objectid;
7639         root_key.type = BTRFS_ROOT_ITEM_KEY;
7640
7641         memcpy(root_item, &root->root_item, sizeof(root_item));
7642         btrfs_set_root_refs(root_item, 0);
7643         btrfs_set_root_bytenr(root_item, eb->start);
7644         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7645         btrfs_set_root_generation(root_item, trans->transid);
7646
7647         btrfs_tree_unlock(eb);
7648         free_extent_buffer(eb);
7649
7650         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7651                                 &root_key, root_item);
7652         BUG_ON(ret);
7653         kfree(root_item);
7654
7655         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7656                                                  &root_key);
7657         BUG_ON(!reloc_root);
7658         reloc_root->last_trans = trans->transid;
7659         reloc_root->commit_root = NULL;
7660         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7661
7662         root->reloc_root = reloc_root;
7663         return 0;
7664 }
7665
7666 /*
7667  * Core function of space balance.
7668  *
7669  * The idea is using reloc trees to relocate tree blocks in reference
7670  * counted roots. There is one reloc tree for each subvol, and all
7671  * reloc trees share same root key objectid. Reloc trees are snapshots
7672  * of the latest committed roots of subvols (root->commit_root).
7673  *
7674  * To relocate a tree block referenced by a subvol, there are two steps.
7675  * COW the block through subvol's reloc tree, then update block pointer
7676  * in the subvol to point to the new block. Since all reloc trees share
7677  * same root key objectid, doing special handing for tree blocks owned
7678  * by them is easy. Once a tree block has been COWed in one reloc tree,
7679  * we can use the resulting new block directly when the same block is
7680  * required to COW again through other reloc trees. By this way, relocated
7681  * tree blocks are shared between reloc trees, so they are also shared
7682  * between subvols.
7683  */
7684 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7685                                       struct btrfs_root *root,
7686                                       struct btrfs_path *path,
7687                                       struct btrfs_key *first_key,
7688                                       struct btrfs_ref_path *ref_path,
7689                                       struct btrfs_block_group_cache *group,
7690                                       struct inode *reloc_inode)
7691 {
7692         struct btrfs_root *reloc_root;
7693         struct extent_buffer *eb = NULL;
7694         struct btrfs_key *keys;
7695         u64 *nodes;
7696         int level;
7697         int shared_level;
7698         int lowest_level = 0;
7699         int ret;
7700
7701         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7702                 lowest_level = ref_path->owner_objectid;
7703
7704         if (!root->ref_cows) {
7705                 path->lowest_level = lowest_level;
7706                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7707                 BUG_ON(ret < 0);
7708                 path->lowest_level = 0;
7709                 btrfs_release_path(root, path);
7710                 return 0;
7711         }
7712
7713         mutex_lock(&root->fs_info->tree_reloc_mutex);
7714         ret = init_reloc_tree(trans, root);
7715         BUG_ON(ret);
7716         reloc_root = root->reloc_root;
7717
7718         shared_level = ref_path->shared_level;
7719         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7720
7721         keys = ref_path->node_keys;
7722         nodes = ref_path->new_nodes;
7723         memset(&keys[shared_level + 1], 0,
7724                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7725         memset(&nodes[shared_level + 1], 0,
7726                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7727
7728         if (nodes[lowest_level] == 0) {
7729                 path->lowest_level = lowest_level;
7730                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7731                                         0, 1);
7732                 BUG_ON(ret);
7733                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7734                         eb = path->nodes[level];
7735                         if (!eb || eb == reloc_root->node)
7736                                 break;
7737                         nodes[level] = eb->start;
7738                         if (level == 0)
7739                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7740                         else
7741                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7742                 }
7743                 if (nodes[0] &&
7744                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7745                         eb = path->nodes[0];
7746                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7747                                                       group, reloc_inode);
7748                         BUG_ON(ret);
7749                 }
7750                 btrfs_release_path(reloc_root, path);
7751         } else {
7752                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7753                                        lowest_level);
7754                 BUG_ON(ret);
7755         }
7756
7757         /*
7758          * replace tree blocks in the fs tree with tree blocks in
7759          * the reloc tree.
