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 __btrfs_free_extent(struct btrfs_trans_handle *trans,
  40                                 struct btrfs_root *root,
  41                                 u64 bytenr, u64 num_bytes, u64 parent,
  42                                 u64 root_objectid, u64 owner_objectid,
  43                                 u64 owner_offset, int refs_to_drop,
  44                                 struct btrfs_delayed_extent_op *extra_op);
  45 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  46                                     struct extent_buffer *leaf,
  47                                     struct btrfs_extent_item *ei);
  48 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  49                                       struct btrfs_root *root,
  50                                       u64 parent, u64 root_objectid,
  51                                       u64 flags, u64 owner, u64 offset,
  52                                       struct btrfs_key *ins, int ref_mod);
  53 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  54                                      struct btrfs_root *root,
  55                                      u64 parent, u64 root_objectid,
  56                                      u64 flags, struct btrfs_disk_key *key,
  57                                      int level, struct btrfs_key *ins);
  58 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  59                           struct btrfs_root *extent_root, u64 alloc_bytes,
  60                           u64 flags, int force);
  61 static int find_next_key(struct btrfs_path *path, int level,
  62                          struct btrfs_key *key);
  63 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  64                             int dump_block_groups);
  65
  66 static noinline int
  67 block_group_cache_done(struct btrfs_block_group_cache *cache)
  68 {
  69         smp_mb();
  70         return cache->cached == BTRFS_CACHE_FINISHED;
  71 }
  72
  73 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  74 {
  75         return (cache->flags & bits) == bits;
  76 }
  77
  78 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  79 {
  80         atomic_inc(&cache->count);
  81 }
  82
  83 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
  84 {
  85         if (atomic_dec_and_test(&cache->count)) {
  86                 WARN_ON(cache->pinned > 0);
  87                 WARN_ON(cache->reserved > 0);
  88                 WARN_ON(cache->reserved_pinned > 0);
  89                 kfree(cache);
  90         }
  91 }
  92
  93 /*
  94  * this adds the block group to the fs_info rb tree for the block group
  95  * cache
  96  */
  97 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  98                                 struct btrfs_block_group_cache *block_group)
  99 {
 100         struct rb_node **p;
 101         struct rb_node *parent = NULL;
 102         struct btrfs_block_group_cache *cache;
 103
 104         spin_lock(&info->block_group_cache_lock);
 105         p = &info->block_group_cache_tree.rb_node;
 106
 107         while (*p) {
 108                 parent = *p;
 109                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 110                                  cache_node);
 111                 if (block_group->key.objectid < cache->key.objectid) {
 112                         p = &(*p)->rb_left;
 113                 } else if (block_group->key.objectid > cache->key.objectid) {
 114                         p = &(*p)->rb_right;
 115                 } else {
 116                         spin_unlock(&info->block_group_cache_lock);
 117                         return -EEXIST;
 118                 }
 119         }
 120
 121         rb_link_node(&block_group->cache_node, parent, p);
 122         rb_insert_color(&block_group->cache_node,
 123                         &info->block_group_cache_tree);
 124         spin_unlock(&info->block_group_cache_lock);
 125
 126         return 0;
 127 }
 128
 129 /*
 130  * This will return the block group at or after bytenr if contains is 0, else
 131  * it will return the block group that contains the bytenr
 132  */
 133 static struct btrfs_block_group_cache *
 134 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 135                               int contains)
 136 {
 137         struct btrfs_block_group_cache *cache, *ret = NULL;
 138         struct rb_node *n;
 139         u64 end, start;
 140
 141         spin_lock(&info->block_group_cache_lock);
 142         n = info->block_group_cache_tree.rb_node;
 143
 144         while (n) {
 145                 cache = rb_entry(n, struct btrfs_block_group_cache,
 146                                  cache_node);
 147                 end = cache->key.objectid + cache->key.offset - 1;
 148                 start = cache->key.objectid;
 149
 150                 if (bytenr < start) {
 151                         if (!contains && (!ret || start < ret->key.objectid))
 152                                 ret = cache;
 153                         n = n->rb_left;
 154                 } else if (bytenr > start) {
 155                         if (contains && bytenr <= end) {
 156                                 ret = cache;
 157                                 break;
 158                         }
 159                         n = n->rb_right;
 160                 } else {
 161                         ret = cache;
 162                         break;
 163                 }
 164         }
 165         if (ret)
 166                 btrfs_get_block_group(ret);
 167         spin_unlock(&info->block_group_cache_lock);
 168
 169         return ret;
 170 }
 171
 172 static int add_excluded_extent(struct btrfs_root *root,
 173                                u64 start, u64 num_bytes)
 174 {
 175         u64 end = start + num_bytes - 1;
 176         set_extent_bits(&root->fs_info->freed_extents[0],
 177                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 178         set_extent_bits(&root->fs_info->freed_extents[1],
 179                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 180         return 0;
 181 }
 182
 183 static void free_excluded_extents(struct btrfs_root *root,
 184                                   struct btrfs_block_group_cache *cache)
 185 {
 186         u64 start, end;
 187
 188         start = cache->key.objectid;
 189         end = start + cache->key.offset - 1;
 190
 191         clear_extent_bits(&root->fs_info->freed_extents[0],
 192                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 193         clear_extent_bits(&root->fs_info->freed_extents[1],
 194                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 195 }
 196
 197 static int exclude_super_stripes(struct btrfs_root *root,
 198                                  struct btrfs_block_group_cache *cache)
 199 {
 200         u64 bytenr;
 201         u64 *logical;
 202         int stripe_len;
 203         int i, nr, ret;
 204
 205         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 206                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 207                 cache->bytes_super += stripe_len;
 208                 ret = add_excluded_extent(root, cache->key.objectid,
 209                                           stripe_len);
 210                 BUG_ON(ret);
 211         }
 212
 213         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 214                 bytenr = btrfs_sb_offset(i);
 215                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 216                                        cache->key.objectid, bytenr,
 217                                        0, &logical, &nr, &stripe_len);
 218                 BUG_ON(ret);
 219
 220                 while (nr--) {
 221                         cache->bytes_super += stripe_len;
 222                         ret = add_excluded_extent(root, logical[nr],
 223                                                   stripe_len);
 224                         BUG_ON(ret);
 225                 }
 226
 227                 kfree(logical);
 228         }
 229         return 0;
 230 }
 231
 232 static struct btrfs_caching_control *
 233 get_caching_control(struct btrfs_block_group_cache *cache)
 234 {
 235         struct btrfs_caching_control *ctl;
 236
 237         spin_lock(&cache->lock);
 238         if (cache->cached != BTRFS_CACHE_STARTED) {
 239                 spin_unlock(&cache->lock);
 240                 return NULL;
 241         }
 242
 243         /* We're loading it the fast way, so we don't have a caching_ctl. */
 244         if (!cache->caching_ctl) {
 245                 spin_unlock(&cache->lock);
 246                 return NULL;
 247         }
 248
 249         ctl = cache->caching_ctl;
 250         atomic_inc(&ctl->count);
 251         spin_unlock(&cache->lock);
 252         return ctl;
 253 }
 254
 255 static void put_caching_control(struct btrfs_caching_control *ctl)
 256 {
 257         if (atomic_dec_and_test(&ctl->count))
 258                 kfree(ctl);
 259 }
 260
 261 /*
 262  * this is only called by cache_block_group, since we could have freed extents
 263  * we need to check the pinned_extents for any extents that can't be used yet
 264  * since their free space will be released as soon as the transaction commits.
 265  */
 266 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 267                               struct btrfs_fs_info *info, u64 start, u64 end)
 268 {
 269         u64 extent_start, extent_end, size, total_added = 0;
 270         int ret;
 271
 272         while (start < end) {
 273                 ret = find_first_extent_bit(info->pinned_extents, start,
 274                                             &extent_start, &extent_end,
 275                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 276                 if (ret)
 277                         break;
 278
 279                 if (extent_start <= start) {
 280                         start = extent_end + 1;
 281                 } else if (extent_start > start && extent_start < end) {
 282                         size = extent_start - start;
 283                         total_added += size;
 284                         ret = btrfs_add_free_space(block_group, start,
 285                                                    size);
 286                         BUG_ON(ret);
 287                         start = extent_end + 1;
 288                 } else {
 289                         break;
 290                 }
 291         }
 292
 293         if (start < end) {
 294                 size = end - start;
 295                 total_added += size;
 296                 ret = btrfs_add_free_space(block_group, start, size);
 297                 BUG_ON(ret);
 298         }
 299
 300         return total_added;
 301 }
 302
 303 static int caching_kthread(void *data)
 304 {
 305         struct btrfs_block_group_cache *block_group = data;
 306         struct btrfs_fs_info *fs_info = block_group->fs_info;
 307         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 308         struct btrfs_root *extent_root = fs_info->extent_root;
 309         struct btrfs_path *path;
 310         struct extent_buffer *leaf;
 311         struct btrfs_key key;
 312         u64 total_found = 0;
 313         u64 last = 0;
 314         u32 nritems;
 315         int ret = 0;
 316
 317         path = btrfs_alloc_path();
 318         if (!path)
 319                 return -ENOMEM;
 320
 321         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 322
 323         /*
 324          * We don't want to deadlock with somebody trying to allocate a new
 325          * extent for the extent root while also trying to search the extent
 326          * root to add free space.  So we skip locking and search the commit
 327          * root, since its read-only
 328          */
 329         path->skip_locking = 1;
 330         path->search_commit_root = 1;
 331         path->reada = 2;
 332
 333         key.objectid = last;
 334         key.offset = 0;
 335         key.type = BTRFS_EXTENT_ITEM_KEY;
 336 again:
 337         mutex_lock(&caching_ctl->mutex);
 338         /* need to make sure the commit_root doesn't disappear */
 339         down_read(&fs_info->extent_commit_sem);
 340
 341         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 342         if (ret < 0)
 343                 goto err;
 344
 345         leaf = path->nodes[0];
 346         nritems = btrfs_header_nritems(leaf);
 347
 348         while (1) {
 349                 smp_mb();
 350                 if (fs_info->closing > 1) {
 351                         last = (u64)-1;
 352                         break;
 353                 }
 354
 355                 if (path->slots[0] < nritems) {
 356                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 357                 } else {
 358                         ret = find_next_key(path, 0, &key);
 359                         if (ret)
 360                                 break;
 361
 362                         caching_ctl->progress = last;
 363                         btrfs_release_path(extent_root, path);
 364                         up_read(&fs_info->extent_commit_sem);
 365                         mutex_unlock(&caching_ctl->mutex);
 366                         if (btrfs_transaction_in_commit(fs_info))
 367                                 schedule_timeout(1);
 368                         else
 369                                 cond_resched();
 370                         goto again;
 371                 }
 372
 373                 if (key.objectid < block_group->key.objectid) {
 374                         path->slots[0]++;
 375                         continue;
 376                 }
 377
 378                 if (key.objectid >= block_group->key.objectid +
 379                     block_group->key.offset)
 380                         break;
 381
 382                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 383                         total_found += add_new_free_space(block_group,
 384                                                           fs_info, last,
 385                                                           key.objectid);
 386                         last = key.objectid + key.offset;
 387
 388                         if (total_found > (1024 * 1024 * 2)) {
 389                                 total_found = 0;
 390                                 wake_up(&caching_ctl->wait);
 391                         }
 392                 }
 393                 path->slots[0]++;
 394         }
 395         ret = 0;
 396
 397         total_found += add_new_free_space(block_group, fs_info, last,
 398                                           block_group->key.objectid +
 399                                           block_group->key.offset);
 400         caching_ctl->progress = (u64)-1;
 401
 402         spin_lock(&block_group->lock);
 403         block_group->caching_ctl = NULL;
 404         block_group->cached = BTRFS_CACHE_FINISHED;
 405         spin_unlock(&block_group->lock);
 406
 407 err:
 408         btrfs_free_path(path);
 409         up_read(&fs_info->extent_commit_sem);
 410
 411         free_excluded_extents(extent_root, block_group);
 412
 413         mutex_unlock(&caching_ctl->mutex);
 414         wake_up(&caching_ctl->wait);
 415
 416         put_caching_control(caching_ctl);
 417         atomic_dec(&block_group->space_info->caching_threads);
 418         btrfs_put_block_group(block_group);
 419
 420         return 0;
 421 }
 422
 423 static int cache_block_group(struct btrfs_block_group_cache *cache,
 424                              struct btrfs_trans_handle *trans,
 425                              struct btrfs_root *root,
 426                              int load_cache_only)
 427 {
 428         struct btrfs_fs_info *fs_info = cache->fs_info;
 429         struct btrfs_caching_control *caching_ctl;
 430         struct task_struct *tsk;
 431         int ret = 0;
 432
 433         smp_mb();
 434         if (cache->cached != BTRFS_CACHE_NO)
 435                 return 0;
 436
 437         /*
 438          * We can't do the read from on-disk cache during a commit since we need
 439          * to have the normal tree locking.  Also if we are currently trying to
 440          * allocate blocks for the tree root we can't do the fast caching since
 441          * we likely hold important locks.
 442          */
 443         if (trans && (!trans->transaction->in_commit) &&
 444             (root && root != root->fs_info->tree_root)) {
 445                 spin_lock(&cache->lock);
 446                 if (cache->cached != BTRFS_CACHE_NO) {
 447                         spin_unlock(&cache->lock);
 448                         return 0;
 449                 }
 450                 cache->cached = BTRFS_CACHE_STARTED;
 451                 spin_unlock(&cache->lock);
 452
 453                 ret = load_free_space_cache(fs_info, cache);
 454
 455                 spin_lock(&cache->lock);
 456                 if (ret == 1) {
 457                         cache->cached = BTRFS_CACHE_FINISHED;
 458                         cache->last_byte_to_unpin = (u64)-1;
 459                 } else {
 460                         cache->cached = BTRFS_CACHE_NO;
 461                 }
 462                 spin_unlock(&cache->lock);
 463                 if (ret == 1) {
 464                         free_excluded_extents(fs_info->extent_root, cache);
 465                         return 0;
 466                 }
 467         }
 468
 469         if (load_cache_only)
 470                 return 0;
 471
 472         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 473         BUG_ON(!caching_ctl);
 474
 475         INIT_LIST_HEAD(&caching_ctl->list);
 476         mutex_init(&caching_ctl->mutex);
 477         init_waitqueue_head(&caching_ctl->wait);
 478         caching_ctl->block_group = cache;
 479         caching_ctl->progress = cache->key.objectid;
 480         /* one for caching kthread, one for caching block group list */
 481         atomic_set(&caching_ctl->count, 2);
 482
 483         spin_lock(&cache->lock);
 484         if (cache->cached != BTRFS_CACHE_NO) {
 485                 spin_unlock(&cache->lock);
 486                 kfree(caching_ctl);
 487                 return 0;
 488         }
 489         cache->caching_ctl = caching_ctl;
 490         cache->cached = BTRFS_CACHE_STARTED;
 491         spin_unlock(&cache->lock);
 492
 493         down_write(&fs_info->extent_commit_sem);
 494         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 495         up_write(&fs_info->extent_commit_sem);
 496
 497         atomic_inc(&cache->space_info->caching_threads);
 498         btrfs_get_block_group(cache);
 499
 500         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 501                           cache->key.objectid);
 502         if (IS_ERR(tsk)) {
 503                 ret = PTR_ERR(tsk);
 504                 printk(KERN_ERR "error running thread %d\n", ret);
 505                 BUG();
 506         }
 507
 508         return ret;
 509 }
 510
 511 /*
 512  * return the block group that starts at or after bytenr
 513  */
 514 static struct btrfs_block_group_cache *
 515 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 516 {
 517         struct btrfs_block_group_cache *cache;
 518
 519         cache = block_group_cache_tree_search(info, bytenr, 0);
 520
 521         return cache;
 522 }
 523
 524 /*
 525  * return the block group that contains the given bytenr
 526  */
 527 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 528                                                  struct btrfs_fs_info *info,
 529                                                  u64 bytenr)
 530 {
 531         struct btrfs_block_group_cache *cache;
 532
 533         cache = block_group_cache_tree_search(info, bytenr, 1);
 534
 535         return cache;
 536 }
 537
 538 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 539                                                   u64 flags)
 540 {
 541         struct list_head *head = &info->space_info;
 542         struct btrfs_space_info *found;
 543
 544         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 545                  BTRFS_BLOCK_GROUP_METADATA;
 546
 547         rcu_read_lock();
 548         list_for_each_entry_rcu(found, head, list) {
 549                 if (found->flags & flags) {
 550                         rcu_read_unlock();
 551                         return found;
 552                 }
 553         }
 554         rcu_read_unlock();
 555         return NULL;
 556 }
 557
 558 /*
 559  * after adding space to the filesystem, we need to clear the full flags
 560  * on all the space infos.
 561  */
 562 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 563 {
 564         struct list_head *head = &info->space_info;
 565         struct btrfs_space_info *found;
 566
 567         rcu_read_lock();
 568         list_for_each_entry_rcu(found, head, list)
 569                 found->full = 0;
 570         rcu_read_unlock();
 571 }
 572
 573 static u64 div_factor(u64 num, int factor)
 574 {
 575         if (factor == 10)
 576                 return num;
 577         num *= factor;
 578         do_div(num, 10);
 579         return num;
 580 }
 581
 582 static u64 div_factor_fine(u64 num, int factor)
 583 {
 584         if (factor == 100)
 585                 return num;
 586         num *= factor;
 587         do_div(num, 100);
 588         return num;
 589 }
 590
 591 u64 btrfs_find_block_group(struct btrfs_root *root,
 592                            u64 search_start, u64 search_hint, int owner)
 593 {
 594         struct btrfs_block_group_cache *cache;
 595         u64 used;
 596         u64 last = max(search_hint, search_start);
 597         u64 group_start = 0;
 598         int full_search = 0;
 599         int factor = 9;
 600         int wrapped = 0;
 601 again:
 602         while (1) {
 603                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 604                 if (!cache)
 605                         break;
 606
 607                 spin_lock(&cache->lock);
 608                 last = cache->key.objectid + cache->key.offset;
 609                 used = btrfs_block_group_used(&cache->item);
 610
 611                 if ((full_search || !cache->ro) &&
 612                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 613                         if (used + cache->pinned + cache->reserved <
 614                             div_factor(cache->key.offset, factor)) {
 615                                 group_start = cache->key.objectid;
 616                                 spin_unlock(&cache->lock);
 617                                 btrfs_put_block_group(cache);
 618                                 goto found;
 619                         }
 620                 }
 621                 spin_unlock(&cache->lock);
 622                 btrfs_put_block_group(cache);
 623                 cond_resched();
 624         }
 625         if (!wrapped) {
 626                 last = search_start;
 627                 wrapped = 1;
 628                 goto again;
 629         }
 630         if (!full_search && factor < 10) {
 631                 last = search_start;
 632                 full_search = 1;
 633                 factor = 10;
 634                 goto again;
 635         }
 636 found:
 637         return group_start;
 638 }
 639
 640 /* simple helper to search for an existing extent at a given offset */
 641 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 642 {
 643         int ret;
 644         struct btrfs_key key;
 645         struct btrfs_path *path;
 646
 647         path = btrfs_alloc_path();
 648         BUG_ON(!path);
 649         key.objectid = start;
 650         key.offset = len;
 651         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 652         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 653                                 0, 0);
 654         btrfs_free_path(path);
 655         return ret;
 656 }
 657
 658 /*
 659  * helper function to lookup reference count and flags of extent.
 660  *
 661  * the head node for delayed ref is used to store the sum of all the
 662  * reference count modifications queued up in the rbtree. the head
 663  * node may also store the extent flags to set. This way you can check
 664  * to see what the reference count and extent flags would be if all of
 665  * the delayed refs are not processed.
 666  */
 667 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 668                              struct btrfs_root *root, u64 bytenr,
 669                              u64 num_bytes, u64 *refs, u64 *flags)
 670 {
 671         struct btrfs_delayed_ref_head *head;
 672         struct btrfs_delayed_ref_root *delayed_refs;
 673         struct btrfs_path *path;
 674         struct btrfs_extent_item *ei;
 675         struct extent_buffer *leaf;
 676         struct btrfs_key key;
 677         u32 item_size;
 678         u64 num_refs;
 679         u64 extent_flags;
 680         int ret;
 681
 682         path = btrfs_alloc_path();
 683         if (!path)
 684                 return -ENOMEM;
 685
 686         key.objectid = bytenr;
 687         key.type = BTRFS_EXTENT_ITEM_KEY;
 688         key.offset = num_bytes;
 689         if (!trans) {
 690                 path->skip_locking = 1;
 691                 path->search_commit_root = 1;
 692         }
 693 again:
 694         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 695                                 &key, path, 0, 0);
 696         if (ret < 0)
 697                 goto out_free;
 698
 699         if (ret == 0) {
 700                 leaf = path->nodes[0];
 701                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 702                 if (item_size >= sizeof(*ei)) {
 703                         ei = btrfs_item_ptr(leaf, path->slots[0],
 704                                             struct btrfs_extent_item);
 705                         num_refs = btrfs_extent_refs(leaf, ei);
 706                         extent_flags = btrfs_extent_flags(leaf, ei);
 707                 } else {
 708 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 709                         struct btrfs_extent_item_v0 *ei0;
 710                         BUG_ON(item_size != sizeof(*ei0));
 711                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 712                                              struct btrfs_extent_item_v0);
 713                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 714                         /* FIXME: this isn't correct for data */
 715                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 716 #else
 717                         BUG();
 718 #endif
 719                 }
 720                 BUG_ON(num_refs == 0);
 721         } else {
 722                 num_refs = 0;
 723                 extent_flags = 0;
 724                 ret = 0;
 725         }
 726
 727         if (!trans)
 728                 goto out;
 729
 730         delayed_refs = &trans->transaction->delayed_refs;
 731         spin_lock(&delayed_refs->lock);
 732         head = btrfs_find_delayed_ref_head(trans, bytenr);
 733         if (head) {
 734                 if (!mutex_trylock(&head->mutex)) {
 735                         atomic_inc(&head->node.refs);
 736                         spin_unlock(&delayed_refs->lock);
 737
 738                         btrfs_release_path(root->fs_info->extent_root, path);
 739
 740                         mutex_lock(&head->mutex);
 741                         mutex_unlock(&head->mutex);
 742                         btrfs_put_delayed_ref(&head->node);
 743                         goto again;
 744                 }
 745                 if (head->extent_op && head->extent_op->update_flags)
 746                         extent_flags |= head->extent_op->flags_to_set;
 747                 else
 748                         BUG_ON(num_refs == 0);
 749
 750                 num_refs += head->node.ref_mod;
 751                 mutex_unlock(&head->mutex);
 752         }
 753         spin_unlock(&delayed_refs->lock);
 754 out:
 755         WARN_ON(num_refs == 0);
 756         if (refs)
 757                 *refs = num_refs;
 758         if (flags)
 759                 *flags = extent_flags;
 760 out_free:
 761         btrfs_free_path(path);
 762         return ret;
 763 }
 764
 765 /*
 766  * Back reference rules.  Back refs have three main goals:
 767  *
 768  * 1) differentiate between all holders of references to an extent so that
 769  *    when a reference is dropped we can make sure it was a valid reference
 770  *    before freeing the extent.
 771  *
 772  * 2) Provide enough information to quickly find the holders of an extent
 773  *    if we notice a given block is corrupted or bad.
 774  *
 775  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 776  *    maintenance.  This is actually the same as #2, but with a slightly
 777  *    different use case.
 778  *
 779  * There are two kinds of back refs. The implicit back refs is optimized
 780  * for pointers in non-shared tree blocks. For a given pointer in a block,
 781  * back refs of this kind provide information about the block's owner tree
 782  * and the pointer's key. These information allow us to find the block by
 783  * b-tree searching. The full back refs is for pointers in tree blocks not
 784  * referenced by their owner trees. The location of tree block is recorded
 785  * in the back refs. Actually the full back refs is generic, and can be
 786  * used in all cases the implicit back refs is used. The major shortcoming
 787  * of the full back refs is its overhead. Every time a tree block gets
 788  * COWed, we have to update back refs entry for all pointers in it.
 789  *
 790  * For a newly allocated tree block, we use implicit back refs for
 791  * pointers in it. This means most tree related operations only involve
 792  * implicit back refs. For a tree block created in old transaction, the
 793  * only way to drop a reference to it is COW it. So we can detect the
 794  * event that tree block loses its owner tree's reference and do the
 795  * back refs conversion.
 796  *
 797  * When a tree block is COW'd through a tree, there are four cases:
 798  *
 799  * The reference count of the block is one and the tree is the block's
 800  * owner tree. Nothing to do in this case.
 801  *
 802  * The reference count of the block is one and the tree is not the
 803  * block's owner tree. In this case, full back refs is used for pointers
 804  * in the block. Remove these full back refs, add implicit back refs for
 805  * every pointers in the new block.
 806  *
 807  * The reference count of the block is greater than one and the tree is
 808  * the block's owner tree. In this case, implicit back refs is used for
 809  * pointers in the block. Add full back refs for every pointers in the
 810  * block, increase lower level extents' reference counts. The original
 811  * implicit back refs are entailed to the new block.
 812  *
 813  * The reference count of the block is greater than one and the tree is
 814  * not the block's owner tree. Add implicit back refs for every pointer in
 815  * the new block, increase lower level extents' reference count.
 816  *
 817  * Back Reference Key composing:
 818  *
 819  * The key objectid corresponds to the first byte in the extent,
 820  * The key type is used to differentiate between types of back refs.
 821  * There are different meanings of the key offset for different types
 822  * of back refs.
 823  *
 824  * File extents can be referenced by:
 825  *
 826  * - multiple snapshots, subvolumes, or different generations in one subvol
 827  * - different files inside a single subvolume
 828  * - different offsets inside a file (bookend extents in file.c)
 829  *
 830  * The extent ref structure for the implicit back refs has fields for:
 831  *
 832  * - Objectid of the subvolume root
 833  * - objectid of the file holding the reference
 834  * - original offset in the file
 835  * - how many bookend extents
 836  *
 837  * The key offset for the implicit back refs is hash of the first
 838  * three fields.
 839  *
 840  * The extent ref structure for the full back refs has field for:
 841  *
 842  * - number of pointers in the tree leaf
 843  *
 844  * The key offset for the implicit back refs is the first byte of
 845  * the tree leaf
 846  *
 847  * When a file extent is allocated, The implicit back refs is used.
 848  * the fields are filled in:
 849  *
 850  *     (root_key.objectid, inode objectid, offset in file, 1)
 851  *
 852  * When a file extent is removed file truncation, we find the
 853  * corresponding implicit back refs and check the following fields:
 854  *
 855  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 856  *
 857  * Btree extents can be referenced by:
 858  *
 859  * - Different subvolumes
 860  *
 861  * Both the implicit back refs and the full back refs for tree blocks
 862  * only consist of key. The key offset for the implicit back refs is
 863  * objectid of block's owner tree. The key offset for the full back refs
 864  * is the first byte of parent block.
 865  *
 866  * When implicit back refs is used, information about the lowest key and
 867  * level of the tree block are required. These information are stored in
 868  * tree block info structure.
 869  */
 870
 871 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 872 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 873                                   struct btrfs_root *root,
 874                                   struct btrfs_path *path,
 875                                   u64 owner, u32 extra_size)
 876 {
 877         struct btrfs_extent_item *item;
 878         struct btrfs_extent_item_v0 *ei0;
 879         struct btrfs_extent_ref_v0 *ref0;
 880         struct btrfs_tree_block_info *bi;
 881         struct extent_buffer *leaf;
 882         struct btrfs_key key;
 883         struct btrfs_key found_key;
 884         u32 new_size = sizeof(*item);
 885         u64 refs;
 886         int ret;
 887
 888         leaf = path->nodes[0];
 889         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 890
 891         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 892         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 893                              struct btrfs_extent_item_v0);
 894         refs = btrfs_extent_refs_v0(leaf, ei0);
 895
 896         if (owner == (u64)-1) {
 897                 while (1) {
 898                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 899                                 ret = btrfs_next_leaf(root, path);
 900                                 if (ret < 0)
 901                                         return ret;
 902                                 BUG_ON(ret > 0);
 903                                 leaf = path->nodes[0];
 904                         }
 905                         btrfs_item_key_to_cpu(leaf, &found_key,
 906                                               path->slots[0]);
 907                         BUG_ON(key.objectid != found_key.objectid);
 908                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 909                                 path->slots[0]++;
 910                                 continue;
 911                         }
 912                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 913                                               struct btrfs_extent_ref_v0);
 914                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 915                         break;
 916                 }
 917         }
 918         btrfs_release_path(root, path);
 919
 920         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 921                 new_size += sizeof(*bi);
 922
 923         new_size -= sizeof(*ei0);
 924         ret = btrfs_search_slot(trans, root, &key, path,
 925                                 new_size + extra_size, 1);
 926         if (ret < 0)
 927                 return ret;
 928         BUG_ON(ret);
 929
 930         ret = btrfs_extend_item(trans, root, path, new_size);
 931         BUG_ON(ret);
 932
 933         leaf = path->nodes[0];
 934         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 935         btrfs_set_extent_refs(leaf, item, refs);
 936         /* FIXME: get real generation */
 937         btrfs_set_extent_generation(leaf, item, 0);
 938         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 939                 btrfs_set_extent_flags(leaf, item,
 940                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 941                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 942                 bi = (struct btrfs_tree_block_info *)(item + 1);
 943                 /* FIXME: get first key of the block */
 944                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 945                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 946         } else {
 947                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 948         }
 949         btrfs_mark_buffer_dirty(leaf);
 950         return 0;
 951 }
 952 #endif
 953
 954 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 955 {
 956         u32 high_crc = ~(u32)0;
 957         u32 low_crc = ~(u32)0;
 958         __le64 lenum;
 959
 960         lenum = cpu_to_le64(root_objectid);
 961         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 962         lenum = cpu_to_le64(owner);
 963         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 964         lenum = cpu_to_le64(offset);
 965         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 966
 967         return ((u64)high_crc << 31) ^ (u64)low_crc;
 968 }
 969
 970 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 971                                      struct btrfs_extent_data_ref *ref)
 972 {
 973         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 974                                     btrfs_extent_data_ref_objectid(leaf, ref),
 975                                     btrfs_extent_data_ref_offset(leaf, ref));
 976 }
 977
 978 static int match_extent_data_ref(struct extent_buffer *leaf,
 979                                  struct btrfs_extent_data_ref *ref,
 980                                  u64 root_objectid, u64 owner, u64 offset)
 981 {
 982         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 983             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 984             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 985                 return 0;
 986         return 1;
 987 }
 988
 989 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 990                                            struct btrfs_root *root,
 991                                            struct btrfs_path *path,
 992                                            u64 bytenr, u64 parent,
 993                                            u64 root_objectid,
 994                                            u64 owner, u64 offset)
 995 {
 996         struct btrfs_key key;
 997         struct btrfs_extent_data_ref *ref;
 998         struct extent_buffer *leaf;
 999         u32 nritems;
1000         int ret;
1001         int recow;
1002         int err = -ENOENT;
1003
1004         key.objectid = bytenr;
1005         if (parent) {
1006                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1007                 key.offset = parent;
1008         } else {
1009                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1010                 key.offset = hash_extent_data_ref(root_objectid,
1011                                                   owner, offset);
1012         }
1013 again:
1014         recow = 0;
1015         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1016         if (ret < 0) {
1017                 err = ret;
1018                 goto fail;
1019         }
1020
1021         if (parent) {
1022                 if (!ret)
1023                         return 0;
1024 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1025                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1026                 btrfs_release_path(root, path);
1027                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1028                 if (ret < 0) {
1029                         err = ret;
1030                         goto fail;
1031                 }
1032                 if (!ret)
1033                         return 0;
1034 #endif
1035                 goto fail;
1036         }
1037
1038         leaf = path->nodes[0];
1039         nritems = btrfs_header_nritems(leaf);
1040         while (1) {
1041                 if (path->slots[0] >= nritems) {
1042                         ret = btrfs_next_leaf(root, path);
1043                         if (ret < 0)
1044                                 err = ret;
1045                         if (ret)
1046                                 goto fail;
1047
1048                         leaf = path->nodes[0];
1049                         nritems = btrfs_header_nritems(leaf);
1050                         recow = 1;
1051                 }
1052
1053                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1054                 if (key.objectid != bytenr ||
1055                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1056                         goto fail;
1057
1058                 ref = btrfs_item_ptr(leaf, path->slots[0],
1059                                      struct btrfs_extent_data_ref);
1060
1061                 if (match_extent_data_ref(leaf, ref, root_objectid,
1062                                           owner, offset)) {
1063                         if (recow) {
1064                                 btrfs_release_path(root, path);
1065                                 goto again;
1066                         }
1067                         err = 0;
1068                         break;
1069                 }
1070                 path->slots[0]++;
1071         }
1072 fail:
1073         return err;
1074 }
1075
1076 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1077                                            struct btrfs_root *root,
1078                                            struct btrfs_path *path,
1079                                            u64 bytenr, u64 parent,
1080                                            u64 root_objectid, u64 owner,
1081                                            u64 offset, int refs_to_add)
1082 {
1083         struct btrfs_key key;
1084         struct extent_buffer *leaf;
1085         u32 size;
1086         u32 num_refs;
1087         int ret;
1088
1089         key.objectid = bytenr;
1090         if (parent) {
1091                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1092                 key.offset = parent;
1093                 size = sizeof(struct btrfs_shared_data_ref);
1094         } else {
1095                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1096                 key.offset = hash_extent_data_ref(root_objectid,
1097                                                   owner, offset);
1098                 size = sizeof(struct btrfs_extent_data_ref);
1099         }
1100
1101         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1102         if (ret && ret != -EEXIST)
1103                 goto fail;
1104
1105         leaf = path->nodes[0];
1106         if (parent) {
1107                 struct btrfs_shared_data_ref *ref;
1108                 ref = btrfs_item_ptr(leaf, path->slots[0],
1109                                      struct btrfs_shared_data_ref);
1110                 if (ret == 0) {
1111                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1112                 } else {
1113                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1114                         num_refs += refs_to_add;
1115                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1116                 }
1117         } else {
1118                 struct btrfs_extent_data_ref *ref;
1119                 while (ret == -EEXIST) {
1120                         ref = btrfs_item_ptr(leaf, path->slots[0],
1121                                              struct btrfs_extent_data_ref);
1122                         if (match_extent_data_ref(leaf, ref, root_objectid,
1123                                                   owner, offset))
1124                                 break;
1125                         btrfs_release_path(root, path);
1126                         key.offset++;
1127                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1128                                                       size);
1129                         if (ret && ret != -EEXIST)
1130                                 goto fail;
1131
1132                         leaf = path->nodes[0];
1133                 }
1134                 ref = btrfs_item_ptr(leaf, path->slots[0],
1135                                      struct btrfs_extent_data_ref);
1136                 if (ret == 0) {
1137                         btrfs_set_extent_data_ref_root(leaf, ref,
1138                                                        root_objectid);
1139                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1140                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1141                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1142                 } else {
1143                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1144                         num_refs += refs_to_add;
1145                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1146                 }
1147         }
1148         btrfs_mark_buffer_dirty(leaf);
1149         ret = 0;
1150 fail:
1151         btrfs_release_path(root, path);
1152         return ret;
1153 }
1154
1155 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1156                                            struct btrfs_root *root,
1157                                            struct btrfs_path *path,
1158                                            int refs_to_drop)
1159 {
1160         struct btrfs_key key;
1161         struct btrfs_extent_data_ref *ref1 = NULL;
1162         struct btrfs_shared_data_ref *ref2 = NULL;
1163         struct extent_buffer *leaf;
1164         u32 num_refs = 0;
1165         int ret = 0;
1166
1167         leaf = path->nodes[0];
1168         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1169
1170         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1171                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1172                                       struct btrfs_extent_data_ref);
1173                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1174         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1175                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1176                                       struct btrfs_shared_data_ref);
1177                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1178 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1179         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1180                 struct btrfs_extent_ref_v0 *ref0;
1181                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1182                                       struct btrfs_extent_ref_v0);
1183                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1184 #endif
1185         } else {
1186                 BUG();
1187         }
1188
1189         BUG_ON(num_refs < refs_to_drop);
1190         num_refs -= refs_to_drop;
1191
1192         if (num_refs == 0) {
1193                 ret = btrfs_del_item(trans, root, path);
1194         } else {
1195                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1196                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1197                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1198                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1199 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1200                 else {
1201                         struct btrfs_extent_ref_v0 *ref0;
1202                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1203                                         struct btrfs_extent_ref_v0);
1204                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1205                 }
1206 #endif
1207                 btrfs_mark_buffer_dirty(leaf);
1208         }
1209         return ret;
1210 }
1211
1212 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1213                                           struct btrfs_path *path,
1214                                           struct btrfs_extent_inline_ref *iref)
1215 {
1216         struct btrfs_key key;
1217         struct extent_buffer *leaf;
1218         struct btrfs_extent_data_ref *ref1;
1219         struct btrfs_shared_data_ref *ref2;
1220         u32 num_refs = 0;
1221
1222         leaf = path->nodes[0];
1223         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1224         if (iref) {
1225                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1226                     BTRFS_EXTENT_DATA_REF_KEY) {
1227                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1228                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1229                 } else {
1230                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1231                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1232                 }
1233         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1234                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1235                                       struct btrfs_extent_data_ref);
1236                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1237         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1238                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1239                                       struct btrfs_shared_data_ref);
1240                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1241 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1242         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1243                 struct btrfs_extent_ref_v0 *ref0;
1244                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1245                                       struct btrfs_extent_ref_v0);
1246                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1247 #endif
1248         } else {
1249                 WARN_ON(1);
1250         }
1251         return num_refs;
1252 }
1253
1254 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1255                                           struct btrfs_root *root,
1256                                           struct btrfs_path *path,
1257                                           u64 bytenr, u64 parent,
1258                                           u64 root_objectid)
1259 {
1260         struct btrfs_key key;
1261         int ret;
1262
1263         key.objectid = bytenr;
1264         if (parent) {
1265                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1266                 key.offset = parent;
1267         } else {
1268                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1269                 key.offset = root_objectid;
1270         }
1271
1272         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1273         if (ret > 0)
1274                 ret = -ENOENT;
1275 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1276         if (ret == -ENOENT && parent) {
1277                 btrfs_release_path(root, path);
1278                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1279                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1280                 if (ret > 0)
1281                         ret = -ENOENT;
1282         }
1283 #endif
1284         return ret;
1285 }
1286
1287 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1288                                           struct btrfs_root *root,
1289                                           struct btrfs_path *path,
1290                                           u64 bytenr, u64 parent,
1291                                           u64 root_objectid)
1292 {
1293         struct btrfs_key key;
1294         int ret;
1295
1296         key.objectid = bytenr;
1297         if (parent) {
1298                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1299                 key.offset = parent;
1300         } else {
1301                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1302                 key.offset = root_objectid;
1303         }
1304
1305         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1306         btrfs_release_path(root, path);
1307         return ret;
1308 }
1309
1310 static inline int extent_ref_type(u64 parent, u64 owner)
1311 {
1312         int type;
1313         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1314                 if (parent > 0)
1315                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1316                 else
1317                         type = BTRFS_TREE_BLOCK_REF_KEY;
1318         } else {
1319                 if (parent > 0)
1320                         type = BTRFS_SHARED_DATA_REF_KEY;
1321                 else
1322                         type = BTRFS_EXTENT_DATA_REF_KEY;
1323         }
1324         return type;
1325 }
1326
1327 static int find_next_key(struct btrfs_path *path, int level,
1328                          struct btrfs_key *key)
1329
1330 {
1331         for (; level < BTRFS_MAX_LEVEL; level++) {
1332                 if (!path->nodes[level])
1333                         break;
1334                 if (path->slots[level] + 1 >=
1335                     btrfs_header_nritems(path->nodes[level]))
1336                         continue;
1337                 if (level == 0)
1338                         btrfs_item_key_to_cpu(path->nodes[level], key,
1339                                               path->slots[level] + 1);
1340                 else
1341                         btrfs_node_key_to_cpu(path->nodes[level], key,
1342                                               path->slots[level] + 1);
1343                 return 0;
1344         }
1345         return 1;
1346 }
1347
1348 /*
1349  * look for inline back ref. if back ref is found, *ref_ret is set
1350  * to the address of inline back ref, and 0 is returned.
