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 /* control flags for do_chunk_alloc's force field
  37  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  38  * if we really need one.
  39  *
  40  * CHUNK_ALLOC_FORCE means it must try to allocate one
  41  *
  42  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  43  * if we have very few chunks already allocated.  This is
  44  * used as part of the clustering code to help make sure
  45  * we have a good pool of storage to cluster in, without
  46  * filling the FS with empty chunks
  47  *
  48  */
  49 enum {
  50         CHUNK_ALLOC_NO_FORCE = 0,
  51         CHUNK_ALLOC_FORCE = 1,
  52         CHUNK_ALLOC_LIMITED = 2,
  53 };
  54
  55 static int update_block_group(struct btrfs_trans_handle *trans,
  56                               struct btrfs_root *root,
  57                               u64 bytenr, u64 num_bytes, int alloc);
  58 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  59                                 struct btrfs_root *root,
  60                                 u64 bytenr, u64 num_bytes, u64 parent,
  61                                 u64 root_objectid, u64 owner_objectid,
  62                                 u64 owner_offset, int refs_to_drop,
  63                                 struct btrfs_delayed_extent_op *extra_op);
  64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  65                                     struct extent_buffer *leaf,
  66                                     struct btrfs_extent_item *ei);
  67 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  68                                       struct btrfs_root *root,
  69                                       u64 parent, u64 root_objectid,
  70                                       u64 flags, u64 owner, u64 offset,
  71                                       struct btrfs_key *ins, int ref_mod);
  72 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  73                                      struct btrfs_root *root,
  74                                      u64 parent, u64 root_objectid,
  75                                      u64 flags, struct btrfs_disk_key *key,
  76                                      int level, struct btrfs_key *ins);
  77 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  78                           struct btrfs_root *extent_root, u64 alloc_bytes,
  79                           u64 flags, int force);
  80 static int find_next_key(struct btrfs_path *path, int level,
  81                          struct btrfs_key *key);
  82 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  83                             int dump_block_groups);
  84
  85 static noinline int
  86 block_group_cache_done(struct btrfs_block_group_cache *cache)
  87 {
  88         smp_mb();
  89         return cache->cached == BTRFS_CACHE_FINISHED;
  90 }
  91
  92 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  93 {
  94         return (cache->flags & bits) == bits;
  95 }
  96
  97 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  98 {
  99         atomic_inc(&cache->count);
 100 }
 101
 102 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 103 {
 104         if (atomic_dec_and_test(&cache->count)) {
 105                 WARN_ON(cache->pinned > 0);
 106                 WARN_ON(cache->reserved > 0);
 107                 WARN_ON(cache->reserved_pinned > 0);
 108                 kfree(cache);
 109         }
 110 }
 111
 112 /*
 113  * this adds the block group to the fs_info rb tree for the block group
 114  * cache
 115  */
 116 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 117                                 struct btrfs_block_group_cache *block_group)
 118 {
 119         struct rb_node **p;
 120         struct rb_node *parent = NULL;
 121         struct btrfs_block_group_cache *cache;
 122
 123         spin_lock(&info->block_group_cache_lock);
 124         p = &info->block_group_cache_tree.rb_node;
 125
 126         while (*p) {
 127                 parent = *p;
 128                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 129                                  cache_node);
 130                 if (block_group->key.objectid < cache->key.objectid) {
 131                         p = &(*p)->rb_left;
 132                 } else if (block_group->key.objectid > cache->key.objectid) {
 133                         p = &(*p)->rb_right;
 134                 } else {
 135                         spin_unlock(&info->block_group_cache_lock);
 136                         return -EEXIST;
 137                 }
 138         }
 139
 140         rb_link_node(&block_group->cache_node, parent, p);
 141         rb_insert_color(&block_group->cache_node,
 142                         &info->block_group_cache_tree);
 143         spin_unlock(&info->block_group_cache_lock);
 144
 145         return 0;
 146 }
 147
 148 /*
 149  * This will return the block group at or after bytenr if contains is 0, else
 150  * it will return the block group that contains the bytenr
 151  */
 152 static struct btrfs_block_group_cache *
 153 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 154                               int contains)