7760          */
7761         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7762         BUG_ON(ret < 0);
7763
7764         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7765                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7766                                         0, 0);
7767                 BUG_ON(ret);
7768                 extent_buffer_get(path->nodes[0]);
7769                 eb = path->nodes[0];
7770                 btrfs_release_path(reloc_root, path);
7771                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7772                 BUG_ON(ret);
7773                 free_extent_buffer(eb);
7774         }
7775
7776         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7777         path->lowest_level = 0;
7778         return 0;
7779 }
7780
7781 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7782                                         struct btrfs_root *root,
7783                                         struct btrfs_path *path,
7784                                         struct btrfs_key *first_key,
7785                                         struct btrfs_ref_path *ref_path)
7786 {
7787         int ret;
7788
7789         ret = relocate_one_path(trans, root, path, first_key,
7790                                 ref_path, NULL, NULL);
7791         BUG_ON(ret);
7792
7793         return 0;
7794 }
7795
7796 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7797                                     struct btrfs_root *extent_root,
7798                                     struct btrfs_path *path,
7799                                     struct btrfs_key *extent_key)
7800 {
7801         int ret;
7802
7803         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7804         if (ret)
7805                 goto out;
7806         ret = btrfs_del_item(trans, extent_root, path);
7807 out:
7808         btrfs_release_path(extent_root, path);
7809         return ret;
7810 }
7811
7812 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7813                                                 struct btrfs_ref_path *ref_path)
7814 {
7815         struct btrfs_key root_key;
7816
7817         root_key.objectid = ref_path->root_objectid;
7818         root_key.type = BTRFS_ROOT_ITEM_KEY;
7819         if (is_cowonly_root(ref_path->root_objectid))
7820                 root_key.offset = 0;
7821         else
7822                 root_key.offset = (u64)-1;
7823
7824         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7825 }
7826
7827 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7828                                         struct btrfs_path *path,
7829                                         struct btrfs_key *extent_key,
7830                                         struct btrfs_block_group_cache *group,
7831                                         struct inode *reloc_inode, int pass)
7832 {
7833         struct btrfs_trans_handle *trans;
7834         struct btrfs_root *found_root;
7835         struct btrfs_ref_path *ref_path = NULL;
7836         struct disk_extent *new_extents = NULL;
7837         int nr_extents = 0;
7838         int loops;
7839         int ret;
7840         int level;
7841         struct btrfs_key first_key;
7842         u64 prev_block = 0;
7843
7844
7845         trans = btrfs_start_transaction(extent_root, 1);
7846         BUG_ON(IS_ERR(trans));
7847
7848         if (extent_key->objectid == 0) {
7849                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7850                 goto out;
7851         }
7852
7853         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7854         if (!ref_path) {
7855                 ret = -ENOMEM;
7856                 goto out;
7857         }
7858
7859         for (loops = 0; ; loops++) {
7860                 if (loops == 0) {
7861                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7862                                                    extent_key->objectid);
7863                 } else {
7864                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7865                 }
7866                 if (ret < 0)
7867                         goto out;
7868                 if (ret > 0)
7869                         break;
7870
7871                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7872                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7873                         continue;
7874
7875                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7876                 BUG_ON(!found_root);
7877                 /*
7878                  * for reference counted tree, only process reference paths
7879                  * rooted at the latest committed root.
7880                  */
7881                 if (found_root->ref_cows &&
7882                     ref_path->root_generation != found_root->root_key.offset)
7883                         continue;
7884
7885                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7886                         if (pass == 0) {
7887                                 /*
7888                                  * copy data extents to new locations
7889                                  */
7890                                 u64 group_start = group->key.objectid;
7891                                 ret = relocate_data_extent(reloc_inode,
7892                                                            extent_key,
7893                                                            group_start);
7894                                 if (ret < 0)
7895                                         goto out;
7896                                 break;
7897                         }
7898                         level = 0;
7899                 } else {
7900                         level = ref_path->owner_objectid;
7901                 }
7902
7903                 if (prev_block != ref_path->nodes[level]) {
7904                         struct extent_buffer *eb;
7905                         u64 block_start = ref_path->nodes[level];
7906                         u64 block_size = btrfs_level_size(found_root, level);
7907
7908                         eb = read_tree_block(found_root, block_start,
7909                                              block_size, 0);
7910                         btrfs_tree_lock(eb);
7911                         BUG_ON(level != btrfs_header_level(eb));
7912
7913                         if (level == 0)
7914                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7915                         else
7916                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7917
7918                         btrfs_tree_unlock(eb);
7919                         free_extent_buffer(eb);
7920                         prev_block = block_start;
7921                 }
7922
7923                 mutex_lock(&extent_root->fs_info->trans_mutex);
7924                 btrfs_record_root_in_trans(found_root);
7925                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7926                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7927                         /*
7928                          * try to update data extent references while
7929                          * keeping metadata shared between snapshots.