1351  *
1352  * if back ref isn't found, *ref_ret is set to the address where it
1353  * should be inserted, and -ENOENT is returned.
1354  *
1355  * if insert is true and there are too many inline back refs, the path
1356  * points to the extent item, and -EAGAIN is returned.
1357  *
1358  * NOTE: inline back refs are ordered in the same way that back ref
1359  *       items in the tree are ordered.
1360  */
1361 static noinline_for_stack
1362 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1363                                  struct btrfs_root *root,
1364                                  struct btrfs_path *path,
1365                                  struct btrfs_extent_inline_ref **ref_ret,
1366                                  u64 bytenr, u64 num_bytes,
1367                                  u64 parent, u64 root_objectid,
1368                                  u64 owner, u64 offset, int insert)
1369 {
1370         struct btrfs_key key;
1371         struct extent_buffer *leaf;
1372         struct btrfs_extent_item *ei;
1373         struct btrfs_extent_inline_ref *iref;
1374         u64 flags;
1375         u64 item_size;
1376         unsigned long ptr;
1377         unsigned long end;
1378         int extra_size;
1379         int type;
1380         int want;
1381         int ret;
1382         int err = 0;
1383
1384         key.objectid = bytenr;
1385         key.type = BTRFS_EXTENT_ITEM_KEY;
1386         key.offset = num_bytes;
1387
1388         want = extent_ref_type(parent, owner);
1389         if (insert) {
1390                 extra_size = btrfs_extent_inline_ref_size(want);
1391                 path->keep_locks = 1;
1392         } else
1393                 extra_size = -1;
1394         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1395         if (ret < 0) {
1396                 err = ret;
1397                 goto out;
1398         }
1399         BUG_ON(ret);
1400
1401         leaf = path->nodes[0];
1402         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1403 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1404         if (item_size < sizeof(*ei)) {
1405                 if (!insert) {
1406                         err = -ENOENT;
1407                         goto out;
1408                 }
1409                 ret = convert_extent_item_v0(trans, root, path, owner,
1410                                              extra_size);
1411                 if (ret < 0) {
1412                         err = ret;
1413                         goto out;
1414                 }
1415                 leaf = path->nodes[0];
1416                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1417         }
1418 #endif
1419         BUG_ON(item_size < sizeof(*ei));
1420
1421         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1422         flags = btrfs_extent_flags(leaf, ei);
1423
1424         ptr = (unsigned long)(ei + 1);
1425         end = (unsigned long)ei + item_size;
1426
1427         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1428                 ptr += sizeof(struct btrfs_tree_block_info);
1429                 BUG_ON(ptr > end);
1430         } else {
1431                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1432         }
1433
1434         err = -ENOENT;
1435         while (1) {
1436                 if (ptr >= end) {
1437                         WARN_ON(ptr > end);
1438                         break;
1439                 }
1440                 iref = (struct btrfs_extent_inline_ref *)ptr;
1441                 type = btrfs_extent_inline_ref_type(leaf, iref);
1442                 if (want < type)
1443                         break;
1444                 if (want > type) {
1445                         ptr += btrfs_extent_inline_ref_size(type);
1446                         continue;
1447                 }
1448
1449                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1450                         struct btrfs_extent_data_ref *dref;
1451                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1452                         if (match_extent_data_ref(leaf, dref, root_objectid,
1453                                                   owner, offset)) {
1454                                 err = 0;
1455                                 break;
1456                         }
1457                         if (hash_extent_data_ref_item(leaf, dref) <
1458                             hash_extent_data_ref(root_objectid, owner, offset))
1459                                 break;
1460                 } else {
1461                         u64 ref_offset;
1462                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1463                         if (parent > 0) {
1464                                 if (parent == ref_offset) {
1465                                         err = 0;
1466                                         break;
1467                                 }
1468                                 if (ref_offset < parent)
1469                                         break;
1470                         } else {
1471                                 if (root_objectid == ref_offset) {
1472                                         err = 0;
1473                                         break;
1474                                 }
1475                                 if (ref_offset < root_objectid)
1476                                         break;
1477                         }
1478                 }
1479                 ptr += btrfs_extent_inline_ref_size(type);
1480         }
1481         if (err == -ENOENT && insert) {
1482                 if (item_size + extra_size >=
1483                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1484                         err = -EAGAIN;
1485                         goto out;
1486                 }
1487                 /*
1488                  * To add new inline back ref, we have to make sure
1489                  * there is no corresponding back ref item.
1490                  * For simplicity, we just do not add new inline back
1491                  * ref if there is any kind of item for this block
1492                  */
1493                 if (find_next_key(path, 0, &key) == 0 &&
1494                     key.objectid == bytenr &&
1495                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1496                         err = -EAGAIN;
1497                         goto out;
1498                 }
1499         }
1500         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1501 out:
1502         if (insert) {
1503                 path->keep_locks = 0;
1504                 btrfs_unlock_up_safe(path, 1);
1505         }
1506         return err;
1507 }
1508
1509 /*
1510  * helper to add new inline back ref
1511  */
1512 static noinline_for_stack
1513 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1514                                 struct btrfs_root *root,
1515                                 struct btrfs_path *path,
1516                                 struct btrfs_extent_inline_ref *iref,
1517                                 u64 parent, u64 root_objectid,
1518                                 u64 owner, u64 offset, int refs_to_add,
1519                                 struct btrfs_delayed_extent_op *extent_op)
1520 {
1521         struct extent_buffer *leaf;
1522         struct btrfs_extent_item *ei;
1523         unsigned long ptr;
1524         unsigned long end;
1525         unsigned long item_offset;
1526         u64 refs;
1527         int size;
1528         int type;
1529         int ret;
1530
1531         leaf = path->nodes[0];
1532         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1533         item_offset = (unsigned long)iref - (unsigned long)ei;
1534
1535         type = extent_ref_type(parent, owner);
1536         size = btrfs_extent_inline_ref_size(type);
1537
1538         ret = btrfs_extend_item(trans, root, path, size);
1539         BUG_ON(ret);
1540
1541         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1542         refs = btrfs_extent_refs(leaf, ei);
1543         refs += refs_to_add;
1544         btrfs_set_extent_refs(leaf, ei, refs);
1545         if (extent_op)
1546                 __run_delayed_extent_op(extent_op, leaf, ei);
1547
1548         ptr = (unsigned long)ei + item_offset;
1549         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1550         if (ptr < end - size)
1551                 memmove_extent_buffer(leaf, ptr + size, ptr,
1552                                       end - size - ptr);
1553
1554         iref = (struct btrfs_extent_inline_ref *)ptr;
1555         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1556         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1557                 struct btrfs_extent_data_ref *dref;
1558                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1559                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1560                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1561                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1562                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1563         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1564                 struct btrfs_shared_data_ref *sref;
1565                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1566                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1567                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1568         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1569                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1570         } else {
1571                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1572         }
1573         btrfs_mark_buffer_dirty(leaf);
1574         return 0;
1575 }
1576
1577 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1578                                  struct btrfs_root *root,
1579                                  struct btrfs_path *path,
1580                                  struct btrfs_extent_inline_ref **ref_ret,
1581                                  u64 bytenr, u64 num_bytes, u64 parent,
1582                                  u64 root_objectid, u64 owner, u64 offset)
1583 {
1584         int ret;
1585
1586         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1587                                            bytenr, num_bytes, parent,
1588                                            root_objectid, owner, offset, 0);
1589         if (ret != -ENOENT)
1590                 return ret;
1591
1592         btrfs_release_path(root, path);
1593         *ref_ret = NULL;
1594
1595         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1596                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1597                                             root_objectid);
1598         } else {
1599                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1600                                              root_objectid, owner, offset);
1601         }
1602         return ret;
1603 }
1604
1605 /*
1606  * helper to update/remove inline back ref
1607  */
1608 static noinline_for_stack
1609 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1610                                  struct btrfs_root *root,
1611                                  struct btrfs_path *path,
1612                                  struct btrfs_extent_inline_ref *iref,
1613                                  int refs_to_mod,
1614                                  struct btrfs_delayed_extent_op *extent_op)
1615 {
1616         struct extent_buffer *leaf;
1617         struct btrfs_extent_item *ei;
1618         struct btrfs_extent_data_ref *dref = NULL;
1619         struct btrfs_shared_data_ref *sref = NULL;
1620         unsigned long ptr;
1621         unsigned long end;
1622         u32 item_size;
1623         int size;
1624         int type;
1625         int ret;
1626         u64 refs;
1627
1628         leaf = path->nodes[0];
1629         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1630         refs = btrfs_extent_refs(leaf, ei);
1631         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1632         refs += refs_to_mod;
1633         btrfs_set_extent_refs(leaf, ei, refs);
1634         if (extent_op)
1635                 __run_delayed_extent_op(extent_op, leaf, ei);
1636
1637         type = btrfs_extent_inline_ref_type(leaf, iref);
1638
1639         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1640                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1641                 refs = btrfs_extent_data_ref_count(leaf, dref);
1642         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1643                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1644                 refs = btrfs_shared_data_ref_count(leaf, sref);
1645         } else {
1646                 refs = 1;
1647                 BUG_ON(refs_to_mod != -1);
1648         }
1649
1650         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1651         refs += refs_to_mod;
1652
1653         if (refs > 0) {
1654                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1655                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1656                 else
1657                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1658         } else {
1659                 size =  btrfs_extent_inline_ref_size(type);
1660                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1661                 ptr = (unsigned long)iref;
1662                 end = (unsigned long)ei + item_size;
1663                 if (ptr + size < end)
1664                         memmove_extent_buffer(leaf, ptr, ptr + size,
1665                                               end - ptr - size);
1666                 item_size -= size;
1667                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1668                 BUG_ON(ret);
1669         }
1670         btrfs_mark_buffer_dirty(leaf);
1671         return 0;
1672 }
1673
1674 static noinline_for_stack
1675 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1676                                  struct btrfs_root *root,
1677                                  struct btrfs_path *path,
1678                                  u64 bytenr, u64 num_bytes, u64 parent,
1679                                  u64 root_objectid, u64 owner,
1680                                  u64 offset, int refs_to_add,
1681                                  struct btrfs_delayed_extent_op *extent_op)
1682 {
1683         struct btrfs_extent_inline_ref *iref;
1684         int ret;
1685
1686         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1687                                            bytenr, num_bytes, parent,
1688                                            root_objectid, owner, offset, 1);
1689         if (ret == 0) {
1690                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1691                 ret = update_inline_extent_backref(trans, root, path, iref,
1692                                                    refs_to_add, extent_op);
1693         } else if (ret == -ENOENT) {
1694                 ret = setup_inline_extent_backref(trans, root, path, iref,
1695                                                   parent, root_objectid,
1696                                                   owner, offset, refs_to_add,
1697                                                   extent_op);
1698         }
1699         return ret;
1700 }
1701
1702 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1703                                  struct btrfs_root *root,
1704                                  struct btrfs_path *path,
1705                                  u64 bytenr, u64 parent, u64 root_objectid,
1706                                  u64 owner, u64 offset, int refs_to_add)
1707 {
1708         int ret;
1709         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1710                 BUG_ON(refs_to_add != 1);
1711                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1712                                             parent, root_objectid);
1713         } else {
1714                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1715                                              parent, root_objectid,
1716                                              owner, offset, refs_to_add);
1717         }
1718         return ret;
1719 }
1720
1721 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1722                                  struct btrfs_root *root,
1723                                  struct btrfs_path *path,
1724                                  struct btrfs_extent_inline_ref *iref,
1725                                  int refs_to_drop, int is_data)
1726 {
1727         int ret;
1728
1729         BUG_ON(!is_data && refs_to_drop != 1);
1730         if (iref) {
1731                 ret = update_inline_extent_backref(trans, root, path, iref,
1732                                                    -refs_to_drop, NULL);
1733         } else if (is_data) {
1734                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1735         } else {
1736                 ret = btrfs_del_item(trans, root, path);
1737         }
1738         return ret;
1739 }
1740
1741 static int btrfs_issue_discard(struct block_device *bdev,
1742                                 u64 start, u64 len)
1743 {
1744         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1745 }
1746
1747 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1748                                 u64 num_bytes, u64 *actual_bytes)
1749 {
1750         int ret;
1751         u64 discarded_bytes = 0;
1752         struct btrfs_multi_bio *multi = NULL;
1753
1754
1755         /* Tell the block device(s) that the sectors can be discarded */
1756         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1757                               bytenr, &num_bytes, &multi, 0);
1758         if (!ret) {
1759                 struct btrfs_bio_stripe *stripe = multi->stripes;
1760                 int i;
1761
1762
1763                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1764                         ret = btrfs_issue_discard(stripe->dev->bdev,
1765                                                   stripe->physical,
1766                                                   stripe->length);
1767                         if (!ret)
1768                                 discarded_bytes += stripe->length;
1769                         else if (ret != -EOPNOTSUPP)
1770                                 break;
1771                 }
1772                 kfree(multi);
1773         }
1774         if (discarded_bytes && ret == -EOPNOTSUPP)
1775                 ret = 0;
1776
1777         if (actual_bytes)
1778                 *actual_bytes = discarded_bytes;
1779
1780
1781         return ret;
1782 }
1783
1784 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1785                          struct btrfs_root *root,
1786                          u64 bytenr, u64 num_bytes, u64 parent,
1787                          u64 root_objectid, u64 owner, u64 offset)
1788 {
1789         int ret;
1790         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1791                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1792
1793         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1794                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1795                                         parent, root_objectid, (int)owner,
1796                                         BTRFS_ADD_DELAYED_REF, NULL);
1797         } else {
1798                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1799                                         parent, root_objectid, owner, offset,
1800                                         BTRFS_ADD_DELAYED_REF, NULL);
1801         }
1802         return ret;
1803 }
1804
1805 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1806                                   struct btrfs_root *root,
1807                                   u64 bytenr, u64 num_bytes,
1808                                   u64 parent, u64 root_objectid,
1809                                   u64 owner, u64 offset, int refs_to_add,
1810                                   struct btrfs_delayed_extent_op *extent_op)
1811 {
1812         struct btrfs_path *path;
1813         struct extent_buffer *leaf;
1814         struct btrfs_extent_item *item;
1815         u64 refs;
1816         int ret;
1817         int err = 0;
1818
1819         path = btrfs_alloc_path();
1820         if (!path)
1821                 return -ENOMEM;
1822
1823         path->reada = 1;
1824         path->leave_spinning = 1;
1825         /* this will setup the path even if it fails to insert the back ref */
1826         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1827                                            path, bytenr, num_bytes, parent,
1828                                            root_objectid, owner, offset,
1829                                            refs_to_add, extent_op);
1830         if (ret == 0)
1831                 goto out;
1832
1833         if (ret != -EAGAIN) {
1834                 err = ret;
1835                 goto out;
1836         }
1837
1838         leaf = path->nodes[0];
1839         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1840         refs = btrfs_extent_refs(leaf, item);
1841         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1842         if (extent_op)
1843                 __run_delayed_extent_op(extent_op, leaf, item);
1844
1845         btrfs_mark_buffer_dirty(leaf);
1846         btrfs_release_path(root->fs_info->extent_root, path);
1847
1848         path->reada = 1;
1849         path->leave_spinning = 1;
1850
1851         /* now insert the actual backref */
1852         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1853                                     path, bytenr, parent, root_objectid,
1854                                     owner, offset, refs_to_add);
1855         BUG_ON(ret);
1856 out:
1857         btrfs_free_path(path);
1858         return err;
1859 }
1860
1861 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1862                                 struct btrfs_root *root,
1863                                 struct btrfs_delayed_ref_node *node,
1864                                 struct btrfs_delayed_extent_op *extent_op,
1865                                 int insert_reserved)
1866 {
1867         int ret = 0;
1868         struct btrfs_delayed_data_ref *ref;
1869         struct btrfs_key ins;
1870         u64 parent = 0;
1871         u64 ref_root = 0;
1872         u64 flags = 0;
1873
1874         ins.objectid = node->bytenr;
1875         ins.offset = node->num_bytes;
1876         ins.type = BTRFS_EXTENT_ITEM_KEY;
1877
1878         ref = btrfs_delayed_node_to_data_ref(node);
1879         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1880                 parent = ref->parent;
1881         else
1882                 ref_root = ref->root;
1883
1884         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1885                 if (extent_op) {
1886                         BUG_ON(extent_op->update_key);
1887                         flags |= extent_op->flags_to_set;
1888                 }
1889                 ret = alloc_reserved_file_extent(trans, root,
1890                                                  parent, ref_root, flags,
1891                                                  ref->objectid, ref->offset,
1892                                                  &ins, node->ref_mod);
1893         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1894                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1895                                              node->num_bytes, parent,
1896                                              ref_root, ref->objectid,
1897                                              ref->offset, node->ref_mod,
1898                                              extent_op);
1899         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1900                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1901                                           node->num_bytes, parent,
1902                                           ref_root, ref->objectid,
1903                                           ref->offset, node->ref_mod,
1904                                           extent_op);
1905         } else {
1906                 BUG();
1907         }
1908         return ret;
1909 }
1910
1911 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1912                                     struct extent_buffer *leaf,
1913                                     struct btrfs_extent_item *ei)
1914 {
1915         u64 flags = btrfs_extent_flags(leaf, ei);
1916         if (extent_op->update_flags) {
1917                 flags |= extent_op->flags_to_set;
1918                 btrfs_set_extent_flags(leaf, ei, flags);
1919         }
1920
1921         if (extent_op->update_key) {
1922                 struct btrfs_tree_block_info *bi;
1923                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1924                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1925                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1926         }
1927 }
1928
1929 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1930                                  struct btrfs_root *root,
1931                                  struct btrfs_delayed_ref_node *node,
1932                                  struct btrfs_delayed_extent_op *extent_op)
1933 {
1934         struct btrfs_key key;
1935         struct btrfs_path *path;
1936         struct btrfs_extent_item *ei;
1937         struct extent_buffer *leaf;
1938         u32 item_size;
1939         int ret;
1940         int err = 0;
1941
1942         path = btrfs_alloc_path();
1943         if (!path)
1944                 return -ENOMEM;
1945
1946         key.objectid = node->bytenr;
1947         key.type = BTRFS_EXTENT_ITEM_KEY;
1948         key.offset = node->num_bytes;
1949
1950         path->reada = 1;
1951         path->leave_spinning = 1;
1952         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1953                                 path, 0, 1);
1954         if (ret < 0) {
1955                 err = ret;
1956                 goto out;
1957         }
1958         if (ret > 0) {
1959                 err = -EIO;
1960                 goto out;
1961         }
1962
1963         leaf = path->nodes[0];
1964         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1965 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1966         if (item_size < sizeof(*ei)) {
1967                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1968                                              path, (u64)-1, 0);
1969                 if (ret < 0) {
1970                         err = ret;
1971                         goto out;
1972                 }
1973                 leaf = path->nodes[0];
1974                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1975         }
1976 #endif
1977         BUG_ON(item_size < sizeof(*ei));
1978         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1979         __run_delayed_extent_op(extent_op, leaf, ei);
1980
1981         btrfs_mark_buffer_dirty(leaf);
1982 out:
1983         btrfs_free_path(path);
1984         return err;
1985 }
1986
1987 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1988                                 struct btrfs_root *root,
1989                                 struct btrfs_delayed_ref_node *node,
1990                                 struct btrfs_delayed_extent_op *extent_op,
1991                                 int insert_reserved)
1992 {
1993         int ret = 0;
1994         struct btrfs_delayed_tree_ref *ref;
1995         struct btrfs_key ins;
1996         u64 parent = 0;
1997         u64 ref_root = 0;
1998
1999         ins.objectid = node->bytenr;
2000         ins.offset = node->num_bytes;
2001         ins.type = BTRFS_EXTENT_ITEM_KEY;
2002
2003         ref = btrfs_delayed_node_to_tree_ref(node);
2004         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2005                 parent = ref->parent;
2006         else
2007                 ref_root = ref->root;
2008
2009         BUG_ON(node->ref_mod != 1);
2010         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2011                 BUG_ON(!extent_op || !extent_op->update_flags ||
2012                        !extent_op->update_key);
2013                 ret = alloc_reserved_tree_block(trans, root,
2014                                                 parent, ref_root,
2015                                                 extent_op->flags_to_set,
2016                                                 &extent_op->key,
2017                                                 ref->level, &ins);
2018         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2019                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2020                                              node->num_bytes, parent, ref_root,
2021                                              ref->level, 0, 1, extent_op);
2022         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2023                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2024                                           node->num_bytes, parent, ref_root,
2025                                           ref->level, 0, 1, extent_op);
2026         } else {
2027                 BUG();
2028         }
2029         return ret;
2030 }
2031
2032 /* helper function to actually process a single delayed ref entry */
2033 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2034                                struct btrfs_root *root,
2035                                struct btrfs_delayed_ref_node *node,
2036                                struct btrfs_delayed_extent_op *extent_op,
2037                                int insert_reserved)
2038 {
2039         int ret;
2040         if (btrfs_delayed_ref_is_head(node)) {
2041                 struct btrfs_delayed_ref_head *head;
2042                 /*
2043                  * we've hit the end of the chain and we were supposed
2044                  * to insert this extent into the tree.  But, it got
2045                  * deleted before we ever needed to insert it, so all
2046                  * we have to do is clean up the accounting
2047                  */
2048                 BUG_ON(extent_op);
2049                 head = btrfs_delayed_node_to_head(node);
2050                 if (insert_reserved) {
2051                         btrfs_pin_extent(root, node->bytenr,
2052                                          node->num_bytes, 1);
2053                         if (head->is_data) {
2054                                 ret = btrfs_del_csums(trans, root,
2055                                                       node->bytenr,
2056                                                       node->num_bytes);
2057                                 BUG_ON(ret);
2058                         }
2059                 }
2060                 mutex_unlock(&head->mutex);
2061                 return 0;
2062         }
2063
2064         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2065             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2066                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2067                                            insert_reserved);
2068         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2069                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2070                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2071                                            insert_reserved);
2072         else
2073                 BUG();
2074         return ret;
2075 }
2076
2077 static noinline struct btrfs_delayed_ref_node *
2078 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2079 {
2080         struct rb_node *node;
2081         struct btrfs_delayed_ref_node *ref;
2082         int action = BTRFS_ADD_DELAYED_REF;
2083 again:
2084         /*
2085          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2086          * this prevents ref count from going down to zero when
2087          * there still are pending delayed ref.
2088          */
2089         node = rb_prev(&head->node.rb_node);
2090         while (1) {
2091                 if (!node)
2092                         break;
2093                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2094                                 rb_node);
2095                 if (ref->bytenr != head->node.bytenr)
2096                         break;
2097                 if (ref->action == action)
2098                         return ref;
2099                 node = rb_prev(node);
2100         }
2101         if (action == BTRFS_ADD_DELAYED_REF) {
2102                 action = BTRFS_DROP_DELAYED_REF;
2103                 goto again;
2104         }
2105         return NULL;
2106 }
2107
2108 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2109                                        struct btrfs_root *root,
2110                                        struct list_head *cluster)
2111 {
2112         struct btrfs_delayed_ref_root *delayed_refs;
2113         struct btrfs_delayed_ref_node *ref;
2114         struct btrfs_delayed_ref_head *locked_ref = NULL;
2115         struct btrfs_delayed_extent_op *extent_op;
2116         int ret;
2117         int count = 0;
2118         int must_insert_reserved = 0;
2119
2120         delayed_refs = &trans->transaction->delayed_refs;
2121         while (1) {
2122                 if (!locked_ref) {
2123                         /* pick a new head ref from the cluster list */
2124                         if (list_empty(cluster))
2125                                 break;
2126
2127                         locked_ref = list_entry(cluster->next,
2128                                      struct btrfs_delayed_ref_head, cluster);
2129
2130                         /* grab the lock that says we are going to process
2131                          * all the refs for this head */
2132                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2133
2134                         /*
2135                          * we may have dropped the spin lock to get the head
2136                          * mutex lock, and that might have given someone else
2137                          * time to free the head.  If that's true, it has been
2138                          * removed from our list and we can move on.
2139                          */
2140                         if (ret == -EAGAIN) {
2141                                 locked_ref = NULL;
2142                                 count++;
2143                                 continue;
2144                         }
2145                 }
2146
2147                 /*
2148                  * record the must insert reserved flag before we
2149                  * drop the spin lock.
2150                  */
2151                 must_insert_reserved = locked_ref->must_insert_reserved;
2152                 locked_ref->must_insert_reserved = 0;
2153
2154                 extent_op = locked_ref->extent_op;
2155                 locked_ref->extent_op = NULL;
2156
2157                 /*
2158                  * locked_ref is the head node, so we have to go one
2159                  * node back for any delayed ref updates
2160                  */
2161                 ref = select_delayed_ref(locked_ref);
2162                 if (!ref) {
2163                         /* All delayed refs have been processed, Go ahead
2164                          * and send the head node to run_one_delayed_ref,
2165                          * so that any accounting fixes can happen
2166                          */
2167                         ref = &locked_ref->node;
2168
2169                         if (extent_op && must_insert_reserved) {
2170                                 kfree(extent_op);
2171                                 extent_op = NULL;
2172                         }
2173
2174                         if (extent_op) {
2175                                 spin_unlock(&delayed_refs->lock);
2176
2177                                 ret = run_delayed_extent_op(trans, root,
2178                                                             ref, extent_op);
2179                                 BUG_ON(ret);
2180                                 kfree(extent_op);
2181
2182                                 cond_resched();
2183                                 spin_lock(&delayed_refs->lock);
2184                                 continue;
2185                         }
2186
2187                         list_del_init(&locked_ref->cluster);
2188                         locked_ref = NULL;
2189                 }
2190
2191                 ref->in_tree = 0;
2192                 rb_erase(&ref->rb_node, &delayed_refs->root);
2193                 delayed_refs->num_entries--;
2194
2195                 spin_unlock(&delayed_refs->lock);
2196
2197                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2198                                           must_insert_reserved);
2199                 BUG_ON(ret);
2200
2201                 btrfs_put_delayed_ref(ref);
2202                 kfree(extent_op);
2203                 count++;
2204
2205                 cond_resched();
2206                 spin_lock(&delayed_refs->lock);
2207         }
2208         return count;
2209 }
2210
2211 /*
2212  * this starts processing the delayed reference count updates and
2213  * extent insertions we have queued up so far.  count can be
2214  * 0, which means to process everything in the tree at the start
2215  * of the run (but not newly added entries), or it can be some target
2216  * number you'd like to process.
2217  */
2218 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2219                            struct btrfs_root *root, unsigned long count)
2220 {
2221         struct rb_node *node;
2222         struct btrfs_delayed_ref_root *delayed_refs;
2223         struct btrfs_delayed_ref_node *ref;
2224         struct list_head cluster;
2225         int ret;
2226         int run_all = count == (unsigned long)-1;
2227         int run_most = 0;
2228
2229         if (root == root->fs_info->extent_root)
2230                 root = root->fs_info->tree_root;
2231
2232         delayed_refs = &trans->transaction->delayed_refs;
2233         INIT_LIST_HEAD(&cluster);
2234 again:
2235         spin_lock(&delayed_refs->lock);
2236         if (count == 0) {
2237                 count = delayed_refs->num_entries * 2;
2238                 run_most = 1;
2239         }
2240         while (1) {
2241                 if (!(run_all || run_most) &&
2242                     delayed_refs->num_heads_ready < 64)
2243                         break;
2244
2245                 /*
2246                  * go find something we can process in the rbtree.  We start at
2247                  * the beginning of the tree, and then build a cluster
2248                  * of refs to process starting at the first one we are able to
2249                  * lock
2250                  */
2251                 ret = btrfs_find_ref_cluster(trans, &cluster,
2252                                              delayed_refs->run_delayed_start);
2253                 if (ret)
2254                         break;
2255
2256                 ret = run_clustered_refs(trans, root, &cluster);
2257                 BUG_ON(ret < 0);
2258
2259                 count -= min_t(unsigned long, ret, count);
2260
2261                 if (count == 0)
2262                         break;
2263         }
2264
2265         if (run_all) {
2266                 node = rb_first(&delayed_refs->root);
2267                 if (!node)
2268                         goto out;
2269                 count = (unsigned long)-1;
2270
2271                 while (node) {
2272                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2273                                        rb_node);
2274                         if (btrfs_delayed_ref_is_head(ref)) {
2275                                 struct btrfs_delayed_ref_head *head;
2276
2277                                 head = btrfs_delayed_node_to_head(ref);
2278                                 atomic_inc(&ref->refs);
2279
2280                                 spin_unlock(&delayed_refs->lock);
2281                                 mutex_lock(&head->mutex);
2282                                 mutex_unlock(&head->mutex);
2283
2284                                 btrfs_put_delayed_ref(ref);
2285                                 cond_resched();
2286                                 goto again;
2287                         }
2288                         node = rb_next(node);
2289                 }
2290                 spin_unlock(&delayed_refs->lock);
2291                 schedule_timeout(1);
2292                 goto again;
2293         }
2294 out:
2295         spin_unlock(&delayed_refs->lock);
2296         return 0;
2297 }
2298
2299 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2300                                 struct btrfs_root *root,
2301                                 u64 bytenr, u64 num_bytes, u64 flags,
2302                                 int is_data)
2303 {
2304         struct btrfs_delayed_extent_op *extent_op;
2305         int ret;
2306
2307         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2308         if (!extent_op)
2309                 return -ENOMEM;
2310
2311         extent_op->flags_to_set = flags;
2312         extent_op->update_flags = 1;
2313         extent_op->update_key = 0;
2314         extent_op->is_data = is_data ? 1 : 0;
2315
2316         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2317         if (ret)
2318                 kfree(extent_op);
2319         return ret;
2320 }
2321
2322 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2323                                       struct btrfs_root *root,
2324                                       struct btrfs_path *path,
2325                                       u64 objectid, u64 offset, u64 bytenr)
2326 {
2327         struct btrfs_delayed_ref_head *head;
2328         struct btrfs_delayed_ref_node *ref;
2329         struct btrfs_delayed_data_ref *data_ref;
2330         struct btrfs_delayed_ref_root *delayed_refs;
2331         struct rb_node *node;
2332         int ret = 0;
2333
2334         ret = -ENOENT;
2335         delayed_refs = &trans->transaction->delayed_refs;
2336         spin_lock(&delayed_refs->lock);
2337         head = btrfs_find_delayed_ref_head(trans, bytenr);
2338         if (!head)
2339                 goto out;
2340
2341         if (!mutex_trylock(&head->mutex)) {
2342                 atomic_inc(&head->node.refs);
2343                 spin_unlock(&delayed_refs->lock);
2344
2345                 btrfs_release_path(root->fs_info->extent_root, path);
2346
2347                 mutex_lock(&head->mutex);
2348                 mutex_unlock(&head->mutex);
2349                 btrfs_put_delayed_ref(&head->node);
2350                 return -EAGAIN;
2351         }
2352
2353         node = rb_prev(&head->node.rb_node);
2354         if (!node)
2355                 goto out_unlock;
2356
2357         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2358
2359         if (ref->bytenr != bytenr)
2360                 goto out_unlock;
2361
2362         ret = 1;
2363         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2364                 goto out_unlock;
2365
2366         data_ref = btrfs_delayed_node_to_data_ref(ref);
2367
2368         node = rb_prev(node);
2369         if (node) {
2370                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2371                 if (ref->bytenr == bytenr)
2372                         goto out_unlock;
2373         }
2374
2375         if (data_ref->root != root->root_key.objectid ||
2376             data_ref->objectid != objectid || data_ref->offset != offset)
2377                 goto out_unlock;
2378
2379         ret = 0;
2380 out_unlock:
2381         mutex_unlock(&head->mutex);
2382 out:
2383         spin_unlock(&delayed_refs->lock);
2384         return ret;
2385 }
2386
2387 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2388                                         struct btrfs_root *root,
2389                                         struct btrfs_path *path,
2390                                         u64 objectid, u64 offset, u64 bytenr)
2391 {
2392         struct btrfs_root *extent_root = root->fs_info->extent_root;
2393         struct extent_buffer *leaf;
2394         struct btrfs_extent_data_ref *ref;
2395         struct btrfs_extent_inline_ref *iref;
2396         struct btrfs_extent_item *ei;
2397         struct btrfs_key key;
2398         u32 item_size;
2399         int ret;
2400
2401         key.objectid = bytenr;
2402         key.offset = (u64)-1;
2403         key.type = BTRFS_EXTENT_ITEM_KEY;
2404
2405         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2406         if (ret < 0)
2407                 goto out;
2408         BUG_ON(ret == 0);
2409
2410         ret = -ENOENT;
2411         if (path->slots[0] == 0)
2412                 goto out;
2413
2414         path->slots[0]--;
2415         leaf = path->nodes[0];
2416         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2417
2418         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2419                 goto out;
2420
2421         ret = 1;
2422         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2423 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2424         if (item_size < sizeof(*ei)) {
2425                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2426                 goto out;
2427         }
2428 #endif
2429         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2430
2431         if (item_size != sizeof(*ei) +
2432             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2433                 goto out;
2434
2435         if (btrfs_extent_generation(leaf, ei) <=
2436             btrfs_root_last_snapshot(&root->root_item))
2437                 goto out;
2438
2439         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2440         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2441             BTRFS_EXTENT_DATA_REF_KEY)
2442                 goto out;
2443
2444         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2445         if (btrfs_extent_refs(leaf, ei) !=
2446             btrfs_extent_data_ref_count(leaf, ref) ||
2447             btrfs_extent_data_ref_root(leaf, ref) !=
2448             root->root_key.objectid ||
2449             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2450             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2451                 goto out;
2452
2453         ret = 0;
2454 out:
2455         return ret;
2456 }
2457
2458 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2459                           struct btrfs_root *root,
2460                           u64 objectid, u64 offset, u64 bytenr)
2461 {
2462         struct btrfs_path *path;
2463         int ret;
2464         int ret2;
2465
2466         path = btrfs_alloc_path();
2467         if (!path)
2468                 return -ENOENT;
2469
2470         do {
2471                 ret = check_committed_ref(trans, root, path, objectid,
2472                                           offset, bytenr);
2473                 if (ret && ret != -ENOENT)
2474                         goto out;
2475
2476                 ret2 = check_delayed_ref(trans, root, path, objectid,
2477                                          offset, bytenr);
2478         } while (ret2 == -EAGAIN);
2479
2480         if (ret2 && ret2 != -ENOENT) {
2481                 ret = ret2;
2482                 goto out;
2483         }
2484
2485         if (ret != -ENOENT || ret2 != -ENOENT)
2486                 ret = 0;
2487 out:
2488         btrfs_free_path(path);
2489         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2490                 WARN_ON(ret > 0);
2491         return ret;
2492 }
2493
2494 #if 0
2495 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2496                     struct extent_buffer *buf, u32 nr_extents)
2497 {
2498         struct btrfs_key key;
2499         struct btrfs_file_extent_item *fi;
2500         u64 root_gen;
2501         u32 nritems;
2502         int i;
2503         int level;
2504         int ret = 0;
2505         int shared = 0;
2506
2507         if (!root->ref_cows)
2508                 return 0;
2509
2510         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2511                 shared = 0;
2512                 root_gen = root->root_key.offset;
2513         } else {
2514                 shared = 1;
2515                 root_gen = trans->transid - 1;
2516         }
2517
2518         level = btrfs_header_level(buf);
2519         nritems = btrfs_header_nritems(buf);
2520
2521         if (level == 0) {
2522                 struct btrfs_leaf_ref *ref;
2523                 struct btrfs_extent_info *info;
2524
2525                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2526                 if (!ref) {
2527                         ret = -ENOMEM;
2528                         goto out;
2529                 }
2530
2531                 ref->root_gen = root_gen;
2532                 ref->bytenr = buf->start;
2533                 ref->owner = btrfs_header_owner(buf);
2534                 ref->generation = btrfs_header_generation(buf);
2535                 ref->nritems = nr_extents;
2536                 info = ref->extents;
2537
2538                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2539                         u64 disk_bytenr;
2540                         btrfs_item_key_to_cpu(buf, &key, i);
2541                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2542                                 continue;
2543                         fi = btrfs_item_ptr(buf, i,
2544                                             struct btrfs_file_extent_item);
2545                         if (btrfs_file_extent_type(buf, fi) ==
2546                             BTRFS_FILE_EXTENT_INLINE)
2547                                 continue;
2548                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2549                         if (disk_bytenr == 0)
2550                                 continue;
2551
2552                         info->bytenr = disk_bytenr;
2553                         info->num_bytes =
2554                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2555                         info->objectid = key.objectid;
2556                         info->offset = key.offset;
2557                         info++;
2558                 }
2559
2560                 ret = btrfs_add_leaf_ref(root, ref, shared);
2561                 if (ret == -EEXIST && shared) {
2562                         struct btrfs_leaf_ref *old;
2563                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2564                         BUG_ON(!old);
2565                         btrfs_remove_leaf_ref(root, old);
2566                         btrfs_free_leaf_ref(root, old);
2567                         ret = btrfs_add_leaf_ref(root, ref, shared);
2568                 }
2569                 WARN_ON(ret);
2570                 btrfs_free_leaf_ref(root, ref);
2571         }
2572 out:
2573         return ret;
2574 }
2575
2576 /* when a block goes through cow, we update the reference counts of
2577  * everything that block points to.  The internal pointers of the block
2578  * can be in just about any order, and it is likely to have clusters of
2579  * things that are close together and clusters of things that are not.
2580  *
2581  * To help reduce the seeks that come with updating all of these reference
2582  * counts, sort them by byte number before actual updates are done.
2583  *
2584  * struct refsort is used to match byte number to slot in the btree block.
2585  * we sort based on the byte number and then use the slot to actually
2586  * find the item.
2587  *
2588  * struct refsort is smaller than strcut btrfs_item and smaller than
2589  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2590  * for a btree block, there's no way for a kmalloc of refsorts for a
2591  * single node to be bigger than a page.
2592  */
2593 struct refsort {
2594         u64 bytenr;
2595         u32 slot;
2596 };
2597
2598 /*
2599  * for passing into sort()
2600  */
2601 static int refsort_cmp(const void *a_void, const void *b_void)
2602 {
2603         const struct refsort *a = a_void;
2604         const struct refsort *b = b_void;
2605
2606         if (a->bytenr < b->bytenr)
2607                 return -1;
2608         if (a->bytenr > b->bytenr)
2609                 return 1;
2610         return 0;
2611 }
2612 #endif
2613
2614 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2615                            struct btrfs_root *root,
2616                            struct extent_buffer *buf,
2617                            int full_backref, int inc)
2618 {
2619         u64 bytenr;
2620         u64 num_bytes;
2621         u64 parent;
2622         u64 ref_root;
2623         u32 nritems;
2624         struct btrfs_key key;
2625         struct btrfs_file_extent_item *fi;
2626         int i;
2627         int level;
2628         int ret = 0;
2629         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2630                             u64, u64, u64, u64, u64, u64);
2631
2632         ref_root = btrfs_header_owner(buf);
2633         nritems = btrfs_header_nritems(buf);
2634         level = btrfs_header_level(buf);
2635
2636         if (!root->ref_cows && level == 0)
2637                 return 0;
2638
2639         if (inc)
2640                 process_func = btrfs_inc_extent_ref;
2641         else
2642                 process_func = btrfs_free_extent;
2643
2644         if (full_backref)
2645                 parent = buf->start;
2646         else
2647                 parent = 0;
2648
2649         for (i = 0; i < nritems; i++) {
2650                 if (level == 0) {
2651                         btrfs_item_key_to_cpu(buf, &key, i);
2652                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2653                                 continue;
2654                         fi = btrfs_item_ptr(buf, i,
2655                                             struct btrfs_file_extent_item);
2656                         if (btrfs_file_extent_type(buf, fi) ==
2657                             BTRFS_FILE_EXTENT_INLINE)
2658                                 continue;
2659                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2660                         if (bytenr == 0)
2661                                 continue;
2662
2663                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2664                         key.offset -= btrfs_file_extent_offset(buf, fi);
2665                         ret = process_func(trans, root, bytenr, num_bytes,
2666                                            parent, ref_root, key.objectid,
2667                                            key.offset);
2668                         if (ret)
2669                                 goto fail;
2670                 } else {
2671                         bytenr = btrfs_node_blockptr(buf, i);
2672                         num_bytes = btrfs_level_size(root, level - 1);
2673                         ret = process_func(trans, root, bytenr, num_bytes,
2674                                            parent, ref_root, level - 1, 0);
2675                         if (ret)
2676                                 goto fail;
2677                 }
2678         }
2679         return 0;
2680 fail:
2681         BUG();
2682         return ret;
2683 }
2684
2685 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2686                   struct extent_buffer *buf, int full_backref)
2687 {
2688         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2689 }
2690
2691 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2692                   struct extent_buffer *buf, int full_backref)
2693 {
2694         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2695 }
2696
2697 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2698                                  struct btrfs_root *root,
2699                                  struct btrfs_path *path,
2700                                  struct btrfs_block_group_cache *cache)
2701 {
2702         int ret;
2703         struct btrfs_root *extent_root = root->fs_info->extent_root;
2704         unsigned long bi;
2705         struct extent_buffer *leaf;
2706
2707         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2708         if (ret < 0)
2709                 goto fail;
2710         BUG_ON(ret);
2711
2712         leaf = path->nodes[0];
2713         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2714         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2715         btrfs_mark_buffer_dirty(leaf);
2716         btrfs_release_path(extent_root, path);
2717 fail:
2718         if (ret)
2719                 return ret;
2720         return 0;
2721
2722 }
2723
2724 static struct btrfs_block_group_cache *
2725 next_block_group(struct btrfs_root *root,
2726                  struct btrfs_block_group_cache *cache)
2727 {
2728         struct rb_node *node;
2729         spin_lock(&root->fs_info->block_group_cache_lock);
2730         node = rb_next(&cache->cache_node);
2731         btrfs_put_block_group(cache);
2732         if (node) {
2733                 cache = rb_entry(node, struct btrfs_block_group_cache,
2734                                  cache_node);
2735                 btrfs_get_block_group(cache);
2736         } else
2737                 cache = NULL;
2738         spin_unlock(&root->fs_info->block_group_cache_lock);
2739         return cache;
2740 }
2741
2742 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2743                             struct btrfs_trans_handle *trans,
2744                             struct btrfs_path *path)
2745 {
2746         struct btrfs_root *root = block_group->fs_info->tree_root;
2747         struct inode *inode = NULL;
2748         u64 alloc_hint = 0;
2749         int dcs = BTRFS_DC_ERROR;
2750         int num_pages = 0;
2751         int retries = 0;
2752         int ret = 0;
2753
2754         /*
2755          * If this block group is smaller than 100 megs don't bother caching the
2756          * block group.
2757          */
2758         if (block_group->key.offset < (100 * 1024 * 1024)) {
2759                 spin_lock(&block_group->lock);
2760                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2761                 spin_unlock(&block_group->lock);
2762                 return 0;
2763         }
2764
2765 again:
2766         inode = lookup_free_space_inode(root, block_group, path);
2767         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2768                 ret = PTR_ERR(inode);
2769                 btrfs_release_path(root, path);
2770                 goto out;
2771         }
2772
2773         if (IS_ERR(inode)) {
2774                 BUG_ON(retries);
2775                 retries++;
2776
2777                 if (block_group->ro)
2778                         goto out_free;
2779
2780                 ret = create_free_space_inode(root, trans, block_group, path);
2781                 if (ret)
2782                         goto out_free;
2783                 goto again;
2784         }
2785
2786         /*
2787          * We want to set the generation to 0, that way if anything goes wrong
2788          * from here on out we know not to trust this cache when we load up next
2789          * time.
2790          */
2791         BTRFS_I(inode)->generation = 0;
2792         ret = btrfs_update_inode(trans, root, inode);
2793         WARN_ON(ret);
2794
2795         if (i_size_read(inode) > 0) {
2796                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2797                                                       inode);
2798                 if (ret)
2799                         goto out_put;
2800         }
2801
2802         spin_lock(&block_group->lock);
2803         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2804                 /* We're not cached, don't bother trying to write stuff out */
2805                 dcs = BTRFS_DC_WRITTEN;
2806                 spin_unlock(&block_group->lock);
2807                 goto out_put;
2808         }
2809         spin_unlock(&block_group->lock);
2810
2811         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2812         if (!num_pages)
2813                 num_pages = 1;
2814
2815         /*
2816          * Just to make absolutely sure we have enough space, we're going to
2817          * preallocate 12 pages worth of space for each block group.  In
2818          * practice we ought to use at most 8, but we need extra space so we can
2819          * add our header and have a terminator between the extents and the
2820          * bitmaps.
2821          */
2822         num_pages *= 16;
2823         num_pages *= PAGE_CACHE_SIZE;
2824
2825         ret = btrfs_check_data_free_space(inode, num_pages);
2826         if (ret)
2827                 goto out_put;
2828
2829         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2830                                               num_pages, num_pages,
2831                                               &alloc_hint);
2832         if (!ret)
2833                 dcs = BTRFS_DC_SETUP;
2834         btrfs_free_reserved_data_space(inode, num_pages);
2835 out_put:
2836         iput(inode);
2837 out_free:
2838         btrfs_release_path(root, path);
2839 out:
2840         spin_lock(&block_group->lock);
2841         block_group->disk_cache_state = dcs;
2842         spin_unlock(&block_group->lock);
2843
2844         return ret;
2845 }
2846
2847 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2848                                    struct btrfs_root *root)
2849 {
2850         struct btrfs_block_group_cache *cache;
2851         int err = 0;
2852         struct btrfs_path *path;
2853         u64 last = 0;
2854
2855         path = btrfs_alloc_path();
2856         if (!path)
2857                 return -ENOMEM;
2858
2859 again:
2860         while (1) {
2861                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2862                 while (cache) {
2863                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2864                                 break;
2865                         cache = next_block_group(root, cache);
2866                 }
2867                 if (!cache) {
2868                         if (last == 0)
2869                                 break;
2870                         last = 0;
2871                         continue;
2872                 }
2873                 err = cache_save_setup(cache, trans, path);
2874                 last = cache->key.objectid + cache->key.offset;
2875                 btrfs_put_block_group(cache);
2876         }
2877
2878         while (1) {
2879                 if (last == 0) {
2880                         err = btrfs_run_delayed_refs(trans, root,
2881                                                      (unsigned long)-1);
2882                         BUG_ON(err);
2883                 }
2884
2885                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2886                 while (cache) {
2887                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2888                                 btrfs_put_block_group(cache);
2889                                 goto again;
2890                         }
2891
2892                         if (cache->dirty)
2893                                 break;
2894                         cache = next_block_group(root, cache);
2895                 }
2896                 if (!cache) {
2897                         if (last == 0)
2898                                 break;
2899                         last = 0;
2900                         continue;
2901                 }
2902
2903                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2904                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2905                 cache->dirty = 0;
2906                 last = cache->key.objectid + cache->key.offset;
2907
2908                 err = write_one_cache_group(trans, root, path, cache);
2909                 BUG_ON(err);
2910                 btrfs_put_block_group(cache);
2911         }
2912
2913         while (1) {
2914                 /*
2915                  * I don't think this is needed since we're just marking our
2916                  * preallocated extent as written, but just in case it can't
2917                  * hurt.
2918                  */
2919                 if (last == 0) {
2920                         err = btrfs_run_delayed_refs(trans, root,
2921                                                      (unsigned long)-1);
2922                         BUG_ON(err);
2923                 }
2924
2925                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2926                 while (cache) {
2927                         /*
2928                          * Really this shouldn't happen, but it could if we
2929                          * couldn't write the entire preallocated extent and
2930                          * splitting the extent resulted in a new block.
2931                          */
2932                         if (cache->dirty) {
2933                                 btrfs_put_block_group(cache);
2934                                 goto again;
2935                         }
2936                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2937                                 break;
2938                         cache = next_block_group(root, cache);
2939                 }
2940                 if (!cache) {
2941                         if (last == 0)
2942                                 break;
2943                         last = 0;
2944                         continue;
2945                 }
2946
2947                 btrfs_write_out_cache(root, trans, cache, path);
2948
2949                 /*
2950                  * If we didn't have an error then the cache state is still
2951                  * NEED_WRITE, so we can set it to WRITTEN.
2952                  */
2953                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2954                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2955                 last = cache->key.objectid + cache->key.offset;
2956                 btrfs_put_block_group(cache);
2957         }
2958
2959         btrfs_free_path(path);
2960         return 0;
2961 }
2962
2963 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2964 {
2965         struct btrfs_block_group_cache *block_group;
2966         int readonly = 0;
2967
2968         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2969         if (!block_group || block_group->ro)
2970                 readonly = 1;
2971         if (block_group)
2972                 btrfs_put_block_group(block_group);
2973         return readonly;
2974 }
2975
2976 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2977                              u64 total_bytes, u64 bytes_used,
2978                              struct btrfs_space_info **space_info)
2979 {
2980         struct btrfs_space_info *found;
2981         int i;
2982         int factor;
2983
2984         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2985                      BTRFS_BLOCK_GROUP_RAID10))
2986                 factor = 2;
2987         else
2988                 factor = 1;
2989
2990         found = __find_space_info(info, flags);
2991         if (found) {
2992                 spin_lock(&found->lock);
2993                 found->total_bytes += total_bytes;
2994                 found->disk_total += total_bytes * factor;
2995                 found->bytes_used += bytes_used;
2996                 found->disk_used += bytes_used * factor;
2997                 found->full = 0;
2998                 spin_unlock(&found->lock);
2999                 *space_info = found;
3000                 return 0;
3001         }
3002         found = kzalloc(sizeof(*found), GFP_NOFS);
3003         if (!found)
3004                 return -ENOMEM;
3005
3006         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3007                 INIT_LIST_HEAD(&found->block_groups[i]);
3008         init_rwsem(&found->groups_sem);
3009         spin_lock_init(&found->lock);
3010         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3011                                 BTRFS_BLOCK_GROUP_SYSTEM |
3012                                 BTRFS_BLOCK_GROUP_METADATA);
3013         found->total_bytes = total_bytes;
3014         found->disk_total = total_bytes * factor;
3015         found->bytes_used = bytes_used;
3016         found->disk_used = bytes_used * factor;
3017         found->bytes_pinned = 0;
3018         found->bytes_reserved = 0;
3019         found->bytes_readonly = 0;
3020         found->bytes_may_use = 0;
3021         found->full = 0;
3022         found->force_alloc = 0;
3023         *space_info = found;
3024         list_add_rcu(&found->list, &info->space_info);
3025         atomic_set(&found->caching_threads, 0);
3026         return 0;
3027 }
3028
3029 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3030 {
3031         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3032                                    BTRFS_BLOCK_GROUP_RAID1 |
3033                                    BTRFS_BLOCK_GROUP_RAID10 |
3034                                    BTRFS_BLOCK_GROUP_DUP);
3035         if (extra_flags) {
3036                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3037                         fs_info->avail_data_alloc_bits |= extra_flags;
3038                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3039                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3040                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3041                         fs_info->avail_system_alloc_bits |= extra_flags;
3042         }
3043 }
3044
3045 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3046 {
3047         /*
3048          * we add in the count of missing devices because we want
3049          * to make sure that any RAID levels on a degraded FS
3050          * continue to be honored.
3051          */
3052         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3053                 root->fs_info->fs_devices->missing_devices;
3054
3055         if (num_devices == 1)
3056                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3057         if (num_devices < 4)
3058                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3059
3060         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3061             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3062                       BTRFS_BLOCK_GROUP_RAID10))) {
3063                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3064         }
3065
3066         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3067             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3068                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3069         }
3070
3071         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3072             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3073              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3074              (flags & BTRFS_BLOCK_GROUP_DUP)))
3075                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3076         return flags;
3077 }
3078
3079 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3080 {
3081         if (flags & BTRFS_BLOCK_GROUP_DATA)
3082                 flags |= root->fs_info->avail_data_alloc_bits &
3083                          root->fs_info->data_alloc_profile;
3084         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3085                 flags |= root->fs_info->avail_system_alloc_bits &
3086                          root->fs_info->system_alloc_profile;
3087         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3088                 flags |= root->fs_info->avail_metadata_alloc_bits &
3089                          root->fs_info->metadata_alloc_profile;
3090         return btrfs_reduce_alloc_profile(root, flags);
3091 }
3092
3093 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3094 {
3095         u64 flags;
3096
3097         if (data)
3098                 flags = BTRFS_BLOCK_GROUP_DATA;
3099         else if (root == root->fs_info->chunk_root)
3100                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3101         else
3102                 flags = BTRFS_BLOCK_GROUP_METADATA;
3103
3104         return get_alloc_profile(root, flags);
3105 }
3106
3107 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3108 {
3109         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3110                                                        BTRFS_BLOCK_GROUP_DATA);
3111 }
3112
3113 /*
3114  * This will check the space that the inode allocates from to make sure we have
3115  * enough space for bytes.
3116  */
3117 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3118 {
3119         struct btrfs_space_info *data_sinfo;
3120         struct btrfs_root *root = BTRFS_I(inode)->root;
3121         u64 used;
3122         int ret = 0, committed = 0, alloc_chunk = 1;
3123
3124         /* make sure bytes are sectorsize aligned */
3125         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3126
3127         if (root == root->fs_info->tree_root) {
3128                 alloc_chunk = 0;
3129                 committed = 1;
3130         }
3131
3132         data_sinfo = BTRFS_I(inode)->space_info;
3133         if (!data_sinfo)
3134                 goto alloc;
3135
3136 again:
3137         /* make sure we have enough space to handle the data first */
3138         spin_lock(&data_sinfo->lock);
3139         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3140                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3141                 data_sinfo->bytes_may_use;
3142
3143         if (used + bytes > data_sinfo->total_bytes) {
3144                 struct btrfs_trans_handle *trans;
3145
3146                 /*
3147                  * if we don't have enough free bytes in this space then we need
3148                  * to alloc a new chunk.
3149                  */
3150                 if (!data_sinfo->full && alloc_chunk) {
3151                         u64 alloc_target;
3152
3153                         data_sinfo->force_alloc = 1;
3154                         spin_unlock(&data_sinfo->lock);
3155 alloc:
3156                         alloc_target = btrfs_get_alloc_profile(root, 1);
3157                         trans = btrfs_join_transaction(root, 1);
3158                         if (IS_ERR(trans))
3159                                 return PTR_ERR(trans);
3160
3161                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3162                                              bytes + 2 * 1024 * 1024,
3163                                              alloc_target, 0);
3164                         btrfs_end_transaction(trans, root);
3165                         if (ret < 0) {
3166                                 if (ret != -ENOSPC)
3167                                         return ret;
3168                                 else
3169                                         goto commit_trans;
3170                         }
3171
3172                         if (!data_sinfo) {
3173                                 btrfs_set_inode_space_info(root, inode);
3174                                 data_sinfo = BTRFS_I(inode)->space_info;
3175                         }
3176                         goto again;
3177                 }
3178                 spin_unlock(&data_sinfo->lock);
3179
3180                 /* commit the current transaction and try again */
3181 commit_trans:
3182                 if (!committed && !root->fs_info->open_ioctl_trans) {
3183                         committed = 1;
3184                         trans = btrfs_join_transaction(root, 1);
3185                         if (IS_ERR(trans))
3186                                 return PTR_ERR(trans);
3187                         ret = btrfs_commit_transaction(trans, root);
3188                         if (ret)
3189                                 return ret;
3190                         goto again;
3191                 }
3192
3193 #if 0 /* I hope we never need this code again, just in case */
3194                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3195                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3196                        "%llu bytes_readonly, %llu may use %llu total\n",
3197                        (unsigned long long)bytes,
3198                        (unsigned long long)data_sinfo->bytes_used,
3199                        (unsigned long long)data_sinfo->bytes_reserved,
3200                        (unsigned long long)data_sinfo->bytes_pinned,
3201                        (unsigned long long)data_sinfo->bytes_readonly,
3202                        (unsigned long long)data_sinfo->bytes_may_use,
3203                        (unsigned long long)data_sinfo->total_bytes);
3204 #endif
3205                 return -ENOSPC;
3206         }
3207         data_sinfo->bytes_may_use += bytes;
3208         BTRFS_I(inode)->reserved_bytes += bytes;
3209         spin_unlock(&data_sinfo->lock);
3210
3211         return 0;
3212 }
3213
3214 /*
3215  * called when we are clearing an delalloc extent from the
3216  * inode's io_tree or there was an error for whatever reason
3217  * after calling btrfs_check_data_free_space
3218  */
3219 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3220 {
3221         struct btrfs_root *root = BTRFS_I(inode)->root;
3222         struct btrfs_space_info *data_sinfo;
3223
3224         /* make sure bytes are sectorsize aligned */
3225         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3226
3227         data_sinfo = BTRFS_I(inode)->space_info;
3228         spin_lock(&data_sinfo->lock);
3229         data_sinfo->bytes_may_use -= bytes;
3230         BTRFS_I(inode)->reserved_bytes -= bytes;
3231         spin_unlock(&data_sinfo->lock);
3232 }
3233
3234 static void force_metadata_allocation(struct btrfs_fs_info *info)
3235 {
3236         struct list_head *head = &info->space_info;
3237         struct btrfs_space_info *found;
3238
3239         rcu_read_lock();
3240         list_for_each_entry_rcu(found, head, list) {
3241                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3242                         found->force_alloc = 1;
3243         }
3244         rcu_read_unlock();
3245 }
3246
3247 static int should_alloc_chunk(struct btrfs_root *root,
3248                               struct btrfs_space_info *sinfo, u64 alloc_bytes)
3249 {
3250         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3251         u64 thresh;
3252
3253         if (sinfo->bytes_used + sinfo->bytes_reserved +
3254             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3255                 return 0;
3256
3257         if (sinfo->bytes_used + sinfo->bytes_reserved +
3258             alloc_bytes < div_factor(num_bytes, 8))
3259                 return 0;
3260
3261         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3262         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3263
3264         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3265                 return 0;
3266
3267         return 1;
3268 }
3269
3270 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3271                           struct btrfs_root *extent_root, u64 alloc_bytes,
3272                           u64 flags, int force)
3273 {
3274         struct btrfs_space_info *space_info;
3275         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3276         int ret = 0;
3277
3278         mutex_lock(&fs_info->chunk_mutex);
3279
3280         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3281
3282         space_info = __find_space_info(extent_root->fs_info, flags);
3283         if (!space_info) {
3284                 ret = update_space_info(extent_root->fs_info, flags,
3285                                         0, 0, &space_info);
3286                 BUG_ON(ret);
3287         }
3288         BUG_ON(!space_info);
3289
3290         spin_lock(&space_info->lock);
3291         if (space_info->force_alloc)
3292                 force = 1;
3293         if (space_info->full) {
3294                 spin_unlock(&space_info->lock);
3295                 goto out;
3296         }
3297
3298         if (!force && !should_alloc_chunk(extent_root, space_info,
3299                                           alloc_bytes)) {
3300                 spin_unlock(&space_info->lock);
3301                 goto out;
3302         }
3303         spin_unlock(&space_info->lock);
3304
3305         /*
3306          * If we have mixed data/metadata chunks we want to make sure we keep
3307          * allocating mixed chunks instead of individual chunks.
3308          */
3309         if (btrfs_mixed_space_info(space_info))
3310                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3311
3312         /*
3313          * if we're doing a data chunk, go ahead and make sure that
3314          * we keep a reasonable number of metadata chunks allocated in the
3315          * FS as well.
3316          */
3317         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3318                 fs_info->data_chunk_allocations++;
3319                 if (!(fs_info->data_chunk_allocations %
3320                       fs_info->metadata_ratio))
3321                         force_metadata_allocation(fs_info);
3322         }
3323
3324         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3325         spin_lock(&space_info->lock);
3326         if (ret)
3327                 space_info->full = 1;
3328         else
3329                 ret = 1;
3330         space_info->force_alloc = 0;
3331         spin_unlock(&space_info->lock);
3332 out:
3333         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3334         return ret;
3335 }
3336
3337 /*
3338  * shrink metadata reservation for delalloc
3339  */
3340 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3341                            struct btrfs_root *root, u64 to_reclaim, int sync)
3342 {
3343         struct btrfs_block_rsv *block_rsv;
3344         struct btrfs_space_info *space_info;
3345         u64 reserved;
3346         u64 max_reclaim;
3347         u64 reclaimed = 0;
3348         long time_left;
3349         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3350         int loops = 0;
3351         unsigned long progress;
3352
3353         block_rsv = &root->fs_info->delalloc_block_rsv;
3354         space_info = block_rsv->space_info;
3355
3356         smp_mb();
3357         reserved = space_info->bytes_reserved;
3358         progress = space_info->reservation_progress;
3359
3360         if (reserved == 0)
3361                 return 0;
3362
3363         max_reclaim = min(reserved, to_reclaim);
3364
3365         while (loops < 1024) {
3366                 /* have the flusher threads jump in and do some IO */
3367                 smp_mb();
3368                 nr_pages = min_t(unsigned long, nr_pages,
3369                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3370                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3371
3372                 spin_lock(&space_info->lock);
3373                 if (reserved > space_info->bytes_reserved)
3374                         reclaimed += reserved - space_info->bytes_reserved;
3375                 reserved = space_info->bytes_reserved;
3376                 spin_unlock(&space_info->lock);
3377
3378                 loops++;
3379
3380                 if (reserved == 0 || reclaimed >= max_reclaim)
3381                         break;
3382
3383                 if (trans && trans->transaction->blocked)
3384                         return -EAGAIN;
3385
3386                 time_left = schedule_timeout_interruptible(1);
3387
3388                 /* We were interrupted, exit */
3389                 if (time_left)
3390                         break;
3391
3392                 /* we've kicked the IO a few times, if anything has been freed,
3393                  * exit.  There is no sense in looping here for a long time
3394                  * when we really need to commit the transaction, or there are
3395                  * just too many writers without enough free space
3396                  */
3397
3398                 if (loops > 3) {
3399                         smp_mb();
3400                         if (progress != space_info->reservation_progress)
3401                                 break;
3402                 }
3403
3404         }
3405         return reclaimed >= to_reclaim;
3406 }
3407
3408 /*
3409  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3410  * idea is if this is the first call (retries == 0) then we will add to our
3411  * reserved count if we can't make the allocation in order to hold our place
3412  * while we go and try and free up space.  That way for retries > 1 we don't try
3413  * and add space, we just check to see if the amount of unused space is >= the
3414  * total space, meaning that our reservation is valid.