7930                          */
7931                         if (pass == 1) {
7932                                 ret = relocate_one_path(trans, found_root,
7933                                                 path, &first_key, ref_path,
7934                                                 group, reloc_inode);
7935                                 if (ret < 0)
7936                                         goto out;
7937                                 continue;
7938                         }
7939                         /*
7940                          * use fallback method to process the remaining
7941                          * references.
7942                          */
7943                         if (!new_extents) {
7944                                 u64 group_start = group->key.objectid;
7945                                 new_extents = kmalloc(sizeof(*new_extents),
7946                                                       GFP_NOFS);
7947                                 nr_extents = 1;
7948                                 ret = get_new_locations(reloc_inode,
7949                                                         extent_key,
7950                                                         group_start, 1,
7951                                                         &new_extents,
7952                                                         &nr_extents);
7953                                 if (ret)
7954                                         goto out;
7955                         }
7956                         ret = replace_one_extent(trans, found_root,
7957                                                 path, extent_key,
7958                                                 &first_key, ref_path,
7959                                                 new_extents, nr_extents);
7960                 } else {
7961                         ret = relocate_tree_block(trans, found_root, path,
7962                                                   &first_key, ref_path);
7963                 }
7964                 if (ret < 0)
7965                         goto out;
7966         }
7967         ret = 0;
7968 out:
7969         btrfs_end_transaction(trans, extent_root);
7970         kfree(new_extents);
7971         kfree(ref_path);
7972         return ret;
7973 }
7974 #endif
7975
7976 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7977 {
7978         u64 num_devices;
7979         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7980                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7981
7982         /*
7983          * we add in the count of missing devices because we want
7984          * to make sure that any RAID levels on a degraded FS
7985          * continue to be honored.
7986          */
7987         num_devices = root->fs_info->fs_devices->rw_devices +
7988                 root->fs_info->fs_devices->missing_devices;
7989
7990         if (num_devices == 1) {
7991                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7992                 stripped = flags & ~stripped;
7993
7994                 /* turn raid0 into single device chunks */
7995                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7996                         return stripped;
7997
7998                 /* turn mirroring into duplication */
7999                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8000                              BTRFS_BLOCK_GROUP_RAID10))
8001                         return stripped | BTRFS_BLOCK_GROUP_DUP;
8002                 return flags;
8003         } else {
8004                 /* they already had raid on here, just return */
8005                 if (flags & stripped)
8006                         return flags;
8007
8008                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8009                 stripped = flags & ~stripped;
8010
8011                 /* switch duplicated blocks with raid1 */
8012                 if (flags & BTRFS_BLOCK_GROUP_DUP)
8013                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
8014
8015                 /* turn single device chunks into raid0 */
8016                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8017         }
8018         return flags;
8019 }
8020
8021 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8022 {
8023         struct btrfs_space_info *sinfo = cache->space_info;
8024         u64 num_bytes;
8025         int ret = -ENOSPC;
8026
8027         if (cache->ro)
8028                 return 0;
8029
8030         spin_lock(&sinfo->lock);
8031         spin_lock(&cache->lock);
8032         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8033                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8034
8035         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8036             sinfo->bytes_may_use + sinfo->bytes_readonly +
8037             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8038                 sinfo->bytes_readonly += num_bytes;
8039                 sinfo->bytes_reserved += cache->reserved_pinned;
8040                 cache->reserved_pinned = 0;
8041                 cache->ro = 1;
8042                 ret = 0;
8043         }
8044
8045         spin_unlock(&cache->lock);
8046         spin_unlock(&sinfo->lock);
8047         return ret;
8048 }
8049
8050 int btrfs_set_block_group_ro(struct btrfs_root *root,
8051                              struct btrfs_block_group_cache *cache)
8052
8053 {
8054         struct btrfs_trans_handle *trans;
8055         u64 alloc_flags;
8056         int ret;
8057
8058         BUG_ON(cache->ro);
8059
8060         trans = btrfs_join_transaction(root, 1);
8061         BUG_ON(IS_ERR(trans));
8062
8063         alloc_flags = update_block_group_flags(root, cache->flags);
8064         if (alloc_flags != cache->flags)
8065                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8066
8067         ret = set_block_group_ro(cache);
8068         if (!ret)
8069                 goto out;
8070         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8071         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8072         if (ret < 0)
8073                 goto out;
8074         ret = set_block_group_ro(cache);
8075 out:
8076         btrfs_end_transaction(trans, root);
8077         return ret;
8078 }
8079
8080 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8081                             struct btrfs_root *root, u64 type)
8082 {
8083         u64 alloc_flags = get_alloc_profile(root, type);
8084         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8085 }
8086
8087 /*
8088  * helper to account the unused space of all the readonly block group in the
8089  * list. takes mirrors into account.
8090  */
8091 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8092 {
8093         struct btrfs_block_group_cache *block_group;
8094         u64 free_bytes = 0;
8095         int factor;
8096
8097         list_for_each_entry(block_group, groups_list, list) {
8098                 spin_lock(&block_group->lock);
8099
8100                 if (!block_group->ro) {
8101                         spin_unlock(&block_group->lock);
8102                         continue;
8103                 }
8104
8105                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8106                                           BTRFS_BLOCK_GROUP_RAID10 |
8107                                           BTRFS_BLOCK_GROUP_DUP))
8108                         factor = 2;
8109                 else
8110                         factor = 1;
8111
8112                 free_bytes += (block_group->key.offset -
8113                                btrfs_block_group_used(&block_group->item)) *
8114                                factor;
8115
8116                 spin_unlock(&block_group->lock);
8117         }
8118
8119         return free_bytes;
8120 }
8121
8122 /*
8123  * helper to account the unused space of all the readonly block group in the
8124  * space_info. takes mirrors into account.
8125  */
8126 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8127 {
8128         int i;
8129         u64 free_bytes = 0;
8130
8131         spin_lock(&sinfo->lock);
8132
8133         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8134                 if (!list_empty(&sinfo->block_groups[i]))
8135                         free_bytes += __btrfs_get_ro_block_group_free_space(
8136                                                 &sinfo->block_groups[i]);
8137
8138         spin_unlock(&sinfo->lock);
8139
8140         return free_bytes;
8141 }
8142
8143 int btrfs_set_block_group_rw(struct btrfs_root *root,
8144                               struct btrfs_block_group_cache *cache)
8145 {
8146         struct btrfs_space_info *sinfo = cache->space_info;
8147         u64 num_bytes;
8148
8149         BUG_ON(!cache->ro);
8150
8151         spin_lock(&sinfo->lock);
8152         spin_lock(&cache->lock);
8153         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8154                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8155         sinfo->bytes_readonly -= num_bytes;
8156         cache->ro = 0;
8157         spin_unlock(&cache->lock);
8158         spin_unlock(&sinfo->lock);
8159         return 0;
8160 }
8161
8162 /*
8163  * checks to see if its even possible to relocate this block group.
8164  *
8165  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8166  * ok to go ahead and try.
8167  */
8168 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8169 {
8170         struct btrfs_block_group_cache *block_group;
8171         struct btrfs_space_info *space_info;
8172         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8173         struct btrfs_device *device;
8174         int full = 0;
8175         int ret = 0;
8176
8177         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8178
8179         /* odd, couldn't find the block group, leave it alone */
8180         if (!block_group)
8181                 return -1;
8182
8183         /* no bytes used, we're good */
8184         if (!btrfs_block_group_used(&block_group->item))
8185                 goto out;
8186
8187         space_info = block_group->space_info;
8188         spin_lock(&space_info->lock);
8189
8190         full = space_info->full;
8191
8192         /*
8193          * if this is the last block group we have in this space, we can't
8194          * relocate it unless we're able to allocate a new chunk below.
8195          *
8196          * Otherwise, we need to make sure we have room in the space to handle
8197          * all of the extents from this block group.  If we can, we're good
8198          */
8199         if ((space_info->total_bytes != block_group->key.offset) &&
8200            (space_info->bytes_used + space_info->bytes_reserved +
8201             space_info->bytes_pinned + space_info->bytes_readonly +
8202             btrfs_block_group_used(&block_group->item) <
8203             space_info->total_bytes)) {
8204                 spin_unlock(&space_info->lock);
8205                 goto out;
8206         }
8207         spin_unlock(&space_info->lock);
8208
8209         /*
8210          * ok we don't have enough space, but maybe we have free space on our
8211          * devices to allocate new chunks for relocation, so loop through our
8212          * alloc devices and guess if we have enough space.  However, if we
8213          * were marked as full, then we know there aren't enough chunks, and we
8214          * can just return.