3415  *
3416  * However if we don't intend to retry this reservation, pass -1 as retries so
3417  * that it short circuits this logic.
3418  */
3419 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3420                                   struct btrfs_root *root,
3421                                   struct btrfs_block_rsv *block_rsv,
3422                                   u64 orig_bytes, int flush)
3423 {
3424         struct btrfs_space_info *space_info = block_rsv->space_info;
3425         u64 unused;
3426         u64 num_bytes = orig_bytes;
3427         int retries = 0;
3428         int ret = 0;
3429         bool reserved = false;
3430         bool committed = false;
3431
3432 again:
3433         ret = -ENOSPC;
3434         if (reserved)
3435                 num_bytes = 0;
3436
3437         spin_lock(&space_info->lock);
3438         unused = space_info->bytes_used + space_info->bytes_reserved +
3439                  space_info->bytes_pinned + space_info->bytes_readonly +
3440                  space_info->bytes_may_use;
3441
3442         /*
3443          * The idea here is that we've not already over-reserved the block group
3444          * then we can go ahead and save our reservation first and then start
3445          * flushing if we need to.  Otherwise if we've already overcommitted
3446          * lets start flushing stuff first and then come back and try to make
3447          * our reservation.
3448          */
3449         if (unused <= space_info->total_bytes) {
3450                 unused = space_info->total_bytes - unused;
3451                 if (unused >= num_bytes) {
3452                         if (!reserved)
3453                                 space_info->bytes_reserved += orig_bytes;
3454                         ret = 0;
3455                 } else {
3456                         /*
3457                          * Ok set num_bytes to orig_bytes since we aren't
3458                          * overocmmitted, this way we only try and reclaim what
3459                          * we need.
3460                          */
3461                         num_bytes = orig_bytes;
3462                 }
3463         } else {
3464                 /*
3465                  * Ok we're over committed, set num_bytes to the overcommitted
3466                  * amount plus the amount of bytes that we need for this
3467                  * reservation.
3468                  */
3469                 num_bytes = unused - space_info->total_bytes +
3470                         (orig_bytes * (retries + 1));
3471         }
3472
3473         /*
3474          * Couldn't make our reservation, save our place so while we're trying
3475          * to reclaim space we can actually use it instead of somebody else
3476          * stealing it from us.
3477          */
3478         if (ret && !reserved) {
3479                 space_info->bytes_reserved += orig_bytes;
3480                 reserved = true;
3481         }
3482
3483         spin_unlock(&space_info->lock);
3484
3485         if (!ret)
3486                 return 0;
3487
3488         if (!flush)
3489                 goto out;
3490
3491         /*
3492          * We do synchronous shrinking since we don't actually unreserve
3493          * metadata until after the IO is completed.
3494          */
3495         ret = shrink_delalloc(trans, root, num_bytes, 1);
3496         if (ret > 0)
3497                 return 0;
3498         else if (ret < 0)
3499                 goto out;
3500
3501         /*
3502          * So if we were overcommitted it's possible that somebody else flushed
3503          * out enough space and we simply didn't have enough space to reclaim,
3504          * so go back around and try again.
3505          */
3506         if (retries < 2) {
3507                 retries++;
3508                 goto again;
3509         }
3510
3511         spin_lock(&space_info->lock);
3512         /*
3513          * Not enough space to be reclaimed, don't bother committing the
3514          * transaction.
3515          */
3516         if (space_info->bytes_pinned < orig_bytes)
3517                 ret = -ENOSPC;
3518         spin_unlock(&space_info->lock);
3519         if (ret)
3520                 goto out;
3521
3522         ret = -EAGAIN;
3523         if (trans || committed)
3524                 goto out;
3525
3526         ret = -ENOSPC;
3527         trans = btrfs_join_transaction(root, 1);
3528         if (IS_ERR(trans))
3529                 goto out;
3530         ret = btrfs_commit_transaction(trans, root);
3531         if (!ret) {
3532                 trans = NULL;
3533                 committed = true;
3534                 goto again;
3535         }
3536
3537 out:
3538         if (reserved) {
3539                 spin_lock(&space_info->lock);
3540                 space_info->bytes_reserved -= orig_bytes;
3541                 spin_unlock(&space_info->lock);
3542         }
3543
3544         return ret;
3545 }
3546
3547 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3548                                              struct btrfs_root *root)
3549 {
3550         struct btrfs_block_rsv *block_rsv;
3551         if (root->ref_cows)
3552                 block_rsv = trans->block_rsv;
3553         else
3554                 block_rsv = root->block_rsv;
3555
3556         if (!block_rsv)
3557                 block_rsv = &root->fs_info->empty_block_rsv;
3558
3559         return block_rsv;
3560 }
3561
3562 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3563                                u64 num_bytes)
3564 {
3565         int ret = -ENOSPC;
3566         spin_lock(&block_rsv->lock);
3567         if (block_rsv->reserved >= num_bytes) {
3568                 block_rsv->reserved -= num_bytes;
3569                 if (block_rsv->reserved < block_rsv->size)
3570                         block_rsv->full = 0;
3571                 ret = 0;
3572         }
3573         spin_unlock(&block_rsv->lock);
3574         return ret;
3575 }
3576
3577 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3578                                 u64 num_bytes, int update_size)
3579 {
3580         spin_lock(&block_rsv->lock);
3581         block_rsv->reserved += num_bytes;
3582         if (update_size)
3583                 block_rsv->size += num_bytes;
3584         else if (block_rsv->reserved >= block_rsv->size)
3585                 block_rsv->full = 1;
3586         spin_unlock(&block_rsv->lock);
3587 }
3588
3589 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3590                              struct btrfs_block_rsv *dest, u64 num_bytes)
3591 {
3592         struct btrfs_space_info *space_info = block_rsv->space_info;
3593
3594         spin_lock(&block_rsv->lock);
3595         if (num_bytes == (u64)-1)
3596                 num_bytes = block_rsv->size;
3597         block_rsv->size -= num_bytes;
3598         if (block_rsv->reserved >= block_rsv->size) {
3599                 num_bytes = block_rsv->reserved - block_rsv->size;
3600                 block_rsv->reserved = block_rsv->size;
3601                 block_rsv->full = 1;
3602         } else {
3603                 num_bytes = 0;
3604         }
3605         spin_unlock(&block_rsv->lock);
3606
3607         if (num_bytes > 0) {
3608                 if (dest) {
3609                         spin_lock(&dest->lock);
3610                         if (!dest->full) {
3611                                 u64 bytes_to_add;
3612
3613                                 bytes_to_add = dest->size - dest->reserved;
3614                                 bytes_to_add = min(num_bytes, bytes_to_add);
3615                                 dest->reserved += bytes_to_add;
3616                                 if (dest->reserved >= dest->size)
3617                                         dest->full = 1;
3618                                 num_bytes -= bytes_to_add;
3619                         }
3620                         spin_unlock(&dest->lock);
3621                 }
3622                 if (num_bytes) {
3623                         spin_lock(&space_info->lock);
3624                         space_info->bytes_reserved -= num_bytes;
3625                         space_info->reservation_progress++;
3626                         spin_unlock(&space_info->lock);
3627                 }
3628         }
3629 }
3630
3631 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3632                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3633 {
3634         int ret;
3635
3636         ret = block_rsv_use_bytes(src, num_bytes);
3637         if (ret)
3638                 return ret;
3639
3640         block_rsv_add_bytes(dst, num_bytes, 1);
3641         return 0;
3642 }
3643
3644 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3645 {
3646         memset(rsv, 0, sizeof(*rsv));
3647         spin_lock_init(&rsv->lock);
3648         atomic_set(&rsv->usage, 1);
3649         rsv->priority = 6;
3650         INIT_LIST_HEAD(&rsv->list);
3651 }
3652
3653 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3654 {
3655         struct btrfs_block_rsv *block_rsv;
3656         struct btrfs_fs_info *fs_info = root->fs_info;
3657
3658         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3659         if (!block_rsv)
3660                 return NULL;
3661
3662         btrfs_init_block_rsv(block_rsv);
3663         block_rsv->space_info = __find_space_info(fs_info,
3664                                                   BTRFS_BLOCK_GROUP_METADATA);
3665         return block_rsv;
3666 }
3667
3668 void btrfs_free_block_rsv(struct btrfs_root *root,
3669                           struct btrfs_block_rsv *rsv)
3670 {
3671         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3672                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3673                 if (!rsv->durable)
3674                         kfree(rsv);
3675         }
3676 }
3677
3678 /*
3679  * make the block_rsv struct be able to capture freed space.
3680  * the captured space will re-add to the the block_rsv struct
3681  * after transaction commit
3682  */
3683 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3684                                  struct btrfs_block_rsv *block_rsv)
3685 {
3686         block_rsv->durable = 1;
3687         mutex_lock(&fs_info->durable_block_rsv_mutex);
3688         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3689         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3690 }
3691
3692 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3693                         struct btrfs_root *root,
3694                         struct btrfs_block_rsv *block_rsv,
3695                         u64 num_bytes)
3696 {
3697         int ret;
3698
3699         if (num_bytes == 0)
3700                 return 0;
3701
3702         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3703         if (!ret) {
3704                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3705                 return 0;
3706         }
3707
3708         return ret;
3709 }
3710
3711 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3712                           struct btrfs_root *root,
3713                           struct btrfs_block_rsv *block_rsv,
3714                           u64 min_reserved, int min_factor)
3715 {
3716         u64 num_bytes = 0;
3717         int commit_trans = 0;
3718         int ret = -ENOSPC;
3719
3720         if (!block_rsv)
3721                 return 0;
3722
3723         spin_lock(&block_rsv->lock);
3724         if (min_factor > 0)
3725                 num_bytes = div_factor(block_rsv->size, min_factor);
3726         if (min_reserved > num_bytes)
3727                 num_bytes = min_reserved;
3728
3729         if (block_rsv->reserved >= num_bytes) {
3730                 ret = 0;
3731         } else {
3732                 num_bytes -= block_rsv->reserved;
3733                 if (block_rsv->durable &&
3734                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3735                         commit_trans = 1;
3736         }
3737         spin_unlock(&block_rsv->lock);
3738         if (!ret)
3739                 return 0;
3740
3741         if (block_rsv->refill_used) {
3742                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3743                                              num_bytes, 0);
3744                 if (!ret) {
3745                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3746                         return 0;
3747                 }
3748         }
3749
3750         if (commit_trans) {
3751                 if (trans)
3752                         return -EAGAIN;
3753
3754                 trans = btrfs_join_transaction(root, 1);
3755                 BUG_ON(IS_ERR(trans));
3756                 ret = btrfs_commit_transaction(trans, root);
3757                 return 0;
3758         }
3759
3760         return -ENOSPC;
3761 }
3762
3763 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3764                             struct btrfs_block_rsv *dst_rsv,
3765                             u64 num_bytes)
3766 {
3767         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3768 }
3769
3770 void btrfs_block_rsv_release(struct btrfs_root *root,
3771                              struct btrfs_block_rsv *block_rsv,
3772                              u64 num_bytes)
3773 {
3774         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3775         if (global_rsv->full || global_rsv == block_rsv ||
3776             block_rsv->space_info != global_rsv->space_info)
3777                 global_rsv = NULL;
3778         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3779 }
3780
3781 /*
3782  * helper to calculate size of global block reservation.
3783  * the desired value is sum of space used by extent tree,
3784  * checksum tree and root tree
3785  */
3786 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3787 {
3788         struct btrfs_space_info *sinfo;
3789         u64 num_bytes;
3790         u64 meta_used;
3791         u64 data_used;
3792         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3793 #if 0
3794         /*
3795          * per tree used space accounting can be inaccuracy, so we
3796          * can't rely on it.
3797          */
3798         spin_lock(&fs_info->extent_root->accounting_lock);
3799         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3800         spin_unlock(&fs_info->extent_root->accounting_lock);
3801
3802         spin_lock(&fs_info->csum_root->accounting_lock);
3803         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3804         spin_unlock(&fs_info->csum_root->accounting_lock);
3805
3806         spin_lock(&fs_info->tree_root->accounting_lock);
3807         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3808         spin_unlock(&fs_info->tree_root->accounting_lock);
3809 #endif
3810         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3811         spin_lock(&sinfo->lock);
3812         data_used = sinfo->bytes_used;
3813         spin_unlock(&sinfo->lock);
3814
3815         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3816         spin_lock(&sinfo->lock);
3817         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3818                 data_used = 0;
3819         meta_used = sinfo->bytes_used;
3820         spin_unlock(&sinfo->lock);
3821
3822         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3823                     csum_size * 2;
3824         num_bytes += div64_u64(data_used + meta_used, 50);
3825
3826         if (num_bytes * 3 > meta_used)
3827                 num_bytes = div64_u64(meta_used, 3);
3828
3829         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3830 }
3831
3832 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3833 {
3834         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3835         struct btrfs_space_info *sinfo = block_rsv->space_info;
3836         u64 num_bytes;
3837
3838         num_bytes = calc_global_metadata_size(fs_info);
3839
3840         spin_lock(&block_rsv->lock);
3841         spin_lock(&sinfo->lock);
3842
3843         block_rsv->size = num_bytes;
3844
3845         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3846                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3847                     sinfo->bytes_may_use;
3848
3849         if (sinfo->total_bytes > num_bytes) {
3850                 num_bytes = sinfo->total_bytes - num_bytes;
3851                 block_rsv->reserved += num_bytes;
3852                 sinfo->bytes_reserved += num_bytes;
3853         }
3854
3855         if (block_rsv->reserved >= block_rsv->size) {
3856                 num_bytes = block_rsv->reserved - block_rsv->size;
3857                 sinfo->bytes_reserved -= num_bytes;
3858                 sinfo->reservation_progress++;
3859                 block_rsv->reserved = block_rsv->size;
3860                 block_rsv->full = 1;
3861         }
3862 #if 0
3863         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3864                 block_rsv->size, block_rsv->reserved);
3865 #endif
3866         spin_unlock(&sinfo->lock);
3867         spin_unlock(&block_rsv->lock);
3868 }
3869
3870 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3871 {
3872         struct btrfs_space_info *space_info;
3873
3874         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3875         fs_info->chunk_block_rsv.space_info = space_info;
3876         fs_info->chunk_block_rsv.priority = 10;
3877
3878         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3879         fs_info->global_block_rsv.space_info = space_info;
3880         fs_info->global_block_rsv.priority = 10;
3881         fs_info->global_block_rsv.refill_used = 1;
3882         fs_info->delalloc_block_rsv.space_info = space_info;
3883         fs_info->trans_block_rsv.space_info = space_info;
3884         fs_info->empty_block_rsv.space_info = space_info;
3885         fs_info->empty_block_rsv.priority = 10;
3886
3887         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3888         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3889         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3890         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3891         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3892
3893         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3894
3895         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3896
3897         update_global_block_rsv(fs_info);
3898 }
3899
3900 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3901 {
3902         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3903         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3904         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3905         WARN_ON(fs_info->trans_block_rsv.size > 0);
3906         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3907         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3908         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3909 }
3910
3911 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3912 {
3913         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3914                 3 * num_items;
3915 }
3916
3917 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3918                                  struct btrfs_root *root,
3919                                  int num_items)
3920 {
3921         u64 num_bytes;
3922         int ret;
3923
3924         if (num_items == 0 || root->fs_info->chunk_root == root)
3925                 return 0;
3926
3927         num_bytes = calc_trans_metadata_size(root, num_items);
3928         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3929                                   num_bytes);
3930         if (!ret) {
3931                 trans->bytes_reserved += num_bytes;
3932                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3933         }
3934         return ret;
3935 }
3936
3937 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3938                                   struct btrfs_root *root)
3939 {
3940         if (!trans->bytes_reserved)
3941                 return;
3942
3943         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3944         btrfs_block_rsv_release(root, trans->block_rsv,
3945                                 trans->bytes_reserved);
3946         trans->bytes_reserved = 0;
3947 }
3948
3949 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3950                                   struct inode *inode)
3951 {
3952         struct btrfs_root *root = BTRFS_I(inode)->root;
3953         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3954         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3955
3956         /*
3957          * one for deleting orphan item, one for updating inode and
3958          * two for calling btrfs_truncate_inode_items.
3959          *
3960          * btrfs_truncate_inode_items is a delete operation, it frees
3961          * more space than it uses in most cases. So two units of
3962          * metadata space should be enough for calling it many times.
3963          * If all of the metadata space is used, we can commit
3964          * transaction and use space it freed.
3965          */
3966         u64 num_bytes = calc_trans_metadata_size(root, 4);
3967         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3968 }
3969
3970 void btrfs_orphan_release_metadata(struct inode *inode)
3971 {
3972         struct btrfs_root *root = BTRFS_I(inode)->root;
3973         u64 num_bytes = calc_trans_metadata_size(root, 4);
3974         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3975 }
3976
3977 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3978                                 struct btrfs_pending_snapshot *pending)
3979 {
3980         struct btrfs_root *root = pending->root;
3981         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3982         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3983         /*
3984          * two for root back/forward refs, two for directory entries
3985          * and one for root of the snapshot.
3986          */
3987         u64 num_bytes = calc_trans_metadata_size(root, 5);
3988         dst_rsv->space_info = src_rsv->space_info;
3989         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3990 }
3991
3992 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3993 {
3994         return num_bytes >>= 3;
3995 }
3996
3997 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3998 {
3999         struct btrfs_root *root = BTRFS_I(inode)->root;
4000         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4001         u64 to_reserve;
4002         int nr_extents;
4003         int reserved_extents;
4004         int ret;
4005
4006         if (btrfs_transaction_in_commit(root->fs_info))
4007                 schedule_timeout(1);
4008
4009         num_bytes = ALIGN(num_bytes, root->sectorsize);
4010
4011         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4012         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4013
4014         if (nr_extents > reserved_extents) {
4015                 nr_extents -= reserved_extents;
4016                 to_reserve = calc_trans_metadata_size(root, nr_extents);
4017         } else {
4018                 nr_extents = 0;
4019                 to_reserve = 0;
4020         }
4021
4022         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4023         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4024         if (ret)
4025                 return ret;
4026
4027         atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
4028         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4029
4030         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4031
4032         if (block_rsv->size > 512 * 1024 * 1024)
4033                 shrink_delalloc(NULL, root, to_reserve, 0);
4034
4035         return 0;
4036 }
4037
4038 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4039 {
4040         struct btrfs_root *root = BTRFS_I(inode)->root;
4041         u64 to_free;
4042         int nr_extents;
4043         int reserved_extents;
4044
4045         num_bytes = ALIGN(num_bytes, root->sectorsize);
4046         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4047         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4048
4049         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4050         do {
4051                 int old, new;
4052
4053                 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4054                 if (nr_extents >= reserved_extents) {
4055                         nr_extents = 0;
4056                         break;
4057                 }
4058                 old = reserved_extents;
4059                 nr_extents = reserved_extents - nr_extents;
4060                 new = reserved_extents - nr_extents;
4061                 old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
4062                                      reserved_extents, new);
4063                 if (likely(old == reserved_extents))
4064                         break;
4065                 reserved_extents = old;
4066         } while (1);
4067
4068         to_free = calc_csum_metadata_size(inode, num_bytes);
4069         if (nr_extents > 0)
4070                 to_free += calc_trans_metadata_size(root, nr_extents);
4071
4072         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4073                                 to_free);
4074 }
4075
4076 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4077 {
4078         int ret;
4079
4080         ret = btrfs_check_data_free_space(inode, num_bytes);
4081         if (ret)
4082                 return ret;
4083
4084         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4085         if (ret) {
4086                 btrfs_free_reserved_data_space(inode, num_bytes);
4087                 return ret;
4088         }
4089
4090         return 0;
4091 }
4092
4093 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4094 {
4095         btrfs_delalloc_release_metadata(inode, num_bytes);
4096         btrfs_free_reserved_data_space(inode, num_bytes);
4097 }
4098
4099 static int update_block_group(struct btrfs_trans_handle *trans,
4100                               struct btrfs_root *root,
4101                               u64 bytenr, u64 num_bytes, int alloc)
4102 {
4103         struct btrfs_block_group_cache *cache = NULL;
4104         struct btrfs_fs_info *info = root->fs_info;
4105         u64 total = num_bytes;
4106         u64 old_val;
4107         u64 byte_in_group;
4108         int factor;
4109
4110         /* block accounting for super block */
4111         spin_lock(&info->delalloc_lock);
4112         old_val = btrfs_super_bytes_used(&info->super_copy);
4113         if (alloc)
4114                 old_val += num_bytes;
4115         else
4116                 old_val -= num_bytes;
4117         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4118         spin_unlock(&info->delalloc_lock);
4119
4120         while (total) {
4121                 cache = btrfs_lookup_block_group(info, bytenr);
4122                 if (!cache)
4123                         return -1;
4124                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4125                                     BTRFS_BLOCK_GROUP_RAID1 |
4126                                     BTRFS_BLOCK_GROUP_RAID10))
4127                         factor = 2;
4128                 else
4129                         factor = 1;
4130                 /*
4131                  * If this block group has free space cache written out, we
4132                  * need to make sure to load it if we are removing space.  This
4133                  * is because we need the unpinning stage to actually add the
4134                  * space back to the block group, otherwise we will leak space.
4135                  */
4136                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4137                         cache_block_group(cache, trans, NULL, 1);
4138
4139                 byte_in_group = bytenr - cache->key.objectid;
4140                 WARN_ON(byte_in_group > cache->key.offset);
4141
4142                 spin_lock(&cache->space_info->lock);
4143                 spin_lock(&cache->lock);
4144
4145                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4146                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4147                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4148
4149                 cache->dirty = 1;
4150                 old_val = btrfs_block_group_used(&cache->item);
4151                 num_bytes = min(total, cache->key.offset - byte_in_group);
4152                 if (alloc) {
4153                         old_val += num_bytes;
4154                         btrfs_set_block_group_used(&cache->item, old_val);
4155                         cache->reserved -= num_bytes;
4156                         cache->space_info->bytes_reserved -= num_bytes;
4157                         cache->space_info->reservation_progress++;
4158                         cache->space_info->bytes_used += num_bytes;
4159                         cache->space_info->disk_used += num_bytes * factor;
4160                         spin_unlock(&cache->lock);
4161                         spin_unlock(&cache->space_info->lock);
4162                 } else {
4163                         old_val -= num_bytes;
4164                         btrfs_set_block_group_used(&cache->item, old_val);
4165                         cache->pinned += num_bytes;
4166                         cache->space_info->bytes_pinned += num_bytes;
4167                         cache->space_info->bytes_used -= num_bytes;
4168                         cache->space_info->disk_used -= num_bytes * factor;
4169                         spin_unlock(&cache->lock);
4170                         spin_unlock(&cache->space_info->lock);
4171
4172                         set_extent_dirty(info->pinned_extents,
4173                                          bytenr, bytenr + num_bytes - 1,
4174                                          GFP_NOFS | __GFP_NOFAIL);
4175                 }
4176                 btrfs_put_block_group(cache);
4177                 total -= num_bytes;
4178                 bytenr += num_bytes;
4179         }
4180         return 0;
4181 }
4182
4183 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4184 {
4185         struct btrfs_block_group_cache *cache;
4186         u64 bytenr;
4187
4188         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4189         if (!cache)
4190                 return 0;
4191
4192         bytenr = cache->key.objectid;
4193         btrfs_put_block_group(cache);
4194
4195         return bytenr;
4196 }
4197
4198 static int pin_down_extent(struct btrfs_root *root,
4199                            struct btrfs_block_group_cache *cache,
4200                            u64 bytenr, u64 num_bytes, int reserved)
4201 {
4202         spin_lock(&cache->space_info->lock);
4203         spin_lock(&cache->lock);
4204         cache->pinned += num_bytes;
4205         cache->space_info->bytes_pinned += num_bytes;
4206         if (reserved) {
4207                 cache->reserved -= num_bytes;
4208                 cache->space_info->bytes_reserved -= num_bytes;
4209                 cache->space_info->reservation_progress++;
4210         }
4211         spin_unlock(&cache->lock);
4212         spin_unlock(&cache->space_info->lock);
4213
4214         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4215                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4216         return 0;
4217 }
4218
4219 /*
4220  * this function must be called within transaction
4221  */
4222 int btrfs_pin_extent(struct btrfs_root *root,
4223                      u64 bytenr, u64 num_bytes, int reserved)
4224 {
4225         struct btrfs_block_group_cache *cache;
4226
4227         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4228         BUG_ON(!cache);
4229
4230         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4231
4232         btrfs_put_block_group(cache);
4233         return 0;
4234 }
4235
4236 /*
4237  * update size of reserved extents. this function may return -EAGAIN
4238  * if 'reserve' is true or 'sinfo' is false.
4239  */
4240 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4241                                 u64 num_bytes, int reserve, int sinfo)
4242 {
4243         int ret = 0;
4244         if (sinfo) {
4245                 struct btrfs_space_info *space_info = cache->space_info;
4246                 spin_lock(&space_info->lock);
4247                 spin_lock(&cache->lock);
4248                 if (reserve) {
4249                         if (cache->ro) {
4250                                 ret = -EAGAIN;
4251                         } else {
4252                                 cache->reserved += num_bytes;
4253                                 space_info->bytes_reserved += num_bytes;
4254                         }
4255                 } else {
4256                         if (cache->ro)
4257                                 space_info->bytes_readonly += num_bytes;
4258                         cache->reserved -= num_bytes;
4259                         space_info->bytes_reserved -= num_bytes;
4260                         space_info->reservation_progress++;
4261                 }
4262                 spin_unlock(&cache->lock);
4263                 spin_unlock(&space_info->lock);
4264         } else {
4265                 spin_lock(&cache->lock);
4266                 if (cache->ro) {
4267                         ret = -EAGAIN;
4268                 } else {
4269                         if (reserve)
4270                                 cache->reserved += num_bytes;
4271                         else
4272                                 cache->reserved -= num_bytes;
4273                 }
4274                 spin_unlock(&cache->lock);
4275         }
4276         return ret;
4277 }
4278
4279 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4280                                 struct btrfs_root *root)
4281 {
4282         struct btrfs_fs_info *fs_info = root->fs_info;
4283         struct btrfs_caching_control *next;
4284         struct btrfs_caching_control *caching_ctl;
4285         struct btrfs_block_group_cache *cache;
4286
4287         down_write(&fs_info->extent_commit_sem);
4288
4289         list_for_each_entry_safe(caching_ctl, next,
4290                                  &fs_info->caching_block_groups, list) {
4291                 cache = caching_ctl->block_group;
4292                 if (block_group_cache_done(cache)) {
4293                         cache->last_byte_to_unpin = (u64)-1;
4294                         list_del_init(&caching_ctl->list);
4295                         put_caching_control(caching_ctl);
4296                 } else {
4297                         cache->last_byte_to_unpin = caching_ctl->progress;
4298                 }
4299         }
4300
4301         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4302                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4303         else
4304                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4305
4306         up_write(&fs_info->extent_commit_sem);
4307
4308         update_global_block_rsv(fs_info);
4309         return 0;
4310 }
4311
4312 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4313 {
4314         struct btrfs_fs_info *fs_info = root->fs_info;
4315         struct btrfs_block_group_cache *cache = NULL;
4316         u64 len;
4317
4318         while (start <= end) {
4319                 if (!cache ||
4320                     start >= cache->key.objectid + cache->key.offset) {
4321                         if (cache)
4322                                 btrfs_put_block_group(cache);
4323                         cache = btrfs_lookup_block_group(fs_info, start);
4324                         BUG_ON(!cache);
4325                 }
4326
4327                 len = cache->key.objectid + cache->key.offset - start;
4328                 len = min(len, end + 1 - start);
4329
4330                 if (start < cache->last_byte_to_unpin) {
4331                         len = min(len, cache->last_byte_to_unpin - start);
4332                         btrfs_add_free_space(cache, start, len);
4333                 }
4334
4335                 start += len;
4336
4337                 spin_lock(&cache->space_info->lock);
4338                 spin_lock(&cache->lock);
4339                 cache->pinned -= len;
4340                 cache->space_info->bytes_pinned -= len;
4341                 if (cache->ro) {
4342                         cache->space_info->bytes_readonly += len;
4343                 } else if (cache->reserved_pinned > 0) {
4344                         len = min(len, cache->reserved_pinned);
4345                         cache->reserved_pinned -= len;
4346                         cache->space_info->bytes_reserved += len;
4347                 }
4348                 spin_unlock(&cache->lock);
4349                 spin_unlock(&cache->space_info->lock);
4350         }
4351
4352         if (cache)
4353                 btrfs_put_block_group(cache);
4354         return 0;
4355 }
4356
4357 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4358                                struct btrfs_root *root)
4359 {
4360         struct btrfs_fs_info *fs_info = root->fs_info;
4361         struct extent_io_tree *unpin;
4362         struct btrfs_block_rsv *block_rsv;
4363         struct btrfs_block_rsv *next_rsv;
4364         u64 start;
4365         u64 end;
4366         int idx;
4367         int ret;
4368
4369         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4370                 unpin = &fs_info->freed_extents[1];
4371         else
4372                 unpin = &fs_info->freed_extents[0];
4373
4374         while (1) {
4375                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4376                                             EXTENT_DIRTY);
4377                 if (ret)
4378                         break;
4379
4380                 if (btrfs_test_opt(root, DISCARD))
4381                         ret = btrfs_discard_extent(root, start,
4382                                                    end + 1 - start, NULL);
4383
4384                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4385                 unpin_extent_range(root, start, end);
4386                 cond_resched();
4387         }
4388
4389         mutex_lock(&fs_info->durable_block_rsv_mutex);
4390         list_for_each_entry_safe(block_rsv, next_rsv,
4391                                  &fs_info->durable_block_rsv_list, list) {
4392
4393                 idx = trans->transid & 0x1;
4394                 if (block_rsv->freed[idx] > 0) {
4395                         block_rsv_add_bytes(block_rsv,
4396                                             block_rsv->freed[idx], 0);
4397                         block_rsv->freed[idx] = 0;
4398                 }
4399                 if (atomic_read(&block_rsv->usage) == 0) {
4400                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4401
4402                         if (block_rsv->freed[0] == 0 &&
4403                             block_rsv->freed[1] == 0) {
4404                                 list_del_init(&block_rsv->list);
4405                                 kfree(block_rsv);
4406                         }
4407                 } else {
4408                         btrfs_block_rsv_release(root, block_rsv, 0);
4409                 }
4410         }
4411         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4412
4413         return 0;
4414 }
4415
4416 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4417                                 struct btrfs_root *root,
4418                                 u64 bytenr, u64 num_bytes, u64 parent,
4419                                 u64 root_objectid, u64 owner_objectid,
4420                                 u64 owner_offset, int refs_to_drop,
4421                                 struct btrfs_delayed_extent_op *extent_op)
4422 {
4423         struct btrfs_key key;
4424         struct btrfs_path *path;
4425         struct btrfs_fs_info *info = root->fs_info;
4426         struct btrfs_root *extent_root = info->extent_root;
4427         struct extent_buffer *leaf;
4428         struct btrfs_extent_item *ei;
4429         struct btrfs_extent_inline_ref *iref;
4430         int ret;
4431         int is_data;
4432         int extent_slot = 0;
4433         int found_extent = 0;
4434         int num_to_del = 1;
4435         u32 item_size;
4436         u64 refs;
4437
4438         path = btrfs_alloc_path();
4439         if (!path)
4440                 return -ENOMEM;
4441
4442         path->reada = 1;
4443         path->leave_spinning = 1;
4444
4445         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4446         BUG_ON(!is_data && refs_to_drop != 1);
4447
4448         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4449                                     bytenr, num_bytes, parent,
4450                                     root_objectid, owner_objectid,
4451                                     owner_offset);
4452         if (ret == 0) {
4453                 extent_slot = path->slots[0];
4454                 while (extent_slot >= 0) {
4455                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4456                                               extent_slot);
4457                         if (key.objectid != bytenr)
4458                                 break;
4459                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4460                             key.offset == num_bytes) {
4461                                 found_extent = 1;
4462                                 break;
4463                         }
4464                         if (path->slots[0] - extent_slot > 5)
4465                                 break;
4466                         extent_slot--;
4467                 }
4468 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4469                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4470                 if (found_extent && item_size < sizeof(*ei))
4471                         found_extent = 0;
4472 #endif
4473                 if (!found_extent) {
4474                         BUG_ON(iref);
4475                         ret = remove_extent_backref(trans, extent_root, path,
4476                                                     NULL, refs_to_drop,
4477                                                     is_data);
4478                         BUG_ON(ret);
4479                         btrfs_release_path(extent_root, path);
4480                         path->leave_spinning = 1;
4481
4482                         key.objectid = bytenr;
4483                         key.type = BTRFS_EXTENT_ITEM_KEY;
4484                         key.offset = num_bytes;
4485
4486                         ret = btrfs_search_slot(trans, extent_root,
4487                                                 &key, path, -1, 1);
4488                         if (ret) {
4489                                 printk(KERN_ERR "umm, got %d back from search"
4490                                        ", was looking for %llu\n", ret,
4491                                        (unsigned long long)bytenr);
4492                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4493                         }
4494                         BUG_ON(ret);
4495                         extent_slot = path->slots[0];
4496                 }
4497         } else {
4498                 btrfs_print_leaf(extent_root, path->nodes[0]);
4499                 WARN_ON(1);
4500                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4501                        "parent %llu root %llu  owner %llu offset %llu\n",
4502                        (unsigned long long)bytenr,
4503                        (unsigned long long)parent,
4504                        (unsigned long long)root_objectid,
4505                        (unsigned long long)owner_objectid,
4506                        (unsigned long long)owner_offset);
4507         }
4508
4509         leaf = path->nodes[0];
4510         item_size = btrfs_item_size_nr(leaf, extent_slot);
4511 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4512         if (item_size < sizeof(*ei)) {
4513                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4514                 ret = convert_extent_item_v0(trans, extent_root, path,
4515                                              owner_objectid, 0);
4516                 BUG_ON(ret < 0);
4517
4518                 btrfs_release_path(extent_root, path);
4519                 path->leave_spinning = 1;
4520
4521                 key.objectid = bytenr;
4522                 key.type = BTRFS_EXTENT_ITEM_KEY;
4523                 key.offset = num_bytes;
4524
4525                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4526                                         -1, 1);
4527                 if (ret) {
4528                         printk(KERN_ERR "umm, got %d back from search"
4529                                ", was looking for %llu\n", ret,
4530                                (unsigned long long)bytenr);
4531                         btrfs_print_leaf(extent_root, path->nodes[0]);
4532                 }
4533                 BUG_ON(ret);
4534                 extent_slot = path->slots[0];
4535                 leaf = path->nodes[0];
4536                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4537         }
4538 #endif
4539         BUG_ON(item_size < sizeof(*ei));
4540         ei = btrfs_item_ptr(leaf, extent_slot,
4541                             struct btrfs_extent_item);
4542         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4543                 struct btrfs_tree_block_info *bi;
4544                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4545                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4546                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4547         }
4548
4549         refs = btrfs_extent_refs(leaf, ei);
4550         BUG_ON(refs < refs_to_drop);
4551         refs -= refs_to_drop;
4552
4553         if (refs > 0) {
4554                 if (extent_op)
4555                         __run_delayed_extent_op(extent_op, leaf, ei);
4556                 /*
4557                  * In the case of inline back ref, reference count will
4558                  * be updated by remove_extent_backref
4559                  */
4560                 if (iref) {
4561                         BUG_ON(!found_extent);
4562                 } else {
4563                         btrfs_set_extent_refs(leaf, ei, refs);
4564                         btrfs_mark_buffer_dirty(leaf);
4565                 }
4566                 if (found_extent) {
4567                         ret = remove_extent_backref(trans, extent_root, path,
4568                                                     iref, refs_to_drop,
4569                                                     is_data);
4570                         BUG_ON(ret);
4571                 }
4572         } else {
4573                 if (found_extent) {
4574                         BUG_ON(is_data && refs_to_drop !=
4575                                extent_data_ref_count(root, path, iref));
4576                         if (iref) {
4577                                 BUG_ON(path->slots[0] != extent_slot);
4578                         } else {
4579                                 BUG_ON(path->slots[0] != extent_slot + 1);
4580                                 path->slots[0] = extent_slot;
4581                                 num_to_del = 2;
4582                         }
4583                 }
4584
4585                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4586                                       num_to_del);
4587                 BUG_ON(ret);
4588                 btrfs_release_path(extent_root, path);
4589
4590                 if (is_data) {
4591                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4592                         BUG_ON(ret);
4593                 } else {
4594                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4595                              bytenr >> PAGE_CACHE_SHIFT,
4596                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4597                 }
4598
4599                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4600                 BUG_ON(ret);
4601         }
4602         btrfs_free_path(path);
4603         return ret;
4604 }
4605
4606 /*
4607  * when we free an block, it is possible (and likely) that we free the last
4608  * delayed ref for that extent as well.  This searches the delayed ref tree for
4609  * a given extent, and if there are no other delayed refs to be processed, it
4610  * removes it from the tree.