8215          */
8216         ret = -1;
8217         if (full)
8218                 goto out;
8219
8220         mutex_lock(&root->fs_info->chunk_mutex);
8221         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8222                 u64 min_free = btrfs_block_group_used(&block_group->item);
8223                 u64 dev_offset;
8224
8225                 /*
8226                  * check to make sure we can actually find a chunk with enough
8227                  * space to fit our block group in.
8228                  */
8229                 if (device->total_bytes > device->bytes_used + min_free) {
8230                         ret = find_free_dev_extent(NULL, device, min_free,
8231                                                    &dev_offset, NULL);
8232                         if (!ret)
8233                                 break;
8234                         ret = -1;
8235                 }
8236         }
8237         mutex_unlock(&root->fs_info->chunk_mutex);
8238 out:
8239         btrfs_put_block_group(block_group);
8240         return ret;
8241 }
8242
8243 static int find_first_block_group(struct btrfs_root *root,
8244                 struct btrfs_path *path, struct btrfs_key *key)
8245 {
8246         int ret = 0;
8247         struct btrfs_key found_key;
8248         struct extent_buffer *leaf;
8249         int slot;
8250
8251         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8252         if (ret < 0)
8253                 goto out;
8254
8255         while (1) {
8256                 slot = path->slots[0];
8257                 leaf = path->nodes[0];
8258                 if (slot >= btrfs_header_nritems(leaf)) {
8259                         ret = btrfs_next_leaf(root, path);
8260                         if (ret == 0)
8261                                 continue;
8262                         if (ret < 0)
8263                                 goto out;
8264                         break;
8265                 }
8266                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8267
8268                 if (found_key.objectid >= key->objectid &&
8269                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8270                         ret = 0;
8271                         goto out;
8272                 }
8273                 path->slots[0]++;
8274         }
8275 out:
8276         return ret;
8277 }
8278
8279 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8280 {
8281         struct btrfs_block_group_cache *block_group;
8282         u64 last = 0;
8283
8284         while (1) {
8285                 struct inode *inode;
8286
8287                 block_group = btrfs_lookup_first_block_group(info, last);
8288                 while (block_group) {
8289                         spin_lock(&block_group->lock);
8290                         if (block_group->iref)
8291                                 break;
8292                         spin_unlock(&block_group->lock);
8293                         block_group = next_block_group(info->tree_root,
8294                                                        block_group);
8295                 }
8296                 if (!block_group) {
8297                         if (last == 0)
8298                                 break;
8299                         last = 0;
8300                         continue;
8301                 }
8302
8303                 inode = block_group->inode;
8304                 block_group->iref = 0;
8305                 block_group->inode = NULL;
8306                 spin_unlock(&block_group->lock);
8307                 iput(inode);
8308                 last = block_group->key.objectid + block_group->key.offset;
8309                 btrfs_put_block_group(block_group);
8310         }
8311 }
8312
8313 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8314 {
8315         struct btrfs_block_group_cache *block_group;
8316         struct btrfs_space_info *space_info;
8317         struct btrfs_caching_control *caching_ctl;
8318         struct rb_node *n;
8319
8320         down_write(&info->extent_commit_sem);
8321         while (!list_empty(&info->caching_block_groups)) {
8322                 caching_ctl = list_entry(info->caching_block_groups.next,
8323                                          struct btrfs_caching_control, list);
8324                 list_del(&caching_ctl->list);
8325                 put_caching_control(caching_ctl);
8326         }
8327         up_write(&info->extent_commit_sem);
8328
8329         spin_lock(&info->block_group_cache_lock);
8330         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8331                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8332                                        cache_node);
8333                 rb_erase(&block_group->cache_node,
8334                          &info->block_group_cache_tree);
8335                 spin_unlock(&info->block_group_cache_lock);
8336
8337                 down_write(&block_group->space_info->groups_sem);
8338                 list_del(&block_group->list);
8339                 up_write(&block_group->space_info->groups_sem);
8340
8341                 if (block_group->cached == BTRFS_CACHE_STARTED)
8342                         wait_block_group_cache_done(block_group);
8343
8344                 /*
8345                  * We haven't cached this block group, which means we could
8346                  * possibly have excluded extents on this block group.
8347                  */
8348                 if (block_group->cached == BTRFS_CACHE_NO)
8349                         free_excluded_extents(info->extent_root, block_group);
8350
8351                 btrfs_remove_free_space_cache(block_group);
8352                 btrfs_put_block_group(block_group);
8353
8354                 spin_lock(&info->block_group_cache_lock);
8355         }
8356         spin_unlock(&info->block_group_cache_lock);
8357
8358         /* now that all the block groups are freed, go through and
8359          * free all the space_info structs.  This is only called during
8360          * the final stages of unmount, and so we know nobody is
8361          * using them.  We call synchronize_rcu() once before we start,
8362          * just to be on the safe side.
8363          */
8364         synchronize_rcu();
8365
8366         release_global_block_rsv(info);
8367
8368         while(!list_empty(&info->space_info)) {
8369                 space_info = list_entry(info->space_info.next,
8370                                         struct btrfs_space_info,
8371                                         list);
8372                 if (space_info->bytes_pinned > 0 ||
8373                     space_info->bytes_reserved > 0) {
8374                         WARN_ON(1);
8375                         dump_space_info(space_info, 0, 0);
8376                 }
8377                 list_del(&space_info->list);
8378                 kfree(space_info);
8379         }
8380         return 0;
8381 }
8382
8383 static void __link_block_group(struct btrfs_space_info *space_info,
8384                                struct btrfs_block_group_cache *cache)
8385 {
8386         int index = get_block_group_index(cache);
8387
8388         down_write(&space_info->groups_sem);
8389         list_add_tail(&cache->list, &space_info->block_groups[index]);
8390         up_write(&space_info->groups_sem);
8391 }
8392
8393 int btrfs_read_block_groups(struct btrfs_root *root)
8394 {
8395         struct btrfs_path *path;
8396         int ret;
8397         struct btrfs_block_group_cache *cache;
8398         struct btrfs_fs_info *info = root->fs_info;
8399         struct btrfs_space_info *space_info;
8400         struct btrfs_key key;
8401         struct btrfs_key found_key;
8402         struct extent_buffer *leaf;
8403         int need_clear = 0;
8404         u64 cache_gen;
8405
8406         root = info->extent_root;
8407         key.objectid = 0;
8408         key.offset = 0;
8409         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8410         path = btrfs_alloc_path();
8411         if (!path)
8412                 return -ENOMEM;
8413
8414         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8415         if (cache_gen != 0 &&
8416             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8417                 need_clear = 1;
8418         if (btrfs_test_opt(root, CLEAR_CACHE))
8419                 need_clear = 1;
8420         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8421                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8422
8423         while (1) {
8424                 ret = find_first_block_group(root, path, &key);
8425                 if (ret > 0)
8426                         break;
8427                 if (ret != 0)
8428                         goto error;
8429                 leaf = path->nodes[0];
8430                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8431                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8432                 if (!cache) {
8433                         ret = -ENOMEM;
8434                         goto error;
8435                 }
8436
8437                 atomic_set(&cache->count, 1);
8438                 spin_lock_init(&cache->lock);
8439                 spin_lock_init(&cache->tree_lock);
8440                 cache->fs_info = info;
8441                 INIT_LIST_HEAD(&cache->list);
8442                 INIT_LIST_HEAD(&cache->cluster_list);
8443
8444                 if (need_clear)
8445                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8446
8447                 /*
8448                  * we only want to have 32k of ram per block group for keeping
8449                  * track of free space, and if we pass 1/2 of that we want to
8450                  * start converting things over to using bitmaps
8451                  */
8452                 cache->extents_thresh = ((1024 * 32) / 2) /
8453                         sizeof(struct btrfs_free_space);
8454
8455                 read_extent_buffer(leaf, &cache->item,
8456                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8457                                    sizeof(cache->item));
8458                 memcpy(&cache->key, &found_key, sizeof(found_key));
8459
8460                 key.objectid = found_key.objectid + found_key.offset;
8461                 btrfs_release_path(root, path);
8462                 cache->flags = btrfs_block_group_flags(&cache->item);
8463                 cache->sectorsize = root->sectorsize;
8464
8465                 /*
8466                  * We need to exclude the super stripes now so that the space
8467                  * info has super bytes accounted for, otherwise we'll think
8468                  * we have more space than we actually do.
8469                  */
8470                 exclude_super_stripes(root, cache);
8471
8472                 /*
8473                  * check for two cases, either we are full, and therefore
8474                  * don't need to bother with the caching work since we won't
8475                  * find any space, or we are empty, and we can just add all
8476                  * the space in and be done with it.  This saves us _alot_ of
8477                  * time, particularly in the full case.