4611  */
4612 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4613                                       struct btrfs_root *root, u64 bytenr)
4614 {
4615         struct btrfs_delayed_ref_head *head;
4616         struct btrfs_delayed_ref_root *delayed_refs;
4617         struct btrfs_delayed_ref_node *ref;
4618         struct rb_node *node;
4619         int ret = 0;
4620
4621         delayed_refs = &trans->transaction->delayed_refs;
4622         spin_lock(&delayed_refs->lock);
4623         head = btrfs_find_delayed_ref_head(trans, bytenr);
4624         if (!head)
4625                 goto out;
4626
4627         node = rb_prev(&head->node.rb_node);
4628         if (!node)
4629                 goto out;
4630
4631         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4632
4633         /* there are still entries for this ref, we can't drop it */
4634         if (ref->bytenr == bytenr)
4635                 goto out;
4636
4637         if (head->extent_op) {
4638                 if (!head->must_insert_reserved)
4639                         goto out;
4640                 kfree(head->extent_op);
4641                 head->extent_op = NULL;
4642         }
4643
4644         /*
4645          * waiting for the lock here would deadlock.  If someone else has it
4646          * locked they are already in the process of dropping it anyway
4647          */
4648         if (!mutex_trylock(&head->mutex))
4649                 goto out;
4650
4651         /*
4652          * at this point we have a head with no other entries.  Go
4653          * ahead and process it.
4654          */
4655         head->node.in_tree = 0;
4656         rb_erase(&head->node.rb_node, &delayed_refs->root);
4657
4658         delayed_refs->num_entries--;
4659
4660         /*
4661          * we don't take a ref on the node because we're removing it from the
4662          * tree, so we just steal the ref the tree was holding.
4663          */
4664         delayed_refs->num_heads--;
4665         if (list_empty(&head->cluster))
4666                 delayed_refs->num_heads_ready--;
4667
4668         list_del_init(&head->cluster);
4669         spin_unlock(&delayed_refs->lock);
4670
4671         BUG_ON(head->extent_op);
4672         if (head->must_insert_reserved)
4673                 ret = 1;
4674
4675         mutex_unlock(&head->mutex);
4676         btrfs_put_delayed_ref(&head->node);
4677         return ret;
4678 out:
4679         spin_unlock(&delayed_refs->lock);
4680         return 0;
4681 }
4682
4683 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4684                            struct btrfs_root *root,
4685                            struct extent_buffer *buf,
4686                            u64 parent, int last_ref)
4687 {
4688         struct btrfs_block_rsv *block_rsv;
4689         struct btrfs_block_group_cache *cache = NULL;
4690         int ret;
4691
4692         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4693                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4694                                                 parent, root->root_key.objectid,
4695                                                 btrfs_header_level(buf),
4696                                                 BTRFS_DROP_DELAYED_REF, NULL);
4697                 BUG_ON(ret);
4698         }
4699
4700         if (!last_ref)
4701                 return;
4702
4703         block_rsv = get_block_rsv(trans, root);
4704         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4705         if (block_rsv->space_info != cache->space_info)
4706                 goto out;
4707
4708         if (btrfs_header_generation(buf) == trans->transid) {
4709                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4710                         ret = check_ref_cleanup(trans, root, buf->start);
4711                         if (!ret)
4712                                 goto pin;
4713                 }
4714
4715                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4716                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4717                         goto pin;
4718                 }
4719
4720                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4721
4722                 btrfs_add_free_space(cache, buf->start, buf->len);
4723                 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
4724                 if (ret == -EAGAIN) {
4725                         /* block group became read-only */
4726                         btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
4727                         goto out;
4728                 }
4729
4730                 ret = 1;
4731                 spin_lock(&block_rsv->lock);
4732                 if (block_rsv->reserved < block_rsv->size) {
4733                         block_rsv->reserved += buf->len;
4734                         ret = 0;
4735                 }
4736                 spin_unlock(&block_rsv->lock);
4737
4738                 if (ret) {
4739                         spin_lock(&cache->space_info->lock);
4740                         cache->space_info->bytes_reserved -= buf->len;
4741                         cache->space_info->reservation_progress++;
4742                         spin_unlock(&cache->space_info->lock);
4743                 }
4744                 goto out;
4745         }
4746 pin:
4747         if (block_rsv->durable && !cache->ro) {
4748                 ret = 0;
4749                 spin_lock(&cache->lock);
4750                 if (!cache->ro) {
4751                         cache->reserved_pinned += buf->len;
4752                         ret = 1;
4753                 }
4754                 spin_unlock(&cache->lock);
4755
4756                 if (ret) {
4757                         spin_lock(&block_rsv->lock);
4758                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4759                         spin_unlock(&block_rsv->lock);
4760                 }
4761         }
4762 out:
4763         /*
4764          * Deleting the buffer, clear the corrupt flag since it doesn't matter
4765          * anymore.
4766          */
4767         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4768         btrfs_put_block_group(cache);
4769 }
4770
4771 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4772                       struct btrfs_root *root,
4773                       u64 bytenr, u64 num_bytes, u64 parent,
4774                       u64 root_objectid, u64 owner, u64 offset)
4775 {
4776         int ret;
4777
4778         /*
4779          * tree log blocks never actually go into the extent allocation
4780          * tree, just update pinning info and exit early.
4781          */
4782         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4783                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4784                 /* unlocks the pinned mutex */
4785                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4786                 ret = 0;
4787         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4788                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4789                                         parent, root_objectid, (int)owner,
4790                                         BTRFS_DROP_DELAYED_REF, NULL);
4791                 BUG_ON(ret);
4792         } else {
4793                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4794                                         parent, root_objectid, owner,
4795                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4796                 BUG_ON(ret);
4797         }
4798         return ret;
4799 }
4800
4801 static u64 stripe_align(struct btrfs_root *root, u64 val)
4802 {
4803         u64 mask = ((u64)root->stripesize - 1);
4804         u64 ret = (val + mask) & ~mask;
4805         return ret;
4806 }
4807
4808 /*
4809  * when we wait for progress in the block group caching, its because
4810  * our allocation attempt failed at least once.  So, we must sleep
4811  * and let some progress happen before we try again.
4812  *
4813  * This function will sleep at least once waiting for new free space to
4814  * show up, and then it will check the block group free space numbers
4815  * for our min num_bytes.  Another option is to have it go ahead
4816  * and look in the rbtree for a free extent of a given size, but this
4817  * is a good start.
4818  */
4819 static noinline int
4820 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4821                                 u64 num_bytes)
4822 {
4823         struct btrfs_caching_control *caching_ctl;
4824         DEFINE_WAIT(wait);
4825
4826         caching_ctl = get_caching_control(cache);
4827         if (!caching_ctl)
4828                 return 0;
4829
4830         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4831                    (cache->free_space >= num_bytes));
4832
4833         put_caching_control(caching_ctl);
4834         return 0;
4835 }
4836
4837 static noinline int
4838 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4839 {
4840         struct btrfs_caching_control *caching_ctl;
4841         DEFINE_WAIT(wait);
4842
4843         caching_ctl = get_caching_control(cache);
4844         if (!caching_ctl)
4845                 return 0;
4846
4847         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4848
4849         put_caching_control(caching_ctl);
4850         return 0;
4851 }
4852
4853 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4854 {
4855         int index;
4856         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4857                 index = 0;
4858         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4859                 index = 1;
4860         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4861                 index = 2;
4862         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4863                 index = 3;
4864         else
4865                 index = 4;
4866         return index;
4867 }
4868
4869 enum btrfs_loop_type {
4870         LOOP_FIND_IDEAL = 0,
4871         LOOP_CACHING_NOWAIT = 1,
4872         LOOP_CACHING_WAIT = 2,
4873         LOOP_ALLOC_CHUNK = 3,
4874         LOOP_NO_EMPTY_SIZE = 4,
4875 };
4876
4877 /*
4878  * walks the btree of allocated extents and find a hole of a given size.
4879  * The key ins is changed to record the hole:
4880  * ins->objectid == block start
4881  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4882  * ins->offset == number of blocks
4883  * Any available blocks before search_start are skipped.
4884  */
4885 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4886                                      struct btrfs_root *orig_root,
4887                                      u64 num_bytes, u64 empty_size,
4888                                      u64 search_start, u64 search_end,
4889                                      u64 hint_byte, struct btrfs_key *ins,
4890                                      int data)
4891 {
4892         int ret = 0;
4893         struct btrfs_root *root = orig_root->fs_info->extent_root;
4894         struct btrfs_free_cluster *last_ptr = NULL;
4895         struct btrfs_block_group_cache *block_group = NULL;
4896         int empty_cluster = 2 * 1024 * 1024;
4897         int allowed_chunk_alloc = 0;
4898         int done_chunk_alloc = 0;
4899         struct btrfs_space_info *space_info;
4900         int last_ptr_loop = 0;
4901         int loop = 0;
4902         int index = 0;
4903         bool found_uncached_bg = false;
4904         bool failed_cluster_refill = false;
4905         bool failed_alloc = false;
4906         bool use_cluster = true;
4907         u64 ideal_cache_percent = 0;
4908         u64 ideal_cache_offset = 0;
4909
4910         WARN_ON(num_bytes < root->sectorsize);
4911         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4912         ins->objectid = 0;
4913         ins->offset = 0;
4914
4915         space_info = __find_space_info(root->fs_info, data);
4916         if (!space_info) {
4917                 printk(KERN_ERR "No space info for %d\n", data);
4918                 return -ENOSPC;
4919         }
4920
4921         /*
4922          * If the space info is for both data and metadata it means we have a
4923          * small filesystem and we can't use the clustering stuff.
4924          */
4925         if (btrfs_mixed_space_info(space_info))
4926                 use_cluster = false;
4927
4928         if (orig_root->ref_cows || empty_size)
4929                 allowed_chunk_alloc = 1;
4930
4931         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4932                 last_ptr = &root->fs_info->meta_alloc_cluster;
4933                 if (!btrfs_test_opt(root, SSD))
4934                         empty_cluster = 64 * 1024;
4935         }
4936
4937         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4938             btrfs_test_opt(root, SSD)) {
4939                 last_ptr = &root->fs_info->data_alloc_cluster;
4940         }
4941
4942         if (last_ptr) {
4943                 spin_lock(&last_ptr->lock);
4944                 if (last_ptr->block_group)
4945                         hint_byte = last_ptr->window_start;
4946                 spin_unlock(&last_ptr->lock);
4947         }
4948
4949         search_start = max(search_start, first_logical_byte(root, 0));
4950         search_start = max(search_start, hint_byte);
4951
4952         if (!last_ptr)
4953                 empty_cluster = 0;
4954
4955         if (search_start == hint_byte) {
4956 ideal_cache:
4957                 block_group = btrfs_lookup_block_group(root->fs_info,
4958                                                        search_start);
4959                 /*
4960                  * we don't want to use the block group if it doesn't match our
4961                  * allocation bits, or if its not cached.
4962                  *
4963                  * However if we are re-searching with an ideal block group
4964                  * picked out then we don't care that the block group is cached.
4965                  */
4966                 if (block_group && block_group_bits(block_group, data) &&
4967                     (block_group->cached != BTRFS_CACHE_NO ||
4968                      search_start == ideal_cache_offset)) {
4969                         down_read(&space_info->groups_sem);
4970                         if (list_empty(&block_group->list) ||
4971                             block_group->ro) {
4972                                 /*
4973                                  * someone is removing this block group,
4974                                  * we can't jump into the have_block_group
4975                                  * target because our list pointers are not
4976                                  * valid
4977                                  */
4978                                 btrfs_put_block_group(block_group);
4979                                 up_read(&space_info->groups_sem);
4980                         } else {
4981                                 index = get_block_group_index(block_group);
4982                                 goto have_block_group;
4983                         }
4984                 } else if (block_group) {
4985                         btrfs_put_block_group(block_group);
4986                 }
4987         }
4988 search:
4989         down_read(&space_info->groups_sem);
4990         list_for_each_entry(block_group, &space_info->block_groups[index],
4991                             list) {
4992                 u64 offset;
4993                 int cached;
4994
4995                 btrfs_get_block_group(block_group);
4996                 search_start = block_group->key.objectid;
4997
4998                 /*
4999                  * this can happen if we end up cycling through all the
5000                  * raid types, but we want to make sure we only allocate
5001                  * for the proper type.
5002                  */
5003                 if (!block_group_bits(block_group, data)) {
5004                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5005                                 BTRFS_BLOCK_GROUP_RAID1 |
5006                                 BTRFS_BLOCK_GROUP_RAID10;
5007
5008                         /*
5009                          * if they asked for extra copies and this block group
5010                          * doesn't provide them, bail.  This does allow us to
5011                          * fill raid0 from raid1.
5012                          */
5013                         if ((data & extra) && !(block_group->flags & extra))
5014                                 goto loop;
5015                 }
5016
5017 have_block_group:
5018                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5019                         u64 free_percent;
5020
5021                         ret = cache_block_group(block_group, trans,
5022                                                 orig_root, 1);
5023                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5024                                 goto have_block_group;
5025
5026                         free_percent = btrfs_block_group_used(&block_group->item);
5027                         free_percent *= 100;
5028                         free_percent = div64_u64(free_percent,
5029                                                  block_group->key.offset);
5030                         free_percent = 100 - free_percent;
5031                         if (free_percent > ideal_cache_percent &&
5032                             likely(!block_group->ro)) {
5033                                 ideal_cache_offset = block_group->key.objectid;
5034                                 ideal_cache_percent = free_percent;
5035                         }
5036
5037                         /*
5038                          * We only want to start kthread caching if we are at
5039                          * the point where we will wait for caching to make
5040                          * progress, or if our ideal search is over and we've
5041                          * found somebody to start caching.
5042                          */
5043                         if (loop > LOOP_CACHING_NOWAIT ||
5044                             (loop > LOOP_FIND_IDEAL &&
5045                              atomic_read(&space_info->caching_threads) < 2)) {
5046                                 ret = cache_block_group(block_group, trans,
5047                                                         orig_root, 0);
5048                                 BUG_ON(ret);
5049                         }
5050                         found_uncached_bg = true;
5051
5052                         /*
5053                          * If loop is set for cached only, try the next block
5054                          * group.
5055                          */
5056                         if (loop == LOOP_FIND_IDEAL)
5057                                 goto loop;
5058                 }
5059
5060                 cached = block_group_cache_done(block_group);
5061                 if (unlikely(!cached))
5062                         found_uncached_bg = true;
5063
5064                 if (unlikely(block_group->ro))
5065                         goto loop;
5066
5067                 /*
5068                  * Ok we want to try and use the cluster allocator, so lets look
5069                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5070                  * have tried the cluster allocator plenty of times at this
5071                  * point and not have found anything, so we are likely way too
5072                  * fragmented for the clustering stuff to find anything, so lets
5073                  * just skip it and let the allocator find whatever block it can
5074                  * find
5075                  */
5076                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5077                         /*
5078                          * the refill lock keeps out other
5079                          * people trying to start a new cluster
5080                          */
5081                         spin_lock(&last_ptr->refill_lock);
5082                         if (last_ptr->block_group &&
5083                             (last_ptr->block_group->ro ||
5084                             !block_group_bits(last_ptr->block_group, data))) {
5085                                 offset = 0;
5086                                 goto refill_cluster;
5087                         }
5088
5089                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5090                                                  num_bytes, search_start);
5091                         if (offset) {
5092                                 /* we have a block, we're done */
5093                                 spin_unlock(&last_ptr->refill_lock);
5094                                 goto checks;
5095                         }
5096
5097                         spin_lock(&last_ptr->lock);
5098                         /*
5099                          * whoops, this cluster doesn't actually point to
5100                          * this block group.  Get a ref on the block
5101                          * group is does point to and try again
5102                          */
5103                         if (!last_ptr_loop && last_ptr->block_group &&
5104                             last_ptr->block_group != block_group) {
5105
5106                                 btrfs_put_block_group(block_group);
5107                                 block_group = last_ptr->block_group;
5108                                 btrfs_get_block_group(block_group);
5109                                 spin_unlock(&last_ptr->lock);
5110                                 spin_unlock(&last_ptr->refill_lock);
5111
5112                                 last_ptr_loop = 1;
5113                                 search_start = block_group->key.objectid;
5114                                 /*
5115                                  * we know this block group is properly
5116                                  * in the list because
5117                                  * btrfs_remove_block_group, drops the
5118                                  * cluster before it removes the block
5119                                  * group from the list
5120                                  */
5121                                 goto have_block_group;
5122                         }
5123                         spin_unlock(&last_ptr->lock);
5124 refill_cluster:
5125                         /*
5126                          * this cluster didn't work out, free it and
5127                          * start over
5128                          */
5129                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5130
5131                         last_ptr_loop = 0;
5132
5133                         /* allocate a cluster in this block group */
5134                         ret = btrfs_find_space_cluster(trans, root,
5135                                                block_group, last_ptr,
5136                                                offset, num_bytes,
5137                                                empty_cluster + empty_size);
5138                         if (ret == 0) {
5139                                 /*
5140                                  * now pull our allocation out of this
5141                                  * cluster
5142                                  */
5143                                 offset = btrfs_alloc_from_cluster(block_group,
5144                                                   last_ptr, num_bytes,
5145                                                   search_start);
5146                                 if (offset) {
5147                                         /* we found one, proceed */
5148                                         spin_unlock(&last_ptr->refill_lock);
5149                                         goto checks;
5150                                 }
5151                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5152                                    && !failed_cluster_refill) {
5153                                 spin_unlock(&last_ptr->refill_lock);
5154
5155                                 failed_cluster_refill = true;
5156                                 wait_block_group_cache_progress(block_group,
5157                                        num_bytes + empty_cluster + empty_size);
5158                                 goto have_block_group;
5159                         }
5160
5161                         /*
5162                          * at this point we either didn't find a cluster
5163                          * or we weren't able to allocate a block from our
5164                          * cluster.  Free the cluster we've been trying
5165                          * to use, and go to the next block group
5166                          */
5167                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5168                         spin_unlock(&last_ptr->refill_lock);
5169                         goto loop;
5170                 }
5171
5172                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5173                                                     num_bytes, empty_size);
5174                 /*
5175                  * If we didn't find a chunk, and we haven't failed on this
5176                  * block group before, and this block group is in the middle of
5177                  * caching and we are ok with waiting, then go ahead and wait
5178                  * for progress to be made, and set failed_alloc to true.
5179                  *
5180                  * If failed_alloc is true then we've already waited on this
5181                  * block group once and should move on to the next block group.
5182                  */
5183                 if (!offset && !failed_alloc && !cached &&
5184                     loop > LOOP_CACHING_NOWAIT) {
5185                         wait_block_group_cache_progress(block_group,
5186                                                 num_bytes + empty_size);
5187                         failed_alloc = true;
5188                         goto have_block_group;
5189                 } else if (!offset) {
5190                         goto loop;
5191                 }
5192 checks:
5193                 search_start = stripe_align(root, offset);
5194                 /* move on to the next group */
5195                 if (search_start + num_bytes >= search_end) {
5196                         btrfs_add_free_space(block_group, offset, num_bytes);
5197                         goto loop;
5198                 }
5199
5200                 /* move on to the next group */
5201                 if (search_start + num_bytes >
5202                     block_group->key.objectid + block_group->key.offset) {
5203                         btrfs_add_free_space(block_group, offset, num_bytes);
5204                         goto loop;
5205                 }
5206
5207                 ins->objectid = search_start;
5208                 ins->offset = num_bytes;
5209
5210                 if (offset < search_start)
5211                         btrfs_add_free_space(block_group, offset,
5212                                              search_start - offset);
5213                 BUG_ON(offset > search_start);
5214
5215                 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
5216                                             (data & BTRFS_BLOCK_GROUP_DATA));
5217                 if (ret == -EAGAIN) {
5218                         btrfs_add_free_space(block_group, offset, num_bytes);
5219                         goto loop;
5220                 }
5221
5222                 /* we are all good, lets return */
5223                 ins->objectid = search_start;
5224                 ins->offset = num_bytes;
5225
5226                 if (offset < search_start)
5227                         btrfs_add_free_space(block_group, offset,
5228                                              search_start - offset);
5229                 BUG_ON(offset > search_start);
5230                 break;
5231 loop:
5232                 failed_cluster_refill = false;
5233                 failed_alloc = false;
5234                 BUG_ON(index != get_block_group_index(block_group));
5235                 btrfs_put_block_group(block_group);
5236         }
5237         up_read(&space_info->groups_sem);
5238
5239         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5240                 goto search;
5241
5242         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5243          *                      for them to make caching progress.  Also
5244          *                      determine the best possible bg to cache
5245          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5246          *                      caching kthreads as we move along
5247          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5248          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5249          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5250          *                      again
5251          */
5252         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5253             (found_uncached_bg || empty_size || empty_cluster ||
5254              allowed_chunk_alloc)) {
5255                 index = 0;
5256                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5257                         found_uncached_bg = false;
5258                         loop++;
5259                         if (!ideal_cache_percent &&
5260                             atomic_read(&space_info->caching_threads))
5261                                 goto search;
5262
5263                         /*
5264                          * 1 of the following 2 things have happened so far
5265                          *
5266                          * 1) We found an ideal block group for caching that
5267                          * is mostly full and will cache quickly, so we might
5268                          * as well wait for it.
5269                          *
5270                          * 2) We searched for cached only and we didn't find
5271                          * anything, and we didn't start any caching kthreads
5272                          * either, so chances are we will loop through and
5273                          * start a couple caching kthreads, and then come back
5274                          * around and just wait for them.  This will be slower
5275                          * because we will have 2 caching kthreads reading at
5276                          * the same time when we could have just started one
5277                          * and waited for it to get far enough to give us an
5278                          * allocation, so go ahead and go to the wait caching
5279                          * loop.
5280                          */
5281                         loop = LOOP_CACHING_WAIT;
5282                         search_start = ideal_cache_offset;
5283                         ideal_cache_percent = 0;
5284                         goto ideal_cache;
5285                 } else if (loop == LOOP_FIND_IDEAL) {
5286                         /*
5287                          * Didn't find a uncached bg, wait on anything we find
5288                          * next.
5289                          */
5290                         loop = LOOP_CACHING_WAIT;
5291                         goto search;
5292                 }
5293
5294                 if (loop < LOOP_CACHING_WAIT) {
5295                         loop++;
5296                         goto search;
5297                 }
5298
5299                 if (loop == LOOP_ALLOC_CHUNK) {
5300                         empty_size = 0;
5301                         empty_cluster = 0;
5302                 }
5303
5304                 if (allowed_chunk_alloc) {
5305                         ret = do_chunk_alloc(trans, root, num_bytes +
5306                                              2 * 1024 * 1024, data, 1);
5307                         allowed_chunk_alloc = 0;
5308                         done_chunk_alloc = 1;
5309                 } else if (!done_chunk_alloc) {
5310                         space_info->force_alloc = 1;
5311                 }
5312
5313                 if (loop < LOOP_NO_EMPTY_SIZE) {
5314                         loop++;
5315                         goto search;
5316                 }
5317                 ret = -ENOSPC;
5318         } else if (!ins->objectid) {
5319                 ret = -ENOSPC;
5320         }
5321
5322         /* we found what we needed */
5323         if (ins->objectid) {
5324                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5325                         trans->block_group = block_group->key.objectid;
5326
5327                 btrfs_put_block_group(block_group);
5328                 ret = 0;
5329         }
5330
5331         return ret;
5332 }
5333
5334 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5335                             int dump_block_groups)
5336 {
5337         struct btrfs_block_group_cache *cache;
5338         int index = 0;
5339
5340         spin_lock(&info->lock);
5341         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5342                (unsigned long long)(info->total_bytes - info->bytes_used -
5343                                     info->bytes_pinned - info->bytes_reserved -
5344                                     info->bytes_readonly),
5345                (info->full) ? "" : "not ");
5346         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5347                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5348                (unsigned long long)info->total_bytes,
5349                (unsigned long long)info->bytes_used,
5350                (unsigned long long)info->bytes_pinned,
5351                (unsigned long long)info->bytes_reserved,
5352                (unsigned long long)info->bytes_may_use,
5353                (unsigned long long)info->bytes_readonly);
5354         spin_unlock(&info->lock);
5355
5356         if (!dump_block_groups)
5357                 return;
5358
5359         down_read(&info->groups_sem);
5360 again:
5361         list_for_each_entry(cache, &info->block_groups[index], list) {
5362                 spin_lock(&cache->lock);
5363                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5364                        "%llu pinned %llu reserved\n",
5365                        (unsigned long long)cache->key.objectid,
5366                        (unsigned long long)cache->key.offset,
5367                        (unsigned long long)btrfs_block_group_used(&cache->item),
5368                        (unsigned long long)cache->pinned,
5369                        (unsigned long long)cache->reserved);
5370                 btrfs_dump_free_space(cache, bytes);
5371                 spin_unlock(&cache->lock);
5372         }
5373         if (++index < BTRFS_NR_RAID_TYPES)
5374                 goto again;
5375         up_read(&info->groups_sem);
5376 }
5377
5378 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5379                          struct btrfs_root *root,
5380                          u64 num_bytes, u64 min_alloc_size,
5381                          u64 empty_size, u64 hint_byte,
5382                          u64 search_end, struct btrfs_key *ins,
5383                          u64 data)
5384 {
5385         int ret;
5386         u64 search_start = 0;
5387
5388         data = btrfs_get_alloc_profile(root, data);
5389 again:
5390         /*
5391          * the only place that sets empty_size is btrfs_realloc_node, which
5392          * is not called recursively on allocations
5393          */
5394         if (empty_size || root->ref_cows)
5395                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5396                                      num_bytes + 2 * 1024 * 1024, data, 0);
5397
5398         WARN_ON(num_bytes < root->sectorsize);
5399         ret = find_free_extent(trans, root, num_bytes, empty_size,
5400                                search_start, search_end, hint_byte,
5401                                ins, data);
5402
5403         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5404                 num_bytes = num_bytes >> 1;
5405                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5406                 num_bytes = max(num_bytes, min_alloc_size);
5407                 do_chunk_alloc(trans, root->fs_info->extent_root,
5408                                num_bytes, data, 1);
5409                 goto again;
5410         }
5411         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5412                 struct btrfs_space_info *sinfo;
5413
5414                 sinfo = __find_space_info(root->fs_info, data);
5415                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5416                        "wanted %llu\n", (unsigned long long)data,
5417                        (unsigned long long)num_bytes);
5418                 dump_space_info(sinfo, num_bytes, 1);
5419         }
5420
5421         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5422
5423         return ret;
5424 }
5425
5426 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5427 {
5428         struct btrfs_block_group_cache *cache;
5429         int ret = 0;
5430
5431         cache = btrfs_lookup_block_group(root->fs_info, start);
5432         if (!cache) {
5433                 printk(KERN_ERR "Unable to find block group for %llu\n",
5434                        (unsigned long long)start);
5435                 return -ENOSPC;
5436         }
5437
5438         if (btrfs_test_opt(root, DISCARD))
5439                 ret = btrfs_discard_extent(root, start, len, NULL);
5440
5441         btrfs_add_free_space(cache, start, len);
5442         btrfs_update_reserved_bytes(cache, len, 0, 1);
5443         btrfs_put_block_group(cache);
5444
5445         trace_btrfs_reserved_extent_free(root, start, len);
5446
5447         return ret;
5448 }
5449
5450 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5451                                       struct btrfs_root *root,
5452                                       u64 parent, u64 root_objectid,
5453                                       u64 flags, u64 owner, u64 offset,
5454                                       struct btrfs_key *ins, int ref_mod)
5455 {
5456         int ret;
5457         struct btrfs_fs_info *fs_info = root->fs_info;
5458         struct btrfs_extent_item *extent_item;
5459         struct btrfs_extent_inline_ref *iref;
5460         struct btrfs_path *path;
5461         struct extent_buffer *leaf;
5462         int type;
5463         u32 size;
5464
5465         if (parent > 0)
5466                 type = BTRFS_SHARED_DATA_REF_KEY;
5467         else
5468                 type = BTRFS_EXTENT_DATA_REF_KEY;
5469
5470         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5471
5472         path = btrfs_alloc_path();
5473         if (!path)
5474                 return -ENOMEM;
5475
5476         path->leave_spinning = 1;
5477         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5478                                       ins, size);
5479         BUG_ON(ret);
5480
5481         leaf = path->nodes[0];
5482         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5483                                      struct btrfs_extent_item);
5484         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5485         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5486         btrfs_set_extent_flags(leaf, extent_item,
5487                                flags | BTRFS_EXTENT_FLAG_DATA);
5488
5489         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5490         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5491         if (parent > 0) {
5492                 struct btrfs_shared_data_ref *ref;
5493                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5494                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5495                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5496         } else {
5497                 struct btrfs_extent_data_ref *ref;
5498                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5499                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5500                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5501                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5502                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5503         }
5504
5505         btrfs_mark_buffer_dirty(path->nodes[0]);
5506         btrfs_free_path(path);
5507
5508         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5509         if (ret) {
5510                 printk(KERN_ERR "btrfs update block group failed for %llu "
5511                        "%llu\n", (unsigned long long)ins->objectid,
5512                        (unsigned long long)ins->offset);
5513                 BUG();
5514         }
5515         return ret;
5516 }
5517
5518 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5519                                      struct btrfs_root *root,
5520                                      u64 parent, u64 root_objectid,
5521                                      u64 flags, struct btrfs_disk_key *key,
5522                                      int level, struct btrfs_key *ins)
5523 {
5524         int ret;
5525         struct btrfs_fs_info *fs_info = root->fs_info;
5526         struct btrfs_extent_item *extent_item;
5527         struct btrfs_tree_block_info *block_info;
5528         struct btrfs_extent_inline_ref *iref;
5529         struct btrfs_path *path;
5530         struct extent_buffer *leaf;
5531         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5532
5533         path = btrfs_alloc_path();
5534         BUG_ON(!path);
5535
5536         path->leave_spinning = 1;
5537         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5538                                       ins, size);
5539         BUG_ON(ret);
5540
5541         leaf = path->nodes[0];
5542         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5543                                      struct btrfs_extent_item);
5544         btrfs_set_extent_refs(leaf, extent_item, 1);
5545         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5546         btrfs_set_extent_flags(leaf, extent_item,
5547                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5548         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5549
5550         btrfs_set_tree_block_key(leaf, block_info, key);
5551         btrfs_set_tree_block_level(leaf, block_info, level);
5552
5553         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5554         if (parent > 0) {
5555                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5556                 btrfs_set_extent_inline_ref_type(leaf, iref,
5557                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5558                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5559         } else {
5560                 btrfs_set_extent_inline_ref_type(leaf, iref,
5561                                                  BTRFS_TREE_BLOCK_REF_KEY);
5562                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5563         }
5564
5565         btrfs_mark_buffer_dirty(leaf);
5566         btrfs_free_path(path);
5567
5568         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5569         if (ret) {
5570                 printk(KERN_ERR "btrfs update block group failed for %llu "
5571                        "%llu\n", (unsigned long long)ins->objectid,
5572                        (unsigned long long)ins->offset);
5573                 BUG();
5574         }
5575         return ret;
5576 }
5577
5578 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5579                                      struct btrfs_root *root,
5580                                      u64 root_objectid, u64 owner,
5581                                      u64 offset, struct btrfs_key *ins)
5582 {
5583         int ret;
5584
5585         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5586
5587         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5588                                          0, root_objectid, owner, offset,
5589                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5590         return ret;
5591 }
5592
5593 /*
5594  * this is used by the tree logging recovery code.  It records that
5595  * an extent has been allocated and makes sure to clear the free
5596  * space cache bits as well
5597  */
5598 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5599                                    struct btrfs_root *root,
5600                                    u64 root_objectid, u64 owner, u64 offset,
5601                                    struct btrfs_key *ins)
5602 {
5603         int ret;
5604         struct btrfs_block_group_cache *block_group;
5605         struct btrfs_caching_control *caching_ctl;
5606         u64 start = ins->objectid;
5607         u64 num_bytes = ins->offset;
5608
5609         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5610         cache_block_group(block_group, trans, NULL, 0);
5611         caching_ctl = get_caching_control(block_group);
5612
5613         if (!caching_ctl) {
5614                 BUG_ON(!block_group_cache_done(block_group));
5615                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5616                 BUG_ON(ret);
5617         } else {
5618                 mutex_lock(&caching_ctl->mutex);
5619
5620                 if (start >= caching_ctl->progress) {
5621                         ret = add_excluded_extent(root, start, num_bytes);
5622                         BUG_ON(ret);
5623                 } else if (start + num_bytes <= caching_ctl->progress) {
5624                         ret = btrfs_remove_free_space(block_group,
5625                                                       start, num_bytes);
5626                         BUG_ON(ret);
5627                 } else {
5628                         num_bytes = caching_ctl->progress - start;
5629                         ret = btrfs_remove_free_space(block_group,
5630                                                       start, num_bytes);
5631                         BUG_ON(ret);
5632
5633                         start = caching_ctl->progress;
5634                         num_bytes = ins->objectid + ins->offset -
5635                                     caching_ctl->progress;
5636                         ret = add_excluded_extent(root, start, num_bytes);
5637                         BUG_ON(ret);
5638                 }
5639
5640                 mutex_unlock(&caching_ctl->mutex);
5641                 put_caching_control(caching_ctl);
5642         }
5643
5644         ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
5645         BUG_ON(ret);
5646         btrfs_put_block_group(block_group);
5647         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5648                                          0, owner, offset, ins, 1);
5649         return ret;
5650 }
5651
5652 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5653                                             struct btrfs_root *root,
5654                                             u64 bytenr, u32 blocksize,
5655                                             int level)
5656 {
5657         struct extent_buffer *buf;
5658
5659         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5660         if (!buf)
5661                 return ERR_PTR(-ENOMEM);
5662         btrfs_set_header_generation(buf, trans->transid);
5663         btrfs_set_buffer_lockdep_class(buf, level);
5664         btrfs_tree_lock(buf);
5665         clean_tree_block(trans, root, buf);
5666
5667         btrfs_set_lock_blocking(buf);
5668         btrfs_set_buffer_uptodate(buf);
5669
5670         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5671                 /*
5672                  * we allow two log transactions at a time, use different
5673                  * EXENT bit to differentiate dirty pages.