8478                  */
8479                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8480                         cache->last_byte_to_unpin = (u64)-1;
8481                         cache->cached = BTRFS_CACHE_FINISHED;
8482                         free_excluded_extents(root, cache);
8483                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8484                         cache->last_byte_to_unpin = (u64)-1;
8485                         cache->cached = BTRFS_CACHE_FINISHED;
8486                         add_new_free_space(cache, root->fs_info,
8487                                            found_key.objectid,
8488                                            found_key.objectid +
8489                                            found_key.offset);
8490                         free_excluded_extents(root, cache);
8491                 }
8492
8493                 ret = update_space_info(info, cache->flags, found_key.offset,
8494                                         btrfs_block_group_used(&cache->item),
8495                                         &space_info);
8496                 BUG_ON(ret);
8497                 cache->space_info = space_info;
8498                 spin_lock(&cache->space_info->lock);
8499                 cache->space_info->bytes_readonly += cache->bytes_super;
8500                 spin_unlock(&cache->space_info->lock);
8501
8502                 __link_block_group(space_info, cache);
8503
8504                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8505                 BUG_ON(ret);
8506
8507                 set_avail_alloc_bits(root->fs_info, cache->flags);
8508                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8509                         set_block_group_ro(cache);
8510         }
8511
8512         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8513                 if (!(get_alloc_profile(root, space_info->flags) &
8514                       (BTRFS_BLOCK_GROUP_RAID10 |
8515                        BTRFS_BLOCK_GROUP_RAID1 |
8516                        BTRFS_BLOCK_GROUP_DUP)))
8517                         continue;
8518                 /*
8519                  * avoid allocating from un-mirrored block group if there are
8520                  * mirrored block groups.
8521                  */
8522                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8523                         set_block_group_ro(cache);
8524                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8525                         set_block_group_ro(cache);
8526         }
8527
8528         init_global_block_rsv(info);
8529         ret = 0;
8530 error:
8531         btrfs_free_path(path);
8532         return ret;
8533 }
8534
8535 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8536                            struct btrfs_root *root, u64 bytes_used,
8537                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8538                            u64 size)
8539 {
8540         int ret;
8541         struct btrfs_root *extent_root;
8542         struct btrfs_block_group_cache *cache;
8543
8544         extent_root = root->fs_info->extent_root;
8545
8546         root->fs_info->last_trans_log_full_commit = trans->transid;
8547
8548         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8549         if (!cache)
8550                 return -ENOMEM;
8551
8552         cache->key.objectid = chunk_offset;
8553         cache->key.offset = size;
8554         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8555         cache->sectorsize = root->sectorsize;
8556         cache->fs_info = root->fs_info;
8557
8558         /*
8559          * we only want to have 32k of ram per block group for keeping track
8560          * of free space, and if we pass 1/2 of that we want to start
8561          * converting things over to using bitmaps
8562          */
8563         cache->extents_thresh = ((1024 * 32) / 2) /
8564                 sizeof(struct btrfs_free_space);
8565         atomic_set(&cache->count, 1);
8566         spin_lock_init(&cache->lock);
8567         spin_lock_init(&cache->tree_lock);
8568         INIT_LIST_HEAD(&cache->list);
8569         INIT_LIST_HEAD(&cache->cluster_list);
8570
8571         btrfs_set_block_group_used(&cache->item, bytes_used);
8572         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8573         cache->flags = type;
8574         btrfs_set_block_group_flags(&cache->item, type);
8575
8576         cache->last_byte_to_unpin = (u64)-1;
8577         cache->cached = BTRFS_CACHE_FINISHED;
8578         exclude_super_stripes(root, cache);
8579
8580         add_new_free_space(cache, root->fs_info, chunk_offset,
8581                            chunk_offset + size);
8582
8583         free_excluded_extents(root, cache);
8584
8585         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8586                                 &cache->space_info);
8587         BUG_ON(ret);
8588
8589         spin_lock(&cache->space_info->lock);
8590         cache->space_info->bytes_readonly += cache->bytes_super;
8591         spin_unlock(&cache->space_info->lock);
8592
8593         __link_block_group(cache->space_info, cache);
8594
8595         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8596         BUG_ON(ret);
8597
8598         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8599                                 sizeof(cache->item));
8600         BUG_ON(ret);
8601
8602         set_avail_alloc_bits(extent_root->fs_info, type);
8603
8604         return 0;
8605 }
8606
8607 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8608                              struct btrfs_root *root, u64 group_start)
8609 {
8610         struct btrfs_path *path;
8611         struct btrfs_block_group_cache *block_group;
8612         struct btrfs_free_cluster *cluster;
8613         struct btrfs_root *tree_root = root->fs_info->tree_root;
8614         struct btrfs_key key;
8615         struct inode *inode;
8616         int ret;
8617         int factor;
8618
8619         root = root->fs_info->extent_root;
8620
8621         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8622         BUG_ON(!block_group);
8623         BUG_ON(!block_group->ro);
8624
8625         memcpy(&key, &block_group->key, sizeof(key));
8626         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8627                                   BTRFS_BLOCK_GROUP_RAID1 |
8628                                   BTRFS_BLOCK_GROUP_RAID10))
8629                 factor = 2;
8630         else
8631                 factor = 1;
8632
8633         /* make sure this block group isn't part of an allocation cluster */
8634         cluster = &root->fs_info->data_alloc_cluster;
8635         spin_lock(&cluster->refill_lock);
8636         btrfs_return_cluster_to_free_space(block_group, cluster);
8637         spin_unlock(&cluster->refill_lock);
8638
8639         /*
8640          * make sure this block group isn't part of a metadata
8641          * allocation cluster
8642          */
8643         cluster = &root->fs_info->meta_alloc_cluster;
8644         spin_lock(&cluster->refill_lock);
8645         btrfs_return_cluster_to_free_space(block_group, cluster);
8646         spin_unlock(&cluster->refill_lock);
8647
8648         path = btrfs_alloc_path();
8649         BUG_ON(!path);
8650
8651         inode = lookup_free_space_inode(root, block_group, path);
8652         if (!IS_ERR(inode)) {
8653                 btrfs_orphan_add(trans, inode);
8654                 clear_nlink(inode);
8655                 /* One for the block groups ref */
8656                 spin_lock(&block_group->lock);
8657                 if (block_group->iref) {
8658                         block_group->iref = 0;
8659                         block_group->inode = NULL;
8660                         spin_unlock(&block_group->lock);
8661                         iput(inode);
8662                 } else {
8663                         spin_unlock(&block_group->lock);
8664                 }
8665                 /* One for our lookup ref */
8666                 iput(inode);
8667         }
8668
8669         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8670         key.offset = block_group->key.objectid;
8671         key.type = 0;
8672
8673         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8674         if (ret < 0)
8675                 goto out;
8676         if (ret > 0)
8677                 btrfs_release_path(tree_root, path);
8678         if (ret == 0) {
8679                 ret = btrfs_del_item(trans, tree_root, path);
8680                 if (ret)
8681                         goto out;
8682                 btrfs_release_path(tree_root, path);
8683         }
8684
8685         spin_lock(&root->fs_info->block_group_cache_lock);
8686         rb_erase(&block_group->cache_node,
8687                  &root->fs_info->block_group_cache_tree);
8688         spin_unlock(&root->fs_info->block_group_cache_lock);
8689
8690         down_write(&block_group->space_info->groups_sem);
8691         /*
8692          * we must use list_del_init so people can check to see if they
8693          * are still on the list after taking the semaphore
8694          */
8695         list_del_init(&block_group->list);
8696         up_write(&block_group->space_info->groups_sem);
8697
8698         if (block_group->cached == BTRFS_CACHE_STARTED)
8699                 wait_block_group_cache_done(block_group);
8700
8701         btrfs_remove_free_space_cache(block_group);
8702
8703         spin_lock(&block_group->space_info->lock);
8704         block_group->space_info->total_bytes -= block_group->key.offset;
8705         block_group->space_info->bytes_readonly -= block_group->key.offset;
8706         block_group->space_info->disk_total -= block_group->key.offset * factor;
8707         spin_unlock(&block_group->space_info->lock);
8708
8709         memcpy(&key, &block_group->key, sizeof(key));
8710
8711         btrfs_clear_space_info_full(root->fs_info);
8712
8713         btrfs_put_block_group(block_group);
8714         btrfs_put_block_group(block_group);
8715
8716         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8717         if (ret > 0)
8718                 ret = -EIO;
8719         if (ret < 0)
8720                 goto out;
8721
8722         ret = btrfs_del_item(trans, root, path);
8723 out:
8724         btrfs_free_path(path);
8725         return ret;
8726 }
8727
8728 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8729 {
8730         return unpin_extent_range(root, start, end);
8731 }
8732
8733 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8734                                u64 num_bytes)
8735 {
8736         return btrfs_discard_extent(root, bytenr, num_bytes);
8737 }