5674                  */
5675                 if (root->log_transid % 2 == 0)
5676                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5677                                         buf->start + buf->len - 1, GFP_NOFS);
5678                 else
5679                         set_extent_new(&root->dirty_log_pages, buf->start,
5680                                         buf->start + buf->len - 1, GFP_NOFS);
5681         } else {
5682                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5683                          buf->start + buf->len - 1, GFP_NOFS);
5684         }
5685         trans->blocks_used++;
5686         /* this returns a buffer locked for blocking */
5687         return buf;
5688 }
5689
5690 static struct btrfs_block_rsv *
5691 use_block_rsv(struct btrfs_trans_handle *trans,
5692               struct btrfs_root *root, u32 blocksize)
5693 {
5694         struct btrfs_block_rsv *block_rsv;
5695         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5696         int ret;
5697
5698         block_rsv = get_block_rsv(trans, root);
5699
5700         if (block_rsv->size == 0) {
5701                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5702                                              blocksize, 0);
5703                 /*
5704                  * If we couldn't reserve metadata bytes try and use some from
5705                  * the global reserve.
5706                  */
5707                 if (ret && block_rsv != global_rsv) {
5708                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5709                         if (!ret)
5710                                 return global_rsv;
5711                         return ERR_PTR(ret);
5712                 } else if (ret) {
5713                         return ERR_PTR(ret);
5714                 }
5715                 return block_rsv;
5716         }
5717
5718         ret = block_rsv_use_bytes(block_rsv, blocksize);
5719         if (!ret)
5720                 return block_rsv;
5721         if (ret) {
5722                 WARN_ON(1);
5723                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5724                                              0);
5725                 if (!ret) {
5726                         spin_lock(&block_rsv->lock);
5727                         block_rsv->size += blocksize;
5728                         spin_unlock(&block_rsv->lock);
5729                         return block_rsv;
5730                 } else if (ret && block_rsv != global_rsv) {
5731                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5732                         if (!ret)
5733                                 return global_rsv;
5734                 }
5735         }
5736
5737         return ERR_PTR(-ENOSPC);
5738 }
5739
5740 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5741 {
5742         block_rsv_add_bytes(block_rsv, blocksize, 0);
5743         block_rsv_release_bytes(block_rsv, NULL, 0);
5744 }
5745
5746 /*
5747  * finds a free extent and does all the dirty work required for allocation
5748  * returns the key for the extent through ins, and a tree buffer for
5749  * the first block of the extent through buf.
5750  *
5751  * returns the tree buffer or NULL.
5752  */
5753 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5754                                         struct btrfs_root *root, u32 blocksize,
5755                                         u64 parent, u64 root_objectid,
5756                                         struct btrfs_disk_key *key, int level,
5757                                         u64 hint, u64 empty_size)
5758 {
5759         struct btrfs_key ins;
5760         struct btrfs_block_rsv *block_rsv;
5761         struct extent_buffer *buf;
5762         u64 flags = 0;
5763         int ret;
5764
5765
5766         block_rsv = use_block_rsv(trans, root, blocksize);
5767         if (IS_ERR(block_rsv))
5768                 return ERR_CAST(block_rsv);
5769
5770         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5771                                    empty_size, hint, (u64)-1, &ins, 0);
5772         if (ret) {
5773                 unuse_block_rsv(block_rsv, blocksize);
5774                 return ERR_PTR(ret);
5775         }
5776
5777         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5778                                     blocksize, level);
5779         BUG_ON(IS_ERR(buf));
5780
5781         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5782                 if (parent == 0)
5783                         parent = ins.objectid;
5784                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5785         } else
5786                 BUG_ON(parent > 0);
5787
5788         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5789                 struct btrfs_delayed_extent_op *extent_op;
5790                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5791                 BUG_ON(!extent_op);
5792                 if (key)
5793                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5794                 else
5795                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5796                 extent_op->flags_to_set = flags;
5797                 extent_op->update_key = 1;
5798                 extent_op->update_flags = 1;
5799                 extent_op->is_data = 0;
5800
5801                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5802                                         ins.offset, parent, root_objectid,
5803                                         level, BTRFS_ADD_DELAYED_EXTENT,
5804                                         extent_op);
5805                 BUG_ON(ret);
5806         }
5807         return buf;
5808 }
5809
5810 struct walk_control {
5811         u64 refs[BTRFS_MAX_LEVEL];
5812         u64 flags[BTRFS_MAX_LEVEL];
5813         struct btrfs_key update_progress;
5814         int stage;
5815         int level;
5816         int shared_level;
5817         int update_ref;
5818         int keep_locks;
5819         int reada_slot;
5820         int reada_count;
5821 };
5822
5823 #define DROP_REFERENCE  1
5824 #define UPDATE_BACKREF  2
5825
5826 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5827                                      struct btrfs_root *root,
5828                                      struct walk_control *wc,
5829                                      struct btrfs_path *path)
5830 {
5831         u64 bytenr;
5832         u64 generation;
5833         u64 refs;
5834         u64 flags;
5835         u32 nritems;
5836         u32 blocksize;
5837         struct btrfs_key key;
5838         struct extent_buffer *eb;
5839         int ret;
5840         int slot;
5841         int nread = 0;
5842
5843         if (path->slots[wc->level] < wc->reada_slot) {
5844                 wc->reada_count = wc->reada_count * 2 / 3;
5845                 wc->reada_count = max(wc->reada_count, 2);
5846         } else {
5847                 wc->reada_count = wc->reada_count * 3 / 2;
5848                 wc->reada_count = min_t(int, wc->reada_count,
5849                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5850         }
5851
5852         eb = path->nodes[wc->level];
5853         nritems = btrfs_header_nritems(eb);
5854         blocksize = btrfs_level_size(root, wc->level - 1);
5855
5856         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5857                 if (nread >= wc->reada_count)
5858                         break;
5859
5860                 cond_resched();
5861                 bytenr = btrfs_node_blockptr(eb, slot);
5862                 generation = btrfs_node_ptr_generation(eb, slot);
5863
5864                 if (slot == path->slots[wc->level])
5865                         goto reada;
5866
5867                 if (wc->stage == UPDATE_BACKREF &&
5868                     generation <= root->root_key.offset)
5869                         continue;
5870
5871                 /* We don't lock the tree block, it's OK to be racy here */
5872                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5873                                                &refs, &flags);
5874                 BUG_ON(ret);
5875                 BUG_ON(refs == 0);
5876
5877                 if (wc->stage == DROP_REFERENCE) {
5878                         if (refs == 1)
5879                                 goto reada;
5880
5881                         if (wc->level == 1 &&
5882                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5883                                 continue;
5884                         if (!wc->update_ref ||
5885                             generation <= root->root_key.offset)
5886                                 continue;
5887                         btrfs_node_key_to_cpu(eb, &key, slot);
5888                         ret = btrfs_comp_cpu_keys(&key,
5889                                                   &wc->update_progress);
5890                         if (ret < 0)
5891                                 continue;
5892                 } else {
5893                         if (wc->level == 1 &&
5894                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5895                                 continue;
5896                 }
5897 reada:
5898                 ret = readahead_tree_block(root, bytenr, blocksize,
5899                                            generation);
5900                 if (ret)
5901                         break;
5902                 nread++;
5903         }
5904         wc->reada_slot = slot;
5905 }
5906
5907 /*
5908  * hepler to process tree block while walking down the tree.
5909  *
5910  * when wc->stage == UPDATE_BACKREF, this function updates
5911  * back refs for pointers in the block.
5912  *
5913  * NOTE: return value 1 means we should stop walking down.
5914  */
5915 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5916                                    struct btrfs_root *root,
5917                                    struct btrfs_path *path,
5918                                    struct walk_control *wc, int lookup_info)
5919 {
5920         int level = wc->level;
5921         struct extent_buffer *eb = path->nodes[level];
5922         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5923         int ret;
5924
5925         if (wc->stage == UPDATE_BACKREF &&
5926             btrfs_header_owner(eb) != root->root_key.objectid)
5927                 return 1;
5928
5929         /*
5930          * when reference count of tree block is 1, it won't increase
5931          * again. once full backref flag is set, we never clear it.
5932          */
5933         if (lookup_info &&
5934             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5935              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5936                 BUG_ON(!path->locks[level]);
5937                 ret = btrfs_lookup_extent_info(trans, root,
5938                                                eb->start, eb->len,
5939                                                &wc->refs[level],
5940                                                &wc->flags[level]);
5941                 BUG_ON(ret);
5942                 BUG_ON(wc->refs[level] == 0);
5943         }
5944
5945         if (wc->stage == DROP_REFERENCE) {
5946                 if (wc->refs[level] > 1)
5947                         return 1;
5948
5949                 if (path->locks[level] && !wc->keep_locks) {
5950                         btrfs_tree_unlock(eb);
5951                         path->locks[level] = 0;
5952                 }
5953                 return 0;
5954         }
5955
5956         /* wc->stage == UPDATE_BACKREF */
5957         if (!(wc->flags[level] & flag)) {
5958                 BUG_ON(!path->locks[level]);
5959                 ret = btrfs_inc_ref(trans, root, eb, 1);
5960                 BUG_ON(ret);
5961                 ret = btrfs_dec_ref(trans, root, eb, 0);
5962                 BUG_ON(ret);
5963                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5964                                                   eb->len, flag, 0);
5965                 BUG_ON(ret);
5966                 wc->flags[level] |= flag;
5967         }
5968
5969         /*
5970          * the block is shared by multiple trees, so it's not good to
5971          * keep the tree lock
5972          */
5973         if (path->locks[level] && level > 0) {
5974                 btrfs_tree_unlock(eb);
5975                 path->locks[level] = 0;
5976         }
5977         return 0;
5978 }
5979
5980 /*
5981  * hepler to process tree block pointer.
5982  *
5983  * when wc->stage == DROP_REFERENCE, this function checks
5984  * reference count of the block pointed to. if the block
5985  * is shared and we need update back refs for the subtree
5986  * rooted at the block, this function changes wc->stage to
5987  * UPDATE_BACKREF. if the block is shared and there is no
5988  * need to update back, this function drops the reference
5989  * to the block.
5990  *
5991  * NOTE: return value 1 means we should stop walking down.
5992  */
5993 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5994                                  struct btrfs_root *root,
5995                                  struct btrfs_path *path,
5996                                  struct walk_control *wc, int *lookup_info)
5997 {
5998         u64 bytenr;
5999         u64 generation;
6000         u64 parent;
6001         u32 blocksize;
6002         struct btrfs_key key;
6003         struct extent_buffer *next;
6004         int level = wc->level;
6005         int reada = 0;
6006         int ret = 0;
6007
6008         generation = btrfs_node_ptr_generation(path->nodes[level],
6009                                                path->slots[level]);
6010         /*
6011          * if the lower level block was created before the snapshot
6012          * was created, we know there is no need to update back refs
6013          * for the subtree
6014          */
6015         if (wc->stage == UPDATE_BACKREF &&
6016             generation <= root->root_key.offset) {
6017                 *lookup_info = 1;
6018                 return 1;
6019         }
6020
6021         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6022         blocksize = btrfs_level_size(root, level - 1);
6023
6024         next = btrfs_find_tree_block(root, bytenr, blocksize);
6025         if (!next) {
6026                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6027                 if (!next)
6028                         return -ENOMEM;
6029                 reada = 1;
6030         }
6031         btrfs_tree_lock(next);
6032         btrfs_set_lock_blocking(next);
6033
6034         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6035                                        &wc->refs[level - 1],
6036                                        &wc->flags[level - 1]);
6037         BUG_ON(ret);
6038         BUG_ON(wc->refs[level - 1] == 0);
6039         *lookup_info = 0;
6040
6041         if (wc->stage == DROP_REFERENCE) {
6042                 if (wc->refs[level - 1] > 1) {
6043                         if (level == 1 &&
6044                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6045                                 goto skip;
6046
6047                         if (!wc->update_ref ||
6048                             generation <= root->root_key.offset)
6049                                 goto skip;
6050
6051                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6052                                               path->slots[level]);
6053                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6054                         if (ret < 0)
6055                                 goto skip;
6056
6057                         wc->stage = UPDATE_BACKREF;
6058                         wc->shared_level = level - 1;
6059                 }
6060         } else {
6061                 if (level == 1 &&
6062                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6063                         goto skip;
6064         }
6065
6066         if (!btrfs_buffer_uptodate(next, generation)) {
6067                 btrfs_tree_unlock(next);
6068                 free_extent_buffer(next);
6069                 next = NULL;
6070                 *lookup_info = 1;
6071         }
6072
6073         if (!next) {
6074                 if (reada && level == 1)
6075                         reada_walk_down(trans, root, wc, path);
6076                 next = read_tree_block(root, bytenr, blocksize, generation);
6077                 if (!next)
6078                         return -EIO;
6079                 btrfs_tree_lock(next);
6080                 btrfs_set_lock_blocking(next);
6081         }
6082
6083         level--;
6084         BUG_ON(level != btrfs_header_level(next));
6085         path->nodes[level] = next;
6086         path->slots[level] = 0;
6087         path->locks[level] = 1;
6088         wc->level = level;
6089         if (wc->level == 1)
6090                 wc->reada_slot = 0;
6091         return 0;
6092 skip:
6093         wc->refs[level - 1] = 0;
6094         wc->flags[level - 1] = 0;
6095         if (wc->stage == DROP_REFERENCE) {
6096                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6097                         parent = path->nodes[level]->start;
6098                 } else {
6099                         BUG_ON(root->root_key.objectid !=
6100                                btrfs_header_owner(path->nodes[level]));
6101                         parent = 0;
6102                 }
6103
6104                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6105                                         root->root_key.objectid, level - 1, 0);
6106                 BUG_ON(ret);
6107         }
6108         btrfs_tree_unlock(next);
6109         free_extent_buffer(next);
6110         *lookup_info = 1;
6111         return 1;
6112 }
6113
6114 /*
6115  * hepler to process tree block while walking up the tree.
6116  *
6117  * when wc->stage == DROP_REFERENCE, this function drops
6118  * reference count on the block.
6119  *
6120  * when wc->stage == UPDATE_BACKREF, this function changes
6121  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6122  * to UPDATE_BACKREF previously while processing the block.
6123  *
6124  * NOTE: return value 1 means we should stop walking up.
6125  */
6126 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6127                                  struct btrfs_root *root,
6128                                  struct btrfs_path *path,
6129                                  struct walk_control *wc)
6130 {
6131         int ret;
6132         int level = wc->level;
6133         struct extent_buffer *eb = path->nodes[level];
6134         u64 parent = 0;
6135
6136         if (wc->stage == UPDATE_BACKREF) {
6137                 BUG_ON(wc->shared_level < level);
6138                 if (level < wc->shared_level)
6139                         goto out;
6140
6141                 ret = find_next_key(path, level + 1, &wc->update_progress);
6142                 if (ret > 0)
6143                         wc->update_ref = 0;
6144
6145                 wc->stage = DROP_REFERENCE;
6146                 wc->shared_level = -1;
6147                 path->slots[level] = 0;
6148
6149                 /*
6150                  * check reference count again if the block isn't locked.
6151                  * we should start walking down the tree again if reference
6152                  * count is one.
6153                  */
6154                 if (!path->locks[level]) {
6155                         BUG_ON(level == 0);
6156                         btrfs_tree_lock(eb);
6157                         btrfs_set_lock_blocking(eb);
6158                         path->locks[level] = 1;
6159
6160                         ret = btrfs_lookup_extent_info(trans, root,
6161                                                        eb->start, eb->len,
6162                                                        &wc->refs[level],
6163                                                        &wc->flags[level]);
6164                         BUG_ON(ret);
6165                         BUG_ON(wc->refs[level] == 0);
6166                         if (wc->refs[level] == 1) {
6167                                 btrfs_tree_unlock(eb);
6168                                 path->locks[level] = 0;
6169                                 return 1;
6170                         }
6171                 }
6172         }
6173
6174         /* wc->stage == DROP_REFERENCE */
6175         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6176
6177         if (wc->refs[level] == 1) {
6178                 if (level == 0) {
6179                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6180                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6181                         else
6182                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6183                         BUG_ON(ret);
6184                 }
6185                 /* make block locked assertion in clean_tree_block happy */
6186                 if (!path->locks[level] &&
6187                     btrfs_header_generation(eb) == trans->transid) {
6188                         btrfs_tree_lock(eb);
6189                         btrfs_set_lock_blocking(eb);
6190                         path->locks[level] = 1;
6191                 }
6192                 clean_tree_block(trans, root, eb);
6193         }
6194
6195         if (eb == root->node) {
6196                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6197                         parent = eb->start;
6198                 else
6199                         BUG_ON(root->root_key.objectid !=
6200                                btrfs_header_owner(eb));
6201         } else {
6202                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6203                         parent = path->nodes[level + 1]->start;
6204                 else
6205                         BUG_ON(root->root_key.objectid !=
6206                                btrfs_header_owner(path->nodes[level + 1]));
6207         }
6208
6209         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6210 out:
6211         wc->refs[level] = 0;
6212         wc->flags[level] = 0;
6213         return 0;
6214 }
6215
6216 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6217                                    struct btrfs_root *root,
6218                                    struct btrfs_path *path,
6219                                    struct walk_control *wc)
6220 {
6221         int level = wc->level;
6222         int lookup_info = 1;
6223         int ret;
6224
6225         while (level >= 0) {
6226                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6227                 if (ret > 0)
6228                         break;
6229
6230                 if (level == 0)
6231                         break;
6232
6233                 if (path->slots[level] >=
6234                     btrfs_header_nritems(path->nodes[level]))
6235                         break;
6236
6237                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6238                 if (ret > 0) {
6239                         path->slots[level]++;
6240                         continue;
6241                 } else if (ret < 0)
6242                         return ret;
6243                 level = wc->level;
6244         }
6245         return 0;
6246 }
6247
6248 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6249                                  struct btrfs_root *root,
6250                                  struct btrfs_path *path,
6251                                  struct walk_control *wc, int max_level)
6252 {
6253         int level = wc->level;
6254         int ret;
6255
6256         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6257         while (level < max_level && path->nodes[level]) {
6258                 wc->level = level;
6259                 if (path->slots[level] + 1 <
6260                     btrfs_header_nritems(path->nodes[level])) {
6261                         path->slots[level]++;
6262                         return 0;
6263                 } else {
6264                         ret = walk_up_proc(trans, root, path, wc);
6265                         if (ret > 0)
6266                                 return 0;
6267
6268                         if (path->locks[level]) {
6269                                 btrfs_tree_unlock(path->nodes[level]);
6270                                 path->locks[level] = 0;
6271                         }
6272                         free_extent_buffer(path->nodes[level]);
6273                         path->nodes[level] = NULL;
6274                         level++;
6275                 }
6276         }
6277         return 1;
6278 }
6279
6280 /*
6281  * drop a subvolume tree.
6282  *
6283  * this function traverses the tree freeing any blocks that only
6284  * referenced by the tree.
6285  *
6286  * when a shared tree block is found. this function decreases its
6287  * reference count by one. if update_ref is true, this function
6288  * also make sure backrefs for the shared block and all lower level
6289  * blocks are properly updated.
6290  */
6291 int btrfs_drop_snapshot(struct btrfs_root *root,
6292                         struct btrfs_block_rsv *block_rsv, int update_ref)
6293 {
6294         struct btrfs_path *path;
6295         struct btrfs_trans_handle *trans;
6296         struct btrfs_root *tree_root = root->fs_info->tree_root;
6297         struct btrfs_root_item *root_item = &root->root_item;
6298         struct walk_control *wc;
6299         struct btrfs_key key;
6300         int err = 0;
6301         int ret;
6302         int level;
6303
6304         path = btrfs_alloc_path();
6305         BUG_ON(!path);
6306
6307         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6308         BUG_ON(!wc);
6309
6310         trans = btrfs_start_transaction(tree_root, 0);
6311         BUG_ON(IS_ERR(trans));
6312
6313         if (block_rsv)
6314                 trans->block_rsv = block_rsv;
6315
6316         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6317                 level = btrfs_header_level(root->node);
6318                 path->nodes[level] = btrfs_lock_root_node(root);
6319                 btrfs_set_lock_blocking(path->nodes[level]);
6320                 path->slots[level] = 0;
6321                 path->locks[level] = 1;
6322                 memset(&wc->update_progress, 0,
6323                        sizeof(wc->update_progress));
6324         } else {
6325                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6326                 memcpy(&wc->update_progress, &key,
6327                        sizeof(wc->update_progress));
6328
6329                 level = root_item->drop_level;
6330                 BUG_ON(level == 0);
6331                 path->lowest_level = level;
6332                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6333                 path->lowest_level = 0;
6334                 if (ret < 0) {
6335                         err = ret;
6336                         goto out;
6337                 }
6338                 WARN_ON(ret > 0);
6339
6340                 /*
6341                  * unlock our path, this is safe because only this
6342                  * function is allowed to delete this snapshot
6343                  */
6344                 btrfs_unlock_up_safe(path, 0);
6345
6346                 level = btrfs_header_level(root->node);
6347                 while (1) {
6348                         btrfs_tree_lock(path->nodes[level]);
6349                         btrfs_set_lock_blocking(path->nodes[level]);
6350
6351                         ret = btrfs_lookup_extent_info(trans, root,
6352                                                 path->nodes[level]->start,
6353                                                 path->nodes[level]->len,
6354                                                 &wc->refs[level],
6355                                                 &wc->flags[level]);
6356                         BUG_ON(ret);
6357                         BUG_ON(wc->refs[level] == 0);
6358
6359                         if (level == root_item->drop_level)
6360                                 break;
6361
6362                         btrfs_tree_unlock(path->nodes[level]);
6363                         WARN_ON(wc->refs[level] != 1);
6364                         level--;
6365                 }
6366         }
6367
6368         wc->level = level;
6369         wc->shared_level = -1;
6370         wc->stage = DROP_REFERENCE;
6371         wc->update_ref = update_ref;
6372         wc->keep_locks = 0;
6373         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6374
6375         while (1) {
6376                 ret = walk_down_tree(trans, root, path, wc);
6377                 if (ret < 0) {
6378                         err = ret;
6379                         break;
6380                 }
6381
6382                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6383                 if (ret < 0) {
6384                         err = ret;
6385                         break;
6386                 }
6387
6388                 if (ret > 0) {
6389                         BUG_ON(wc->stage != DROP_REFERENCE);
6390                         break;
6391                 }
6392
6393                 if (wc->stage == DROP_REFERENCE) {
6394                         level = wc->level;
6395                         btrfs_node_key(path->nodes[level],
6396                                        &root_item->drop_progress,
6397                                        path->slots[level]);
6398                         root_item->drop_level = level;
6399                 }
6400
6401                 BUG_ON(wc->level == 0);
6402                 if (btrfs_should_end_transaction(trans, tree_root)) {
6403                         ret = btrfs_update_root(trans, tree_root,
6404                                                 &root->root_key,
6405                                                 root_item);
6406                         BUG_ON(ret);
6407
6408                         btrfs_end_transaction_throttle(trans, tree_root);
6409                         trans = btrfs_start_transaction(tree_root, 0);
6410                         BUG_ON(IS_ERR(trans));
6411                         if (block_rsv)
6412                                 trans->block_rsv = block_rsv;
6413                 }
6414         }
6415         btrfs_release_path(root, path);
6416         BUG_ON(err);
6417
6418         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6419         BUG_ON(ret);
6420
6421         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6422                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6423                                            NULL, NULL);
6424                 BUG_ON(ret < 0);
6425                 if (ret > 0) {
6426                         /* if we fail to delete the orphan item this time
6427                          * around, it'll get picked up the next time.
6428                          *
6429                          * The most common failure here is just -ENOENT.
6430                          */
6431                         btrfs_del_orphan_item(trans, tree_root,
6432                                               root->root_key.objectid);
6433                 }
6434         }
6435
6436         if (root->in_radix) {
6437                 btrfs_free_fs_root(tree_root->fs_info, root);
6438         } else {
6439                 free_extent_buffer(root->node);
6440                 free_extent_buffer(root->commit_root);
6441                 kfree(root);
6442         }
6443 out:
6444         btrfs_end_transaction_throttle(trans, tree_root);
6445         kfree(wc);
6446         btrfs_free_path(path);
6447         return err;
6448 }
6449
6450 /*
6451  * drop subtree rooted at tree block 'node'.
6452  *
6453  * NOTE: this function will unlock and release tree block 'node'
6454  */
6455 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6456                         struct btrfs_root *root,
6457                         struct extent_buffer *node,
6458                         struct extent_buffer *parent)
6459 {
6460         struct btrfs_path *path;
6461         struct walk_control *wc;
6462         int level;
6463         int parent_level;
6464         int ret = 0;
6465         int wret;
6466
6467         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6468
6469         path = btrfs_alloc_path();
6470         if (!path)
6471                 return -ENOMEM;
6472
6473         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6474         if (!wc) {
6475                 btrfs_free_path(path);
6476                 return -ENOMEM;
6477         }
6478
6479         btrfs_assert_tree_locked(parent);
6480         parent_level = btrfs_header_level(parent);
6481         extent_buffer_get(parent);
6482         path->nodes[parent_level] = parent;
6483         path->slots[parent_level] = btrfs_header_nritems(parent);
6484
6485         btrfs_assert_tree_locked(node);
6486         level = btrfs_header_level(node);
6487         path->nodes[level] = node;
6488         path->slots[level] = 0;
6489         path->locks[level] = 1;
6490
6491         wc->refs[parent_level] = 1;
6492         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6493         wc->level = level;
6494         wc->shared_level = -1;
6495         wc->stage = DROP_REFERENCE;
6496         wc->update_ref = 0;
6497         wc->keep_locks = 1;
6498         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6499
6500         while (1) {
6501                 wret = walk_down_tree(trans, root, path, wc);
6502                 if (wret < 0) {
6503                         ret = wret;
6504                         break;
6505                 }
6506
6507                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6508                 if (wret < 0)
6509                         ret = wret;
6510                 if (wret != 0)
6511                         break;
6512         }
6513
6514         kfree(wc);
6515         btrfs_free_path(path);
6516         return ret;
6517 }
6518
6519 #if 0
6520 static unsigned long calc_ra(unsigned long start, unsigned long last,
6521                              unsigned long nr)
6522 {
6523         return min(last, start + nr - 1);
6524 }
6525
6526 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6527                                          u64 len)
6528 {
6529         u64 page_start;
6530         u64 page_end;
6531         unsigned long first_index;
6532         unsigned long last_index;
6533         unsigned long i;
6534         struct page *page;
6535         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6536         struct file_ra_state *ra;
6537         struct btrfs_ordered_extent *ordered;
6538         unsigned int total_read = 0;
6539         unsigned int total_dirty = 0;
6540         int ret = 0;
6541
6542         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6543         if (!ra)
6544                 return -ENOMEM;
6545
6546         mutex_lock(&inode->i_mutex);
6547         first_index = start >> PAGE_CACHE_SHIFT;
6548         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6549
6550         /* make sure the dirty trick played by the caller work */
6551         ret = invalidate_inode_pages2_range(inode->i_mapping,
6552                                             first_index, last_index);
6553         if (ret)
6554                 goto out_unlock;
6555
6556         file_ra_state_init(ra, inode->i_mapping);
6557
6558         for (i = first_index ; i <= last_index; i++) {
6559                 if (total_read % ra->ra_pages == 0) {
6560                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6561                                        calc_ra(i, last_index, ra->ra_pages));
6562                 }
6563                 total_read++;
6564 again:
6565                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6566                         BUG_ON(1);
6567                 page = grab_cache_page(inode->i_mapping, i);
6568                 if (!page) {
6569                         ret = -ENOMEM;
6570                         goto out_unlock;
6571                 }
6572                 if (!PageUptodate(page)) {
6573                         btrfs_readpage(NULL, page);
6574                         lock_page(page);
6575                         if (!PageUptodate(page)) {
6576                                 unlock_page(page);
6577                                 page_cache_release(page);
6578                                 ret = -EIO;
6579                                 goto out_unlock;
6580                         }
6581                 }
6582                 wait_on_page_writeback(page);
6583
6584                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6585                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6586                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6587
6588                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6589                 if (ordered) {
6590                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6591                         unlock_page(page);
6592                         page_cache_release(page);
6593                         btrfs_start_ordered_extent(inode, ordered, 1);
6594                         btrfs_put_ordered_extent(ordered);
6595                         goto again;
6596                 }
6597                 set_page_extent_mapped(page);
6598
6599                 if (i == first_index)
6600                         set_extent_bits(io_tree, page_start, page_end,
6601                                         EXTENT_BOUNDARY, GFP_NOFS);
6602                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6603
6604                 set_page_dirty(page);
6605                 total_dirty++;
6606
6607                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6608                 unlock_page(page);
6609                 page_cache_release(page);
6610         }
6611
6612 out_unlock:
6613         kfree(ra);
6614         mutex_unlock(&inode->i_mutex);
6615         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6616         return ret;
6617 }
6618
6619 static noinline int relocate_data_extent(struct inode *reloc_inode,
6620                                          struct btrfs_key *extent_key,
6621                                          u64 offset)
6622 {
6623         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6624         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6625         struct extent_map *em;
6626         u64 start = extent_key->objectid - offset;
6627         u64 end = start + extent_key->offset - 1;
6628
6629         em = alloc_extent_map(GFP_NOFS);
6630         BUG_ON(!em);
6631
6632         em->start = start;
6633         em->len = extent_key->offset;
6634         em->block_len = extent_key->offset;
6635         em->block_start = extent_key->objectid;
6636         em->bdev = root->fs_info->fs_devices->latest_bdev;
6637         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6638
6639         /* setup extent map to cheat btrfs_readpage */
6640         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6641         while (1) {
6642                 int ret;
6643                 write_lock(&em_tree->lock);
6644                 ret = add_extent_mapping(em_tree, em);
6645                 write_unlock(&em_tree->lock);
6646                 if (ret != -EEXIST) {
6647                         free_extent_map(em);
6648                         break;
6649                 }
6650                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6651         }
6652         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6653
6654         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6655 }
6656
6657 struct btrfs_ref_path {
6658         u64 extent_start;
6659         u64 nodes[BTRFS_MAX_LEVEL];
6660         u64 root_objectid;
6661         u64 root_generation;
6662         u64 owner_objectid;
6663         u32 num_refs;
6664         int lowest_level;
6665         int current_level;
6666         int shared_level;
6667
6668         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6669         u64 new_nodes[BTRFS_MAX_LEVEL];
6670 };
6671
6672 struct disk_extent {
6673         u64 ram_bytes;
6674         u64 disk_bytenr;
6675         u64 disk_num_bytes;
6676         u64 offset;
6677         u64 num_bytes;
6678         u8 compression;
6679         u8 encryption;
6680         u16 other_encoding;
6681 };
6682
6683 static int is_cowonly_root(u64 root_objectid)
6684 {
6685         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6686             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6687             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6688             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6689             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6690             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6691                 return 1;
6692         return 0;
6693 }
6694
6695 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6696                                     struct btrfs_root *extent_root,
6697                                     struct btrfs_ref_path *ref_path,
6698                                     int first_time)
6699 {
6700         struct extent_buffer *leaf;
6701         struct btrfs_path *path;
6702         struct btrfs_extent_ref *ref;
6703         struct btrfs_key key;
6704         struct btrfs_key found_key;
6705         u64 bytenr;
6706         u32 nritems;
6707         int level;
6708         int ret = 1;
6709
6710         path = btrfs_alloc_path();
6711         if (!path)
6712                 return -ENOMEM;
6713
6714         if (first_time) {
6715                 ref_path->lowest_level = -1;
6716                 ref_path->current_level = -1;
6717                 ref_path->shared_level = -1;
6718                 goto walk_up;
6719         }
6720 walk_down:
6721         level = ref_path->current_level - 1;
6722         while (level >= -1) {
6723                 u64 parent;
6724                 if (level < ref_path->lowest_level)
6725                         break;
6726
6727                 if (level >= 0)
6728                         bytenr = ref_path->nodes[level];
6729                 else
6730                         bytenr = ref_path->extent_start;
6731                 BUG_ON(bytenr == 0);
6732
6733                 parent = ref_path->nodes[level + 1];
6734                 ref_path->nodes[level + 1] = 0;
6735                 ref_path->current_level = level;
6736                 BUG_ON(parent == 0);
6737
6738                 key.objectid = bytenr;
6739                 key.offset = parent + 1;
6740                 key.type = BTRFS_EXTENT_REF_KEY;
6741
6742                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6743                 if (ret < 0)
6744                         goto out;
6745                 BUG_ON(ret == 0);
6746
6747                 leaf = path->nodes[0];
6748                 nritems = btrfs_header_nritems(leaf);
6749                 if (path->slots[0] >= nritems) {
6750                         ret = btrfs_next_leaf(extent_root, path);
6751                         if (ret < 0)
6752                                 goto out;
6753                         if (ret > 0)
6754                                 goto next;
6755                         leaf = path->nodes[0];
6756                 }
6757
6758                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6759                 if (found_key.objectid == bytenr &&
6760                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6761                         if (level < ref_path->shared_level)
6762                                 ref_path->shared_level = level;
6763                         goto found;
6764                 }
6765 next:
6766                 level--;
6767                 btrfs_release_path(extent_root, path);
6768                 cond_resched();
6769         }
6770         /* reached lowest level */
6771         ret = 1;
6772         goto out;
6773 walk_up:
6774         level = ref_path->current_level;
6775         while (level < BTRFS_MAX_LEVEL - 1) {
6776                 u64 ref_objectid;
6777
6778                 if (level >= 0)
6779                         bytenr = ref_path->nodes[level];
6780                 else
6781                         bytenr = ref_path->extent_start;
6782
6783                 BUG_ON(bytenr == 0);
6784
6785                 key.objectid = bytenr;
6786                 key.offset = 0;
6787                 key.type = BTRFS_EXTENT_REF_KEY;
6788
6789                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6790                 if (ret < 0)
6791                         goto out;
6792
6793                 leaf = path->nodes[0];
6794                 nritems = btrfs_header_nritems(leaf);
6795                 if (path->slots[0] >= nritems) {
6796                         ret = btrfs_next_leaf(extent_root, path);
6797                         if (ret < 0)
6798                                 goto out;
6799                         if (ret > 0) {
6800                                 /* the extent was freed by someone */
6801                                 if (ref_path->lowest_level == level)
6802                                         goto out;
6803                                 btrfs_release_path(extent_root, path);
6804                                 goto walk_down;
6805                         }
6806                         leaf = path->nodes[0];
6807                 }
6808
6809                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6810                 if (found_key.objectid != bytenr ||
6811                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6812                         /* the extent was freed by someone */
6813                         if (ref_path->lowest_level == level) {
6814                                 ret = 1;
6815                                 goto out;
6816                         }
6817                         btrfs_release_path(extent_root, path);
6818                         goto walk_down;
6819                 }
6820 found:
6821                 ref = btrfs_item_ptr(leaf, path->slots[0],
6822                                 struct btrfs_extent_ref);
6823                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6824                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6825                         if (first_time) {
6826                                 level = (int)ref_objectid;
6827                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6828                                 ref_path->lowest_level = level;
6829                                 ref_path->current_level = level;
6830                                 ref_path->nodes[level] = bytenr;
6831                         } else {
6832                                 WARN_ON(ref_objectid != level);
6833                         }
6834                 } else {
6835                         WARN_ON(level != -1);
6836                 }
6837                 first_time = 0;
6838
6839                 if (ref_path->lowest_level == level) {
6840                         ref_path->owner_objectid = ref_objectid;
6841                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6842                 }
6843
6844                 /*
6845                  * the block is tree root or the block isn't in reference
6846                  * counted tree.
6847                  */
6848                 if (found_key.objectid == found_key.offset ||
6849                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6850                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6851                         ref_path->root_generation =
6852                                 btrfs_ref_generation(leaf, ref);
6853                         if (level < 0) {
6854                                 /* special reference from the tree log */
6855                                 ref_path->nodes[0] = found_key.offset;
6856                                 ref_path->current_level = 0;
6857                         }
6858                         ret = 0;
6859                         goto out;
6860                 }
6861
6862                 level++;
6863                 BUG_ON(ref_path->nodes[level] != 0);
6864                 ref_path->nodes[level] = found_key.offset;
6865                 ref_path->current_level = level;
6866
6867                 /*
6868                  * the reference was created in the running transaction,
6869                  * no need to continue walking up.
6870                  */
6871                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6872                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6873                         ref_path->root_generation =
6874                                 btrfs_ref_generation(leaf, ref);
6875                         ret = 0;
6876                         goto out;
6877                 }
6878
6879                 btrfs_release_path(extent_root, path);
6880                 cond_resched();
6881         }
6882         /* reached max tree level, but no tree root found. */
6883         BUG();
6884 out:
6885         btrfs_free_path(path);
6886         return ret;
6887 }
6888
6889 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6890                                 struct btrfs_root *extent_root,
6891                                 struct btrfs_ref_path *ref_path,
6892                                 u64 extent_start)
6893 {
6894         memset(ref_path, 0, sizeof(*ref_path));
6895         ref_path->extent_start = extent_start;
6896
6897         return __next_ref_path(trans, extent_root, ref_path, 1);
6898 }
6899
6900 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6901                                struct btrfs_root *extent_root,
6902                                struct btrfs_ref_path *ref_path)
6903 {
6904         return __next_ref_path(trans, extent_root, ref_path, 0);
6905 }
6906
6907 static noinline int get_new_locations(struct inode *reloc_inode,
6908                                       struct btrfs_key *extent_key,
6909                                       u64 offset, int no_fragment,
6910                                       struct disk_extent **extents,
6911                                       int *nr_extents)
6912 {
6913         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6914         struct btrfs_path *path;
6915         struct btrfs_file_extent_item *fi;
6916         struct extent_buffer *leaf;
6917         struct disk_extent *exts = *extents;
6918         struct btrfs_key found_key;
6919         u64 cur_pos;
6920         u64 last_byte;
6921         u32 nritems;
6922         int nr = 0;
6923         int max = *nr_extents;
6924         int ret;
6925
6926         WARN_ON(!no_fragment && *extents);
6927         if (!exts) {
6928                 max = 1;
6929                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6930                 if (!exts)
6931                         return -ENOMEM;
6932         }
6933
6934         path = btrfs_alloc_path();
6935         if (!path) {
6936                 if (exts != *extents)
6937                         kfree(exts);
6938                 return -ENOMEM;
6939         }
6940
6941         cur_pos = extent_key->objectid - offset;
6942         last_byte = extent_key->objectid + extent_key->offset;
6943         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6944                                        cur_pos, 0);
6945         if (ret < 0)
6946                 goto out;
6947         if (ret > 0) {
6948                 ret = -ENOENT;
6949                 goto out;
6950         }
6951
6952         while (1) {
6953                 leaf = path->nodes[0];
6954                 nritems = btrfs_header_nritems(leaf);
6955                 if (path->slots[0] >= nritems) {
6956                         ret = btrfs_next_leaf(root, path);
6957                         if (ret < 0)
6958                                 goto out;
6959                         if (ret > 0)
6960                                 break;
6961                         leaf = path->nodes[0];
6962                 }
6963
6964                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6965                 if (found_key.offset != cur_pos ||
6966                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6967                     found_key.objectid != reloc_inode->i_ino)
6968                         break;
6969
6970                 fi = btrfs_item_ptr(leaf, path->slots[0],
6971                                     struct btrfs_file_extent_item);
6972                 if (btrfs_file_extent_type(leaf, fi) !=
6973                     BTRFS_FILE_EXTENT_REG ||
6974                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6975                         break;
6976
6977                 if (nr == max) {
6978                         struct disk_extent *old = exts;
6979                         max *= 2;
6980                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6981                         if (!exts) {
6982                                 ret = -ENOMEM;
6983                                 goto out;
6984                         }
6985                         memcpy(exts, old, sizeof(*exts) * nr);
6986                         if (old != *extents)
6987                                 kfree(old);
6988                 }
6989
6990                 exts[nr].disk_bytenr =
6991                         btrfs_file_extent_disk_bytenr(leaf, fi);
6992                 exts[nr].disk_num_bytes =
6993                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6994                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6995                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6996                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6997                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6998                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6999                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
7000                                                                            fi);
7001                 BUG_ON(exts[nr].offset > 0);
7002                 BUG_ON(exts[nr].compression || exts[nr].encryption);
7003                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
7004
7005                 cur_pos += exts[nr].num_bytes;
7006                 nr++;
7007
7008                 if (cur_pos + offset >= last_byte)
7009                         break;
7010
7011                 if (no_fragment) {
7012                         ret = 1;
7013                         goto out;
7014                 }
7015                 path->slots[0]++;
7016         }
7017
7018         BUG_ON(cur_pos + offset > last_byte);
7019         if (cur_pos + offset < last_byte) {
7020                 ret = -ENOENT;
7021                 goto out;
7022         }
7023         ret = 0;
7024 out:
7025         btrfs_free_path(path);
7026         if (ret) {
7027                 if (exts != *extents)
7028                         kfree(exts);
7029         } else {
7030                 *extents = exts;
7031                 *nr_extents = nr;
7032         }
7033         return ret;
7034 }
7035
7036 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
7037                                         struct btrfs_root *root,
7038                                         struct btrfs_path *path,
7039                                         struct btrfs_key *extent_key,
7040                                         struct btrfs_key *leaf_key,
7041                                         struct btrfs_ref_path *ref_path,
7042                                         struct disk_extent *new_extents,
7043                                         int nr_extents)
7044 {
7045         struct extent_buffer *leaf;
7046         struct btrfs_file_extent_item *fi;
7047         struct inode *inode = NULL;
7048         struct btrfs_key key;
7049         u64 lock_start = 0;
7050         u64 lock_end = 0;
7051         u64 num_bytes;
7052         u64 ext_offset;
7053         u64 search_end = (u64)-1;
7054         u32 nritems;
7055         int nr_scaned = 0;
7056         int extent_locked = 0;
7057         int extent_type;
7058         int ret;
7059
7060         memcpy(&key, leaf_key, sizeof(key));
7061         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7062                 if (key.objectid < ref_path->owner_objectid ||
7063                     (key.objectid == ref_path->owner_objectid &&
7064                      key.type < BTRFS_EXTENT_DATA_KEY)) {
7065                         key.objectid = ref_path->owner_objectid;
7066                         key.type = BTRFS_EXTENT_DATA_KEY;
7067                         key.offset = 0;
7068                 }
7069         }
7070
7071         while (1) {
7072                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7073                 if (ret < 0)
7074                         goto out;
7075
7076                 leaf = path->nodes[0];
7077                 nritems = btrfs_header_nritems(leaf);
7078 next:
7079                 if (extent_locked && ret > 0) {
7080                         /*
7081                          * the file extent item was modified by someone
7082                          * before the extent got locked.
7083                          */
7084                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7085                                       lock_end, GFP_NOFS);
7086                         extent_locked = 0;
7087                 }
7088
7089                 if (path->slots[0] >= nritems) {
7090                         if (++nr_scaned > 2)
7091                                 break;
7092
7093                         BUG_ON(extent_locked);
7094                         ret = btrfs_next_leaf(root, path);
7095                         if (ret < 0)
7096                                 goto out;
7097                         if (ret > 0)
7098                                 break;
7099                         leaf = path->nodes[0];
7100                         nritems = btrfs_header_nritems(leaf);
7101                 }
7102
7103                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7104
7105                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7106                         if ((key.objectid > ref_path->owner_objectid) ||
7107                             (key.objectid == ref_path->owner_objectid &&
7108                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7109                             key.offset >= search_end)
7110                                 break;
7111                 }
7112
7113                 if (inode && key.objectid != inode->i_ino) {
7114                         BUG_ON(extent_locked);
7115                         btrfs_release_path(root, path);
7116                         mutex_unlock(&inode->i_mutex);
7117                         iput(inode);
7118                         inode = NULL;
7119                         continue;
7120                 }
7121
7122                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7123                         path->slots[0]++;
7124                         ret = 1;
7125                         goto next;
7126                 }
7127                 fi = btrfs_item_ptr(leaf, path->slots[0],
7128                                     struct btrfs_file_extent_item);
7129                 extent_type = btrfs_file_extent_type(leaf, fi);
7130                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7131                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7132                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7133                      extent_key->objectid)) {
7134                         path->slots[0]++;
7135                         ret = 1;
7136                         goto next;
7137                 }
7138
7139                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7140                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7141
7142                 if (search_end == (u64)-1) {
7143                         search_end = key.offset - ext_offset +
7144                                 btrfs_file_extent_ram_bytes(leaf, fi);
7145                 }
7146
7147                 if (!extent_locked) {
7148                         lock_start = key.offset;
7149                         lock_end = lock_start + num_bytes - 1;
7150                 } else {
7151                         if (lock_start > key.offset ||
7152                             lock_end + 1 < key.offset + num_bytes) {
7153                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7154                                               lock_start, lock_end, GFP_NOFS);
7155                                 extent_locked = 0;
7156                         }
7157                 }
7158
7159                 if (!inode) {
7160                         btrfs_release_path(root, path);
7161
7162                         inode = btrfs_iget_locked(root->fs_info->sb,
7163                                                   key.objectid, root);
7164                         if (inode->i_state & I_NEW) {
7165                                 BTRFS_I(inode)->root = root;
7166                                 BTRFS_I(inode)->location.objectid =
7167                                         key.objectid;
7168                                 BTRFS_I(inode)->location.type =
7169                                         BTRFS_INODE_ITEM_KEY;
7170                                 BTRFS_I(inode)->location.offset = 0;
7171                                 btrfs_read_locked_inode(inode);
7172                                 unlock_new_inode(inode);
7173                         }
7174                         /*
7175                          * some code call btrfs_commit_transaction while
7176                          * holding the i_mutex, so we can't use mutex_lock
7177                          * here.
7178                          */
7179                         if (is_bad_inode(inode) ||
7180                             !mutex_trylock(&inode->i_mutex)) {
7181                                 iput(inode);
7182                                 inode = NULL;
7183                                 key.offset = (u64)-1;
7184                                 goto skip;
7185                         }
7186                 }
7187
7188                 if (!extent_locked) {
7189                         struct btrfs_ordered_extent *ordered;
7190
7191                         btrfs_release_path(root, path);
7192
7193                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7194                                     lock_end, GFP_NOFS);
7195                         ordered = btrfs_lookup_first_ordered_extent(inode,
7196                                                                     lock_end);
7197                         if (ordered &&
7198                             ordered->file_offset <= lock_end &&
7199                             ordered->file_offset + ordered->len > lock_start) {
7200                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7201                                               lock_start, lock_end, GFP_NOFS);
7202                                 btrfs_start_ordered_extent(inode, ordered, 1);
7203                                 btrfs_put_ordered_extent(ordered);
7204                                 key.offset += num_bytes;
7205                                 goto skip;
7206                         }
7207                         if (ordered)
7208                                 btrfs_put_ordered_extent(ordered);
7209
7210                         extent_locked = 1;
7211                         continue;
7212                 }
7213
7214                 if (nr_extents == 1) {
7215                         /* update extent pointer in place */
7216                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7217                                                 new_extents[0].disk_bytenr);
7218                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7219                                                 new_extents[0].disk_num_bytes);
7220                         btrfs_mark_buffer_dirty(leaf);
7221
7222                         btrfs_drop_extent_cache(inode, key.offset,
7223                                                 key.offset + num_bytes - 1, 0);
7224
7225                         ret = btrfs_inc_extent_ref(trans, root,
7226                                                 new_extents[0].disk_bytenr,
7227                                                 new_extents[0].disk_num_bytes,
7228                                                 leaf->start,
7229                                                 root->root_key.objectid,
7230                                                 trans->transid,
7231                                                 key.objectid);
7232                         BUG_ON(ret);
7233
7234                         ret = btrfs_free_extent(trans, root,
7235                                                 extent_key->objectid,
7236                                                 extent_key->offset,
7237                                                 leaf->start,
7238                                                 btrfs_header_owner(leaf),
7239                                                 btrfs_header_generation(leaf),
7240                                                 key.objectid, 0);
7241                         BUG_ON(ret);
7242
7243                         btrfs_release_path(root, path);
7244                         key.offset += num_bytes;
7245                 } else {
7246                         BUG_ON(1);
7247 #if 0
7248                         u64 alloc_hint;
7249                         u64 extent_len;
7250                         int i;
7251                         /*
7252                          * drop old extent pointer at first, then insert the
7253                          * new pointers one bye one
7254                          */
7255                         btrfs_release_path(root, path);
7256                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7257                                                  key.offset + num_bytes,
7258                                                  key.offset, &alloc_hint);
7259                         BUG_ON(ret);
7260
7261                         for (i = 0; i < nr_extents; i++) {
7262                                 if (ext_offset >= new_extents[i].num_bytes) {
7263                                         ext_offset -= new_extents[i].num_bytes;
7264                                         continue;
7265                                 }
7266                                 extent_len = min(new_extents[i].num_bytes -
7267                                                  ext_offset, num_bytes);
7268
7269                                 ret = btrfs_insert_empty_item(trans, root,
7270                                                               path, &key,
7271                                                               sizeof(*fi));
7272                                 BUG_ON(ret);
7273
7274                                 leaf = path->nodes[0];
7275                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7276                                                 struct btrfs_file_extent_item);
7277                                 btrfs_set_file_extent_generation(leaf, fi,
7278                                                         trans->transid);
7279                                 btrfs_set_file_extent_type(leaf, fi,
7280                                                         BTRFS_FILE_EXTENT_REG);
7281                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7282                                                 new_extents[i].disk_bytenr);
7283                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7284                                                 new_extents[i].disk_num_bytes);
7285                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7286                                                 new_extents[i].ram_bytes);
7287
7288                                 btrfs_set_file_extent_compression(leaf, fi,
7289                                                 new_extents[i].compression);
7290                                 btrfs_set_file_extent_encryption(leaf, fi,
7291                                                 new_extents[i].encryption);
7292                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7293                                                 new_extents[i].other_encoding);
7294
7295                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7296                                                         extent_len);
7297                                 ext_offset += new_extents[i].offset;
7298                                 btrfs_set_file_extent_offset(leaf, fi,
7299                                                         ext_offset);
7300                                 btrfs_mark_buffer_dirty(leaf);
7301
7302                                 btrfs_drop_extent_cache(inode, key.offset,
7303                                                 key.offset + extent_len - 1, 0);
7304
7305                                 ret = btrfs_inc_extent_ref(trans, root,
7306                                                 new_extents[i].disk_bytenr,
7307                                                 new_extents[i].disk_num_bytes,
7308                                                 leaf->start,
7309                                                 root->root_key.objectid,
7310                                                 trans->transid, key.objectid);
7311                                 BUG_ON(ret);
7312                                 btrfs_release_path(root, path);
7313
7314                                 inode_add_bytes(inode, extent_len);
7315
7316                                 ext_offset = 0;
7317                                 num_bytes -= extent_len;
7318                                 key.offset += extent_len;
7319
7320                                 if (num_bytes == 0)
7321                                         break;
7322                         }
7323                         BUG_ON(i >= nr_extents);
7324 #endif
7325                 }
7326
7327                 if (extent_locked) {
7328                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7329                                       lock_end, GFP_NOFS);
7330                         extent_locked = 0;
7331                 }
7332 skip:
7333                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7334                     key.offset >= search_end)
7335                         break;
7336
7337                 cond_resched();
7338         }
7339         ret = 0;
7340 out:
7341         btrfs_release_path(root, path);
7342         if (inode) {
7343                 mutex_unlock(&inode->i_mutex);
7344                 if (extent_locked) {
7345                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7346                                       lock_end, GFP_NOFS);
7347                 }
7348                 iput(inode);
7349         }
7350         return ret;
7351 }
7352
7353 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7354                                struct btrfs_root *root,
7355                                struct extent_buffer *buf, u64 orig_start)
7356 {
7357         int level;
7358         int ret;
7359
7360         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7361         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7362
7363         level = btrfs_header_level(buf);
7364         if (level == 0) {
7365                 struct btrfs_leaf_ref *ref;
7366                 struct btrfs_leaf_ref *orig_ref;
7367
7368                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7369                 if (!orig_ref)
7370                         return -ENOENT;
7371
7372                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7373                 if (!ref) {
7374                         btrfs_free_leaf_ref(root, orig_ref);
7375                         return -ENOMEM;
7376                 }
7377
7378                 ref->nritems = orig_ref->nritems;
7379                 memcpy(ref->extents, orig_ref->extents,
7380                         sizeof(ref->extents[0]) * ref->nritems);
7381
7382                 btrfs_free_leaf_ref(root, orig_ref);
7383
7384                 ref->root_gen = trans->transid;
7385                 ref->bytenr = buf->start;
7386                 ref->owner = btrfs_header_owner(buf);
7387                 ref->generation = btrfs_header_generation(buf);
7388
7389                 ret = btrfs_add_leaf_ref(root, ref, 0);
7390                 WARN_ON(ret);
7391                 btrfs_free_leaf_ref(root, ref);
7392         }
7393         return 0;
7394 }
7395
7396 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7397                                         struct extent_buffer *leaf,
7398                                         struct btrfs_block_group_cache *group,
7399                                         struct btrfs_root *target_root)
7400 {
7401         struct btrfs_key key;
7402         struct inode *inode = NULL;
7403         struct btrfs_file_extent_item *fi;
7404         struct extent_state *cached_state = NULL;
7405         u64 num_bytes;
7406         u64 skip_objectid = 0;
7407         u32 nritems;
7408         u32 i;
7409
7410         nritems = btrfs_header_nritems(leaf);
7411         for (i = 0; i < nritems; i++) {
7412                 btrfs_item_key_to_cpu(leaf, &key, i);
7413                 if (key.objectid == skip_objectid ||
7414                     key.type != BTRFS_EXTENT_DATA_KEY)
7415                         continue;
7416                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7417                 if (btrfs_file_extent_type(leaf, fi) ==
7418                     BTRFS_FILE_EXTENT_INLINE)
7419                         continue;
7420                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7421                         continue;
7422                 if (!inode || inode->i_ino != key.objectid) {
7423                         iput(inode);
7424                         inode = btrfs_ilookup(target_root->fs_info->sb,
7425                                               key.objectid, target_root, 1);
7426                 }
7427                 if (!inode) {
7428                         skip_objectid = key.objectid;
7429                         continue;
7430                 }
7431                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7432
7433                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7434                                  key.offset + num_bytes - 1, 0, &cached_state,
7435                                  GFP_NOFS);
7436                 btrfs_drop_extent_cache(inode, key.offset,
7437                                         key.offset + num_bytes - 1, 1);
7438                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7439                                      key.offset + num_bytes - 1, &cached_state,
7440                                      GFP_NOFS);
7441                 cond_resched();
7442         }
7443         iput(inode);
7444         return 0;
7445 }
7446
7447 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7448                                         struct btrfs_root *root,
7449                                         struct extent_buffer *leaf,
7450                                         struct btrfs_block_group_cache *group,
7451                                         struct inode *reloc_inode)
7452 {
7453         struct btrfs_key key;
7454         struct btrfs_key extent_key;
7455         struct btrfs_file_extent_item *fi;
7456         struct btrfs_leaf_ref *ref;
7457         struct disk_extent *new_extent;
7458         u64 bytenr;
7459         u64 num_bytes;
7460         u32 nritems;
7461         u32 i;
7462         int ext_index;
7463         int nr_extent;
7464         int ret;
7465
7466         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7467         if (!new_extent)
7468                 return -ENOMEM;
7469
7470         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7471         BUG_ON(!ref);
7472
7473         ext_index = -1;
7474         nritems = btrfs_header_nritems(leaf);
7475         for (i = 0; i < nritems; i++) {
7476                 btrfs_item_key_to_cpu(leaf, &key, i);
7477                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7478                         continue;
7479                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7480                 if (btrfs_file_extent_type(leaf, fi) ==
7481                     BTRFS_FILE_EXTENT_INLINE)
7482                         continue;
7483                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7484                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7485                 if (bytenr == 0)
7486                         continue;
7487
7488                 ext_index++;
7489                 if (bytenr >= group->key.objectid + group->key.offset ||
7490                     bytenr + num_bytes <= group->key.objectid)
7491                         continue;
7492
7493                 extent_key.objectid = bytenr;
7494                 extent_key.offset = num_bytes;
7495                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7496                 nr_extent = 1;
7497                 ret = get_new_locations(reloc_inode, &extent_key,
7498                                         group->key.objectid, 1,
7499                                         &new_extent, &nr_extent);
7500                 if (ret > 0)
7501                         continue;
7502                 BUG_ON(ret < 0);
7503
7504                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7505                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7506                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7507                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7508
7509                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7510                                                 new_extent->disk_bytenr);
7511                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7512                                                 new_extent->disk_num_bytes);
7513                 btrfs_mark_buffer_dirty(leaf);
7514
7515                 ret = btrfs_inc_extent_ref(trans, root,
7516                                         new_extent->disk_bytenr,
7517                                         new_extent->disk_num_bytes,
7518                                         leaf->start,
7519                                         root->root_key.objectid,
7520                                         trans->transid, key.objectid);
7521                 BUG_ON(ret);
7522
7523                 ret = btrfs_free_extent(trans, root,
7524                                         bytenr, num_bytes, leaf->start,
7525                                         btrfs_header_owner(leaf),
7526                                         btrfs_header_generation(leaf),
7527                                         key.objectid, 0);
7528                 BUG_ON(ret);
7529                 cond_resched();
7530         }
7531         kfree(new_extent);
7532         BUG_ON(ext_index + 1 != ref->nritems);
7533         btrfs_free_leaf_ref(root, ref);
7534         return 0;
7535 }
7536
7537 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7538                           struct btrfs_root *root)
7539 {
7540         struct btrfs_root *reloc_root;
7541         int ret;
7542
7543         if (root->reloc_root) {
7544                 reloc_root = root->reloc_root;
7545                 root->reloc_root = NULL;
7546                 list_add(&reloc_root->dead_list,
7547                          &root->fs_info->dead_reloc_roots);
7548
7549                 btrfs_set_root_bytenr(&reloc_root->root_item,
7550                                       reloc_root->node->start);
7551                 btrfs_set_root_level(&root->root_item,
7552                                      btrfs_header_level(reloc_root->node));
7553                 memset(&reloc_root->root_item.drop_progress, 0,
7554                         sizeof(struct btrfs_disk_key));
7555                 reloc_root->root_item.drop_level = 0;
7556
7557                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7558                                         &reloc_root->root_key,
7559                                         &reloc_root->root_item);
7560                 BUG_ON(ret);
7561         }
7562         return 0;
7563 }
7564
7565 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7566 {
7567         struct btrfs_trans_handle *trans;
7568         struct btrfs_root *reloc_root;
7569         struct btrfs_root *prev_root = NULL;
7570         struct list_head dead_roots;
7571         int ret;
7572         unsigned long nr;
7573
7574         INIT_LIST_HEAD(&dead_roots);
7575         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7576
7577         while (!list_empty(&dead_roots)) {
7578                 reloc_root = list_entry(dead_roots.prev,
7579                                         struct btrfs_root, dead_list);
7580                 list_del_init(&reloc_root->dead_list);
7581
7582                 BUG_ON(reloc_root->commit_root != NULL);
7583                 while (1) {
7584                         trans = btrfs_join_transaction(root, 1);
7585                         BUG_ON(IS_ERR(trans));
7586
7587                         mutex_lock(&root->fs_info->drop_mutex);
7588                         ret = btrfs_drop_snapshot(trans, reloc_root);
7589                         if (ret != -EAGAIN)
7590                                 break;
7591                         mutex_unlock(&root->fs_info->drop_mutex);
7592
7593                         nr = trans->blocks_used;
7594                         ret = btrfs_end_transaction(trans, root);
7595                         BUG_ON(ret);
7596                         btrfs_btree_balance_dirty(root, nr);
7597                 }
7598
7599                 free_extent_buffer(reloc_root->node);
7600
7601                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7602                                      &reloc_root->root_key);
7603                 BUG_ON(ret);
7604                 mutex_unlock(&root->fs_info->drop_mutex);
7605
7606                 nr = trans->blocks_used;
7607                 ret = btrfs_end_transaction(trans, root);
7608                 BUG_ON(ret);
7609                 btrfs_btree_balance_dirty(root, nr);
7610
7611                 kfree(prev_root);
7612                 prev_root = reloc_root;
7613         }
7614         if (prev_root) {
7615                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7616                 kfree(prev_root);
7617         }
7618         return 0;
7619 }
7620
7621 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7622 {
7623         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7624         return 0;
7625 }
7626
7627 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7628 {
7629         struct btrfs_root *reloc_root;
7630         struct btrfs_trans_handle *trans;
7631         struct btrfs_key location;
7632         int found;
7633         int ret;
7634
7635         mutex_lock(&root->fs_info->tree_reloc_mutex);
7636         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7637         BUG_ON(ret);
7638         found = !list_empty(&root->fs_info->dead_reloc_roots);
7639         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7640
7641         if (found) {
7642                 trans = btrfs_start_transaction(root, 1);
7643                 BUG_ON(IS_ERR(trans));
7644                 ret = btrfs_commit_transaction(trans, root);
7645                 BUG_ON(ret);
7646         }
7647
7648         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7649         location.offset = (u64)-1;
7650         location.type = BTRFS_ROOT_ITEM_KEY;
7651
7652         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7653         BUG_ON(!reloc_root);
7654         ret = btrfs_orphan_cleanup(reloc_root);
7655         BUG_ON(ret);
7656         return 0;
7657 }
7658
7659 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7660                                     struct btrfs_root *root)
7661 {
7662         struct btrfs_root *reloc_root;
7663         struct extent_buffer *eb;
7664         struct btrfs_root_item *root_item;
7665         struct btrfs_key root_key;
7666         int ret;
7667
7668         BUG_ON(!root->ref_cows);
7669         if (root->reloc_root)
7670                 return 0;
7671
7672         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7673         if (!root_item)
7674                 return -ENOMEM;
7675
7676         ret = btrfs_copy_root(trans, root, root->commit_root,
7677                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7678         BUG_ON(ret);
7679
7680         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7681         root_key.offset = root->root_key.objectid;
7682         root_key.type = BTRFS_ROOT_ITEM_KEY;
7683
7684         memcpy(root_item, &root->root_item, sizeof(root_item));
7685         btrfs_set_root_refs(root_item, 0);
7686         btrfs_set_root_bytenr(root_item, eb->start);
7687         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7688         btrfs_set_root_generation(root_item, trans->transid);
7689
7690         btrfs_tree_unlock(eb);
7691         free_extent_buffer(eb);
7692
7693         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7694                                 &root_key, root_item);
7695         BUG_ON(ret);
7696         kfree(root_item);
7697
7698         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7699                                                  &root_key);
7700         BUG_ON(IS_ERR(reloc_root));
7701         reloc_root->last_trans = trans->transid;
7702         reloc_root->commit_root = NULL;
7703         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7704
7705         root->reloc_root = reloc_root;
7706         return 0;
7707 }
7708
7709 /*
7710  * Core function of space balance.
7711  *
7712  * The idea is using reloc trees to relocate tree blocks in reference
7713  * counted roots. There is one reloc tree for each subvol, and all
7714  * reloc trees share same root key objectid. Reloc trees are snapshots
7715  * of the latest committed roots of subvols (root->commit_root).
7716  *
7717  * To relocate a tree block referenced by a subvol, there are two steps.
7718  * COW the block through subvol's reloc tree, then update block pointer
7719  * in the subvol to point to the new block. Since all reloc trees share
7720  * same root key objectid, doing special handing for tree blocks owned
7721  * by them is easy. Once a tree block has been COWed in one reloc tree,
7722  * we can use the resulting new block directly when the same block is
7723  * required to COW again through other reloc trees. By this way, relocated
7724  * tree blocks are shared between reloc trees, so they are also shared
7725  * between subvols.
7726  */
7727 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7728                                       struct btrfs_root *root,
7729                                       struct btrfs_path *path,
7730                                       struct btrfs_key *first_key,
7731                                       struct btrfs_ref_path *ref_path,
7732                                       struct btrfs_block_group_cache *group,
7733                                       struct inode *reloc_inode)
7734 {
7735         struct btrfs_root *reloc_root;
7736         struct extent_buffer *eb = NULL;
7737         struct btrfs_key *keys;
7738         u64 *nodes;
7739         int level;
7740         int shared_level;
7741         int lowest_level = 0;
7742         int ret;
7743
7744         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7745                 lowest_level = ref_path->owner_objectid;
7746
7747         if (!root->ref_cows) {
7748                 path->lowest_level = lowest_level;
7749                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7750                 BUG_ON(ret < 0);
7751                 path->lowest_level = 0;
7752                 btrfs_release_path(root, path);
7753                 return 0;
7754         }
7755
7756         mutex_lock(&root->fs_info->tree_reloc_mutex);
7757         ret = init_reloc_tree(trans, root);
7758         BUG_ON(ret);
7759         reloc_root = root->reloc_root;
7760
7761         shared_level = ref_path->shared_level;
7762         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7763
7764         keys = ref_path->node_keys;
7765         nodes = ref_path->new_nodes;
7766         memset(&keys[shared_level + 1], 0,
7767                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7768         memset(&nodes[shared_level + 1], 0,
7769                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7770
7771         if (nodes[lowest_level] == 0) {
7772                 path->lowest_level = lowest_level;
7773                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7774                                         0, 1);
7775                 BUG_ON(ret);
7776                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7777                         eb = path->nodes[level];
7778                         if (!eb || eb == reloc_root->node)
7779                                 break;
7780                         nodes[level] = eb->start;
7781                         if (level == 0)
7782                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7783                         else
7784                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7785                 }
7786                 if (nodes[0] &&
7787                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7788                         eb = path->nodes[0];
7789                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7790                                                       group, reloc_inode);
7791                         BUG_ON(ret);
7792                 }
7793                 btrfs_release_path(reloc_root, path);
7794         } else {
7795                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7796                                        lowest_level);
7797                 BUG_ON(ret);
7798         }
7799
7800         /*
7801          * replace tree blocks in the fs tree with tree blocks in
7802          * the reloc tree.
7803          */
7804         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7805         BUG_ON(ret < 0);
7806
7807         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7808                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7809                                         0, 0);
7810                 BUG_ON(ret);
7811                 extent_buffer_get(path->nodes[0]);
7812                 eb = path->nodes[0];
7813                 btrfs_release_path(reloc_root, path);
7814                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7815                 BUG_ON(ret);
7816                 free_extent_buffer(eb);
7817         }
7818
7819         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7820         path->lowest_level = 0;
7821         return 0;
7822 }
7823
7824 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7825                                         struct btrfs_root *root,
7826                                         struct btrfs_path *path,
7827                                         struct btrfs_key *first_key,
7828                                         struct btrfs_ref_path *ref_path)
7829 {
7830         int ret;
7831
7832         ret = relocate_one_path(trans, root, path, first_key,
7833                                 ref_path, NULL, NULL);
7834         BUG_ON(ret);
7835
7836         return 0;
7837 }
7838
7839 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7840                                     struct btrfs_root *extent_root,
7841                                     struct btrfs_path *path,
7842                                     struct btrfs_key *extent_key)
7843 {
7844         int ret;
7845
7846         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7847         if (ret)
7848                 goto out;
7849         ret = btrfs_del_item(trans, extent_root, path);
7850 out:
7851         btrfs_release_path(extent_root, path);
7852         return ret;
7853 }
7854
7855 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7856                                                 struct btrfs_ref_path *ref_path)
7857 {
7858         struct btrfs_key root_key;
7859
7860         root_key.objectid = ref_path->root_objectid;
7861         root_key.type = BTRFS_ROOT_ITEM_KEY;
7862         if (is_cowonly_root(ref_path->root_objectid))
7863                 root_key.offset = 0;
7864         else
7865                 root_key.offset = (u64)-1;
7866
7867         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7868 }
7869
7870 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7871                                         struct btrfs_path *path,
7872                                         struct btrfs_key *extent_key,
7873                                         struct btrfs_block_group_cache *group,
7874                                         struct inode *reloc_inode, int pass)
7875 {
7876         struct btrfs_trans_handle *trans;
7877         struct btrfs_root *found_root;
7878         struct btrfs_ref_path *ref_path = NULL;
7879         struct disk_extent *new_extents = NULL;
7880         int nr_extents = 0;
7881         int loops;
7882         int ret;
7883         int level;
7884         struct btrfs_key first_key;
7885         u64 prev_block = 0;
7886
7887
7888         trans = btrfs_start_transaction(extent_root, 1);
7889         BUG_ON(IS_ERR(trans));
7890
7891         if (extent_key->objectid == 0) {
7892                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7893                 goto out;
7894         }
7895
7896         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7897         if (!ref_path) {
7898                 ret = -ENOMEM;
7899                 goto out;
7900         }
7901
7902         for (loops = 0; ; loops++) {
7903                 if (loops == 0) {
7904                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7905                                                    extent_key->objectid);
7906                 } else {
7907                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7908                 }
7909                 if (ret < 0)
7910                         goto out;
7911                 if (ret > 0)
7912                         break;
7913
7914                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7915                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7916                         continue;
7917
7918                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7919                 BUG_ON(!found_root);
7920                 /*
7921                  * for reference counted tree, only process reference paths
7922                  * rooted at the latest committed root.
7923                  */
7924                 if (found_root->ref_cows &&
7925                     ref_path->root_generation != found_root->root_key.offset)
7926                         continue;
7927
7928                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7929                         if (pass == 0) {
7930                                 /*
7931                                  * copy data extents to new locations
7932                                  */
7933                                 u64 group_start = group->key.objectid;
7934                                 ret = relocate_data_extent(reloc_inode,
7935                                                            extent_key,
7936                                                            group_start);
7937                                 if (ret < 0)
7938                                         goto out;
7939                                 break;
7940                         }
7941                         level = 0;
7942                 } else {
7943                         level = ref_path->owner_objectid;
7944                 }
7945
7946                 if (prev_block != ref_path->nodes[level]) {
7947                         struct extent_buffer *eb;
7948                         u64 block_start = ref_path->nodes[level];
7949                         u64 block_size = btrfs_level_size(found_root, level);
7950
7951                         eb = read_tree_block(found_root, block_start,
7952                                              block_size, 0);
7953                         if (!eb) {
7954                                 ret = -EIO;
7955                                 goto out;
7956                         }
7957                         btrfs_tree_lock(eb);
7958                         BUG_ON(level != btrfs_header_level(eb));
7959
7960                         if (level == 0)
7961                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7962                         else
7963                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7964
7965                         btrfs_tree_unlock(eb);
7966                         free_extent_buffer(eb);
7967                         prev_block = block_start;
7968                 }
7969
7970                 mutex_lock(&extent_root->fs_info->trans_mutex);
7971                 btrfs_record_root_in_trans(found_root);
7972                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7973                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7974                         /*
7975                          * try to update data extent references while
7976                          * keeping metadata shared between snapshots.
7977                          */
7978                         if (pass == 1) {
7979                                 ret = relocate_one_path(trans, found_root,
7980                                                 path, &first_key, ref_path,
7981                                                 group, reloc_inode);
7982                                 if (ret < 0)
7983                                         goto out;
7984                                 continue;
7985                         }
7986                         /*
7987                          * use fallback method to process the remaining
7988                          * references.
7989                          */
7990                         if (!new_extents) {
7991                                 u64 group_start = group->key.objectid;
7992                                 new_extents = kmalloc(sizeof(*new_extents),
7993                                                       GFP_NOFS);
7994                                 nr_extents = 1;
7995                                 ret = get_new_locations(reloc_inode,
7996                                                         extent_key,
7997                                                         group_start, 1,
7998                                                         &new_extents,
7999                                                         &nr_extents);
8000                                 if (ret)
8001                                         goto out;
8002                         }
8003                         ret = replace_one_extent(trans, found_root,
8004                                                 path, extent_key,
8005                                                 &first_key, ref_path,
8006                                                 new_extents, nr_extents);
8007                 } else {
8008                         ret = relocate_tree_block(trans, found_root, path,
8009                                                   &first_key, ref_path);
8010                 }
8011                 if (ret < 0)
8012                         goto out;
8013         }
8014         ret = 0;
8015 out:
8016         btrfs_end_transaction(trans, extent_root);
8017         kfree(new_extents);
8018         kfree(ref_path);
8019         return ret;
8020 }
8021 #endif
8022
8023 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8024 {
8025         u64 num_devices;
8026         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8027                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8028
8029         /*
8030          * we add in the count of missing devices because we want
8031          * to make sure that any RAID levels on a degraded FS
8032          * continue to be honored.
8033          */
8034         num_devices = root->fs_info->fs_devices->rw_devices +
8035                 root->fs_info->fs_devices->missing_devices;
8036
8037         if (num_devices == 1) {
8038                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8039                 stripped = flags & ~stripped;
8040
8041                 /* turn raid0 into single device chunks */
8042                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8043                         return stripped;
8044
8045                 /* turn mirroring into duplication */
8046                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8047                              BTRFS_BLOCK_GROUP_RAID10))
8048                         return stripped | BTRFS_BLOCK_GROUP_DUP;
8049                 return flags;
8050         } else {
8051                 /* they already had raid on here, just return */
8052                 if (flags & stripped)
8053                         return flags;
8054
8055                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8056                 stripped = flags & ~stripped;
8057
8058                 /* switch duplicated blocks with raid1 */
8059                 if (flags & BTRFS_BLOCK_GROUP_DUP)
8060                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
8061
8062                 /* turn single device chunks into raid0 */
8063                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8064         }
8065         return flags;
8066 }
8067
8068 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8069 {
8070         struct btrfs_space_info *sinfo = cache->space_info;
8071         u64 num_bytes;
8072         int ret = -ENOSPC;
8073
8074         if (cache->ro)
8075                 return 0;
8076
8077         spin_lock(&sinfo->lock);
8078         spin_lock(&cache->lock);
8079         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8080                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8081
8082         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8083             sinfo->bytes_may_use + sinfo->bytes_readonly +
8084             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8085                 sinfo->bytes_readonly += num_bytes;
8086                 sinfo->bytes_reserved += cache->reserved_pinned;
8087                 cache->reserved_pinned = 0;
8088                 cache->ro = 1;
8089                 ret = 0;
8090         }
8091
8092         spin_unlock(&cache->lock);
8093         spin_unlock(&sinfo->lock);
8094         return ret;
8095 }
8096
8097 int btrfs_set_block_group_ro(struct btrfs_root *root,
8098                              struct btrfs_block_group_cache *cache)
8099
8100 {
8101         struct btrfs_trans_handle *trans;
8102         u64 alloc_flags;
8103         int ret;
8104
8105         BUG_ON(cache->ro);
8106
8107         trans = btrfs_join_transaction(root, 1);
8108         BUG_ON(IS_ERR(trans));
8109
8110         alloc_flags = update_block_group_flags(root, cache->flags);
8111         if (alloc_flags != cache->flags)
8112                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8113
8114         ret = set_block_group_ro(cache);
8115         if (!ret)
8116                 goto out;
8117         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8118         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8119         if (ret < 0)
8120                 goto out;
8121         ret = set_block_group_ro(cache);
8122 out:
8123         btrfs_end_transaction(trans, root);
8124         return ret;
8125 }
8126
8127 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8128                             struct btrfs_root *root, u64 type)
8129 {
8130         u64 alloc_flags = get_alloc_profile(root, type);
8131         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8132 }
8133
8134 /*
8135  * helper to account the unused space of all the readonly block group in the
8136  * list. takes mirrors into account.
8137  */
8138 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8139 {
8140         struct btrfs_block_group_cache *block_group;
8141         u64 free_bytes = 0;
8142         int factor;
8143
8144         list_for_each_entry(block_group, groups_list, list) {
8145                 spin_lock(&block_group->lock);
8146
8147                 if (!block_group->ro) {
8148                         spin_unlock(&block_group->lock);
8149                         continue;
8150                 }
8151
8152                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8153                                           BTRFS_BLOCK_GROUP_RAID10 |
8154                                           BTRFS_BLOCK_GROUP_DUP))
8155                         factor = 2;
8156                 else
8157                         factor = 1;
8158
8159                 free_bytes += (block_group->key.offset -
8160                                btrfs_block_group_used(&block_group->item)) *
8161                                factor;
8162
8163                 spin_unlock(&block_group->lock);
8164         }
8165
8166         return free_bytes;
8167 }
8168
8169 /*
8170  * helper to account the unused space of all the readonly block group in the
8171  * space_info. takes mirrors into account.
8172  */
8173 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8174 {
8175         int i;
8176         u64 free_bytes = 0;
8177
8178         spin_lock(&sinfo->lock);
8179
8180         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8181                 if (!list_empty(&sinfo->block_groups[i]))
8182                         free_bytes += __btrfs_get_ro_block_group_free_space(
8183                                                 &sinfo->block_groups[i]);
8184
8185         spin_unlock(&sinfo->lock);
8186
8187         return free_bytes;
8188 }
8189
8190 int btrfs_set_block_group_rw(struct btrfs_root *root,
8191                               struct btrfs_block_group_cache *cache)
8192 {
8193         struct btrfs_space_info *sinfo = cache->space_info;
8194         u64 num_bytes;
8195
8196         BUG_ON(!cache->ro);
8197
8198         spin_lock(&sinfo->lock);
8199         spin_lock(&cache->lock);
8200         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8201                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8202         sinfo->bytes_readonly -= num_bytes;
8203         cache->ro = 0;
8204         spin_unlock(&cache->lock);
8205         spin_unlock(&sinfo->lock);
8206         return 0;
8207 }
8208
8209 /*
8210  * checks to see if its even possible to relocate this block group.
8211  *
8212  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8213  * ok to go ahead and try.
8214  */
8215 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8216 {
8217         struct btrfs_block_group_cache *block_group;
8218         struct btrfs_space_info *space_info;
8219         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8220         struct btrfs_device *device;
8221         int full = 0;
8222         int ret = 0;
8223
8224         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8225
8226         /* odd, couldn't find the block group, leave it alone */
8227         if (!block_group)
8228                 return -1;
8229
8230         /* no bytes used, we're good */
8231         if (!btrfs_block_group_used(&block_group->item))
8232                 goto out;
8233
8234         space_info = block_group->space_info;
8235         spin_lock(&space_info->lock);
8236
8237         full = space_info->full;
8238
8239         /*
8240          * if this is the last block group we have in this space, we can't
8241          * relocate it unless we're able to allocate a new chunk below.
8242          *
8243          * Otherwise, we need to make sure we have room in the space to handle
8244          * all of the extents from this block group.  If we can, we're good
8245          */
8246         if ((space_info->total_bytes != block_group->key.offset) &&
8247            (space_info->bytes_used + space_info->bytes_reserved +
8248             space_info->bytes_pinned + space_info->bytes_readonly +
8249             btrfs_block_group_used(&block_group->item) <
8250             space_info->total_bytes)) {
8251                 spin_unlock(&space_info->lock);
8252                 goto out;
8253         }
8254         spin_unlock(&space_info->lock);
8255
8256         /*
8257          * ok we don't have enough space, but maybe we have free space on our
8258          * devices to allocate new chunks for relocation, so loop through our
8259          * alloc devices and guess if we have enough space.  However, if we
8260          * were marked as full, then we know there aren't enough chunks, and we
8261          * can just return.
8262          */
8263         ret = -1;
8264         if (full)
8265                 goto out;
8266
8267         mutex_lock(&root->fs_info->chunk_mutex);
8268         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8269                 u64 min_free = btrfs_block_group_used(&block_group->item);
8270                 u64 dev_offset;
8271
8272                 /*
8273                  * check to make sure we can actually find a chunk with enough
8274                  * space to fit our block group in.
8275                  */
8276                 if (device->total_bytes > device->bytes_used + min_free) {
8277                         ret = find_free_dev_extent(NULL, device, min_free,
8278                                                    &dev_offset, NULL);
8279                         if (!ret)
8280                                 break;
8281                         ret = -1;
8282                 }
8283         }
8284         mutex_unlock(&root->fs_info->chunk_mutex);
8285 out:
8286         btrfs_put_block_group(block_group);
8287         return ret;
8288 }
8289
8290 static int find_first_block_group(struct btrfs_root *root,
8291                 struct btrfs_path *path, struct btrfs_key *key)
8292 {
8293         int ret = 0;
8294         struct btrfs_key found_key;
8295         struct extent_buffer *leaf;
8296         int slot;
8297
8298         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8299         if (ret < 0)
8300                 goto out;
8301
8302         while (1) {
8303                 slot = path->slots[0];
8304                 leaf = path->nodes[0];
8305                 if (slot >= btrfs_header_nritems(leaf)) {
8306                         ret = btrfs_next_leaf(root, path);
8307                         if (ret == 0)
8308                                 continue;
8309                         if (ret < 0)
8310                                 goto out;
8311                         break;
8312                 }
8313                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8314
8315                 if (found_key.objectid >= key->objectid &&
8316                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8317                         ret = 0;
8318                         goto out;
8319                 }
8320                 path->slots[0]++;
8321         }
8322 out:
8323         return ret;
8324 }
8325
8326 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8327 {
8328         struct btrfs_block_group_cache *block_group;
8329         u64 last = 0;
8330
8331         while (1) {
8332                 struct inode *inode;
8333
8334                 block_group = btrfs_lookup_first_block_group(info, last);
8335                 while (block_group) {
8336                         spin_lock(&block_group->lock);
8337                         if (block_group->iref)
8338                                 break;
8339                         spin_unlock(&block_group->lock);
8340                         block_group = next_block_group(info->tree_root,
8341                                                        block_group);
8342                 }
8343                 if (!block_group) {
8344                         if (last == 0)
8345                                 break;
8346                         last = 0;
8347                         continue;
8348                 }
8349
8350                 inode = block_group->inode;
8351                 block_group->iref = 0;
8352                 block_group->inode = NULL;
8353                 spin_unlock(&block_group->lock);
8354                 iput(inode);
8355                 last = block_group->key.objectid + block_group->key.offset;
8356                 btrfs_put_block_group(block_group);
8357         }
8358 }
8359
8360 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8361 {
8362         struct btrfs_block_group_cache *block_group;
8363         struct btrfs_space_info *space_info;
8364         struct btrfs_caching_control *caching_ctl;
8365         struct rb_node *n;
8366
8367         down_write(&info->extent_commit_sem);
8368         while (!list_empty(&info->caching_block_groups)) {
8369                 caching_ctl = list_entry(info->caching_block_groups.next,
8370                                          struct btrfs_caching_control, list);
8371                 list_del(&caching_ctl->list);
8372                 put_caching_control(caching_ctl);
8373         }
8374         up_write(&info->extent_commit_sem);
8375
8376         spin_lock(&info->block_group_cache_lock);
8377         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8378                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8379                                        cache_node);
8380                 rb_erase(&block_group->cache_node,
8381                          &info->block_group_cache_tree);
8382                 spin_unlock(&info->block_group_cache_lock);
8383
8384                 down_write(&block_group->space_info->groups_sem);
8385                 list_del(&block_group->list);
8386                 up_write(&block_group->space_info->groups_sem);
8387
8388                 if (block_group->cached == BTRFS_CACHE_STARTED)
8389                         wait_block_group_cache_done(block_group);
8390
8391                 /*
8392                  * We haven't cached this block group, which means we could
8393                  * possibly have excluded extents on this block group.
8394                  */
8395                 if (block_group->cached == BTRFS_CACHE_NO)
8396                         free_excluded_extents(info->extent_root, block_group);
8397
8398                 btrfs_remove_free_space_cache(block_group);
8399                 btrfs_put_block_group(block_group);
8400
8401                 spin_lock(&info->block_group_cache_lock);
8402         }
8403         spin_unlock(&info->block_group_cache_lock);
8404
8405         /* now that all the block groups are freed, go through and
8406          * free all the space_info structs.  This is only called during
8407          * the final stages of unmount, and so we know nobody is
8408          * using them.  We call synchronize_rcu() once before we start,
8409          * just to be on the safe side.
8410          */
8411         synchronize_rcu();
8412
8413         release_global_block_rsv(info);
8414
8415         while(!list_empty(&info->space_info)) {
8416                 space_info = list_entry(info->space_info.next,
8417                                         struct btrfs_space_info,
8418                                         list);
8419                 if (space_info->bytes_pinned > 0 ||
8420                     space_info->bytes_reserved > 0) {
8421                         WARN_ON(1);
8422                         dump_space_info(space_info, 0, 0);
8423                 }
8424                 list_del(&space_info->list);
8425                 kfree(space_info);
8426         }
8427         return 0;
8428 }
8429
8430 static void __link_block_group(struct btrfs_space_info *space_info,
8431                                struct btrfs_block_group_cache *cache)
8432 {
8433         int index = get_block_group_index(cache);
8434
8435         down_write(&space_info->groups_sem);
8436         list_add_tail(&cache->list, &space_info->block_groups[index]);
8437         up_write(&space_info->groups_sem);
8438 }
8439
8440 int btrfs_read_block_groups(struct btrfs_root *root)
8441 {
8442         struct btrfs_path *path;
8443         int ret;
8444         struct btrfs_block_group_cache *cache;
8445         struct btrfs_fs_info *info = root->fs_info;
8446         struct btrfs_space_info *space_info;
8447         struct btrfs_key key;
8448         struct btrfs_key found_key;
8449         struct extent_buffer *leaf;
8450         int need_clear = 0;
8451         u64 cache_gen;
8452
8453         root = info->extent_root;
8454         key.objectid = 0;
8455         key.offset = 0;
8456         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8457         path = btrfs_alloc_path();
8458         if (!path)
8459                 return -ENOMEM;
8460
8461         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8462         if (cache_gen != 0 &&
8463             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8464                 need_clear = 1;
8465         if (btrfs_test_opt(root, CLEAR_CACHE))
8466                 need_clear = 1;
8467         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8468                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8469
8470         while (1) {
8471                 ret = find_first_block_group(root, path, &key);
8472                 if (ret > 0)
8473                         break;
8474                 if (ret != 0)
8475                         goto error;
8476                 leaf = path->nodes[0];
8477                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8478                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8479                 if (!cache) {
8480                         ret = -ENOMEM;
8481                         goto error;
8482                 }
8483
8484                 atomic_set(&cache->count, 1);
8485                 spin_lock_init(&cache->lock);
8486                 spin_lock_init(&cache->tree_lock);
8487                 cache->fs_info = info;
8488                 INIT_LIST_HEAD(&cache->list);
8489                 INIT_LIST_HEAD(&cache->cluster_list);
8490
8491                 if (need_clear)
8492                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8493
8494                 /*
8495                  * we only want to have 32k of ram per block group for keeping
8496                  * track of free space, and if we pass 1/2 of that we want to
8497                  * start converting things over to using bitmaps
8498                  */
8499                 cache->extents_thresh = ((1024 * 32) / 2) /
8500                         sizeof(struct btrfs_free_space);
8501
8502                 read_extent_buffer(leaf, &cache->item,
8503                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8504                                    sizeof(cache->item));
8505                 memcpy(&cache->key, &found_key, sizeof(found_key));
8506
8507                 key.objectid = found_key.objectid + found_key.offset;
8508                 btrfs_release_path(root, path);
8509                 cache->flags = btrfs_block_group_flags(&cache->item);
8510                 cache->sectorsize = root->sectorsize;
8511
8512                 /*
8513                  * We need to exclude the super stripes now so that the space
8514                  * info has super bytes accounted for, otherwise we'll think
8515                  * we have more space than we actually do.
8516                  */
8517                 exclude_super_stripes(root, cache);
8518
8519                 /*
8520                  * check for two cases, either we are full, and therefore
8521                  * don't need to bother with the caching work since we won't
8522                  * find any space, or we are empty, and we can just add all
8523                  * the space in and be done with it.  This saves us _alot_ of
8524                  * time, particularly in the full case.
8525                  */
8526                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8527                         cache->last_byte_to_unpin = (u64)-1;
8528                         cache->cached = BTRFS_CACHE_FINISHED;
8529                         free_excluded_extents(root, cache);
8530                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8531                         cache->last_byte_to_unpin = (u64)-1;
8532                         cache->cached = BTRFS_CACHE_FINISHED;
8533                         add_new_free_space(cache, root->fs_info,
8534                                            found_key.objectid,
8535                                            found_key.objectid +
8536                                            found_key.offset);
8537                         free_excluded_extents(root, cache);
8538                 }
8539
8540                 ret = update_space_info(info, cache->flags, found_key.offset,
8541                                         btrfs_block_group_used(&cache->item),
8542                                         &space_info);
8543                 BUG_ON(ret);
8544                 cache->space_info = space_info;
8545                 spin_lock(&cache->space_info->lock);
8546                 cache->space_info->bytes_readonly += cache->bytes_super;
8547                 spin_unlock(&cache->space_info->lock);
8548
8549                 __link_block_group(space_info, cache);
8550
8551                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8552                 BUG_ON(ret);
8553
8554                 set_avail_alloc_bits(root->fs_info, cache->flags);
8555                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8556                         set_block_group_ro(cache);
8557         }
8558
8559         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8560                 if (!(get_alloc_profile(root, space_info->flags) &
8561                       (BTRFS_BLOCK_GROUP_RAID10 |
8562                        BTRFS_BLOCK_GROUP_RAID1 |
8563                        BTRFS_BLOCK_GROUP_DUP)))
8564                         continue;
8565                 /*
8566                  * avoid allocating from un-mirrored block group if there are
8567                  * mirrored block groups.
8568                  */
8569                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8570                         set_block_group_ro(cache);
8571                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8572                         set_block_group_ro(cache);
8573         }
8574
8575         init_global_block_rsv(info);
8576         ret = 0;
8577 error:
8578         btrfs_free_path(path);
8579         return ret;
8580 }
8581
8582 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8583                            struct btrfs_root *root, u64 bytes_used,
8584                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8585                            u64 size)
8586 {
8587         int ret;
8588         struct btrfs_root *extent_root;
8589         struct btrfs_block_group_cache *cache;
8590
8591         extent_root = root->fs_info->extent_root;
8592
8593         root->fs_info->last_trans_log_full_commit = trans->transid;
8594
8595         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8596         if (!cache)
8597                 return -ENOMEM;
8598
8599         cache->key.objectid = chunk_offset;
8600         cache->key.offset = size;
8601         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8602         cache->sectorsize = root->sectorsize;
8603         cache->fs_info = root->fs_info;
8604
8605         /*
8606          * we only want to have 32k of ram per block group for keeping track
8607          * of free space, and if we pass 1/2 of that we want to start
8608          * converting things over to using bitmaps
8609          */
8610         cache->extents_thresh = ((1024 * 32) / 2) /
8611                 sizeof(struct btrfs_free_space);
8612         atomic_set(&cache->count, 1);
8613         spin_lock_init(&cache->lock);
8614         spin_lock_init(&cache->tree_lock);
8615         INIT_LIST_HEAD(&cache->list);
8616         INIT_LIST_HEAD(&cache->cluster_list);
8617
8618         btrfs_set_block_group_used(&cache->item, bytes_used);
8619         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8620         cache->flags = type;
8621         btrfs_set_block_group_flags(&cache->item, type);
8622
8623         cache->last_byte_to_unpin = (u64)-1;
8624         cache->cached = BTRFS_CACHE_FINISHED;
8625         exclude_super_stripes(root, cache);
8626
8627         add_new_free_space(cache, root->fs_info, chunk_offset,
8628                            chunk_offset + size);
8629
8630         free_excluded_extents(root, cache);
8631
8632         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8633                                 &cache->space_info);
8634         BUG_ON(ret);
8635
8636         spin_lock(&cache->space_info->lock);
8637         cache->space_info->bytes_readonly += cache->bytes_super;
8638         spin_unlock(&cache->space_info->lock);
8639
8640         __link_block_group(cache->space_info, cache);
8641
8642         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8643         BUG_ON(ret);
8644
8645         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8646                                 sizeof(cache->item));
8647         BUG_ON(ret);
8648
8649         set_avail_alloc_bits(extent_root->fs_info, type);
8650
8651         return 0;
8652 }
8653
8654 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8655                              struct btrfs_root *root, u64 group_start)
8656 {
8657         struct btrfs_path *path;
8658         struct btrfs_block_group_cache *block_group;
8659         struct btrfs_free_cluster *cluster;
8660         struct btrfs_root *tree_root = root->fs_info->tree_root;
8661         struct btrfs_key key;
8662         struct inode *inode;
8663         int ret;
8664         int factor;
8665
8666         root = root->fs_info->extent_root;
8667
8668         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8669         BUG_ON(!block_group);
8670         BUG_ON(!block_group->ro);
8671
8672         /*
8673          * Free the reserved super bytes from this block group before
8674          * remove it.
8675          */
8676         free_excluded_extents(root, block_group);
8677
8678         memcpy(&key, &block_group->key, sizeof(key));
8679         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8680                                   BTRFS_BLOCK_GROUP_RAID1 |
8681                                   BTRFS_BLOCK_GROUP_RAID10))
8682                 factor = 2;
8683         else
8684                 factor = 1;
8685
8686         /* make sure this block group isn't part of an allocation cluster */
8687         cluster = &root->fs_info->data_alloc_cluster;
8688         spin_lock(&cluster->refill_lock);
8689         btrfs_return_cluster_to_free_space(block_group, cluster);
8690         spin_unlock(&cluster->refill_lock);
8691
8692         /*
8693          * make sure this block group isn't part of a metadata
8694          * allocation cluster
8695          */
8696         cluster = &root->fs_info->meta_alloc_cluster;
8697         spin_lock(&cluster->refill_lock);
8698         btrfs_return_cluster_to_free_space(block_group, cluster);
8699         spin_unlock(&cluster->refill_lock);
8700
8701         path = btrfs_alloc_path();
8702         BUG_ON(!path);
8703
8704         inode = lookup_free_space_inode(root, block_group, path);
8705         if (!IS_ERR(inode)) {
8706                 btrfs_orphan_add(trans, inode);
8707                 clear_nlink(inode);
8708                 /* One for the block groups ref */
8709                 spin_lock(&block_group->lock);
8710                 if (block_group->iref) {
8711                         block_group->iref = 0;
8712                         block_group->inode = NULL;
8713                         spin_unlock(&block_group->lock);
8714                         iput(inode);
8715                 } else {
8716                         spin_unlock(&block_group->lock);
8717                 }
8718                 /* One for our lookup ref */
8719                 iput(inode);
8720         }
8721
8722         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8723         key.offset = block_group->key.objectid;
8724         key.type = 0;
8725
8726         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8727         if (ret < 0)
8728                 goto out;
8729         if (ret > 0)
8730                 btrfs_release_path(tree_root, path);
8731         if (ret == 0) {
8732                 ret = btrfs_del_item(trans, tree_root, path);
8733                 if (ret)
8734                         goto out;
8735                 btrfs_release_path(tree_root, path);
8736         }
8737
8738         spin_lock(&root->fs_info->block_group_cache_lock);
8739         rb_erase(&block_group->cache_node,
8740                  &root->fs_info->block_group_cache_tree);
8741         spin_unlock(&root->fs_info->block_group_cache_lock);
8742
8743         down_write(&block_group->space_info->groups_sem);
8744         /*
8745          * we must use list_del_init so people can check to see if they
8746          * are still on the list after taking the semaphore
8747          */
8748         list_del_init(&block_group->list);
8749         up_write(&block_group->space_info->groups_sem);
8750
8751         if (block_group->cached == BTRFS_CACHE_STARTED)
8752                 wait_block_group_cache_done(block_group);
8753
8754         btrfs_remove_free_space_cache(block_group);
8755
8756         spin_lock(&block_group->space_info->lock);
8757         block_group->space_info->total_bytes -= block_group->key.offset;
8758         block_group->space_info->bytes_readonly -= block_group->key.offset;
8759         block_group->space_info->disk_total -= block_group->key.offset * factor;
8760         spin_unlock(&block_group->space_info->lock);
8761
8762         memcpy(&key, &block_group->key, sizeof(key));
8763
8764         btrfs_clear_space_info_full(root->fs_info);
8765
8766         btrfs_put_block_group(block_group);
8767         btrfs_put_block_group(block_group);
8768
8769         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8770         if (ret > 0)
8771                 ret = -EIO;
8772         if (ret < 0)
8773                 goto out;
8774
8775         ret = btrfs_del_item(trans, root, path);
8776 out:
8777         btrfs_free_path(path);
8778         return ret;
8779 }
8780
8781 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8782 {
8783         struct btrfs_space_info *space_info;
8784         int ret;
8785
8786         ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM, 0, 0,
8787                                                                  &space_info);
8788         if (ret)
8789                 return ret;
8790
8791         ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA, 0, 0,
8792                                                                  &space_info);
8793         if (ret)
8794                 return ret;
8795
8796         ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA, 0, 0,
8797                                                                  &space_info);
8798         if (ret)
8799                 return ret;
8800
8801         return ret;
8802 }
8803
8804 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8805 {
8806         return unpin_extent_range(root, start, end);
8807 }
8808
8809 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8810                                u64 num_bytes, u64 *actual_bytes)
8811 {
8812         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8813 }
8814
8815 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8816 {
8817         struct btrfs_fs_info *fs_info = root->fs_info;
8818         struct btrfs_block_group_cache *cache = NULL;
8819         u64 group_trimmed;
8820         u64 start;
8821         u64 end;
8822         u64 trimmed = 0;
8823         int ret = 0;
8824
8825         cache = btrfs_lookup_block_group(fs_info, range->start);
8826
8827         while (cache) {
8828                 if (cache->key.objectid >= (range->start + range->len)) {
8829                         btrfs_put_block_group(cache);
8830                         break;
8831                 }
8832
8833                 start = max(range->start, cache->key.objectid);
8834                 end = min(range->start + range->len,
8835                                 cache->key.objectid + cache->key.offset);
8836
8837                 if (end - start >= range->minlen) {
8838                         if (!block_group_cache_done(cache)) {
8839                                 ret = cache_block_group(cache, NULL, root, 0);
8840                                 if (!ret)
8841                                         wait_block_group_cache_done(cache);
8842                         }
8843                         ret = btrfs_trim_block_group(cache,
8844                                                      &group_trimmed,
8845                                                      start,
8846                                                      end,
8847                                                      range->minlen);
8848
8849                         trimmed += group_trimmed;
8850                         if (ret) {
8851                                 btrfs_put_block_group(cache);
8852                                 break;
8853                         }
8854                 }
8855
8856                 cache = next_block_group(fs_info->tree_root, cache);
8857         }
8858
8859         range->len = trimmed;
8860         return ret;
8861 }