 155 {
 156         struct btrfs_block_group_cache *cache, *ret = NULL;
 157         struct rb_node *n;
 158         u64 end, start;
 159
 160         spin_lock(&info->block_group_cache_lock);
 161         n = info->block_group_cache_tree.rb_node;
 162
 163         while (n) {
 164                 cache = rb_entry(n, struct btrfs_block_group_cache,
 165                                  cache_node);
 166                 end = cache->key.objectid + cache->key.offset - 1;
 167                 start = cache->key.objectid;
 168
 169                 if (bytenr < start) {
 170                         if (!contains && (!ret || start < ret->key.objectid))
 171                                 ret = cache;
 172                         n = n->rb_left;
 173                 } else if (bytenr > start) {
 174                         if (contains && bytenr <= end) {
 175                                 ret = cache;
 176                                 break;
 177                         }
 178                         n = n->rb_right;
 179                 } else {
 180                         ret = cache;
 181                         break;
 182                 }
 183         }
 184         if (ret)
 185                 btrfs_get_block_group(ret);
 186         spin_unlock(&info->block_group_cache_lock);
 187
 188         return ret;
 189 }
 190
 191 static int add_excluded_extent(struct btrfs_root *root,
 192                                u64 start, u64 num_bytes)
 193 {
 194         u64 end = start + num_bytes - 1;
 195         set_extent_bits(&root->fs_info->freed_extents[0],
 196                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 197         set_extent_bits(&root->fs_info->freed_extents[1],
 198                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 199         return 0;
 200 }
 201
 202 static void free_excluded_extents(struct btrfs_root *root,
 203                                   struct btrfs_block_group_cache *cache)
 204 {
 205         u64 start, end;
 206
 207         start = cache->key.objectid;
 208         end = start + cache->key.offset - 1;
 209
 210         clear_extent_bits(&root->fs_info->freed_extents[0],
 211                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 212         clear_extent_bits(&root->fs_info->freed_extents[1],
 213                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 214 }
 215
 216 static int exclude_super_stripes(struct btrfs_root *root,
 217                                  struct btrfs_block_group_cache *cache)
 218 {
 219         u64 bytenr;
 220         u64 *logical;
 221         int stripe_len;
 222         int i, nr, ret;
 223
 224         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 225                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 226                 cache->bytes_super += stripe_len;
 227                 ret = add_excluded_extent(root, cache->key.objectid,
 228                                           stripe_len);
 229                 BUG_ON(ret);
 230         }
 231
 232         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 233                 bytenr = btrfs_sb_offset(i);
 234                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 235                                        cache->key.objectid, bytenr,
 236                                        0, &logical, &nr, &stripe_len);
 237                 BUG_ON(ret);
 238
 239                 while (nr--) {
 240                         cache->bytes_super += stripe_len;
 241                         ret = add_excluded_extent(root, logical[nr],
 242                                                   stripe_len);
 243                         BUG_ON(ret);
 244                 }
 245
 246                 kfree(logical);
 247         }
 248         return 0;
 249 }
 250
 251 static struct btrfs_caching_control *
 252 get_caching_control(struct btrfs_block_group_cache *cache)
 253 {
 254         struct btrfs_caching_control *ctl;
 255
 256         spin_lock(&cache->lock);
 257         if (cache->cached != BTRFS_CACHE_STARTED) {
 258                 spin_unlock(&cache->lock);
 259                 return NULL;
 260         }
 261
 262         /* We're loading it the fast way, so we don't have a caching_ctl. */
 263         if (!cache->caching_ctl) {
 264                 spin_unlock(&cache->lock);
 265                 return NULL;
 266         }
 267
 268         ctl = cache->caching_ctl;
 269         atomic_inc(&ctl->count);
 270         spin_unlock(&cache->lock);
 271         return ctl;
 272 }
 273
 274 static void put_caching_control(struct btrfs_caching_control *ctl)
 275 {
 276         if (atomic_dec_and_test(&ctl->count))
 277                 kfree(ctl);
 278 }
 279
 280 /*
 281  * this is only called by cache_block_group, since we could have freed extents
 282  * we need to check the pinned_extents for any extents that can't be used yet
 283  * since their free space will be released as soon as the transaction commits.
 284  */
 285 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 286                               struct btrfs_fs_info *info, u64 start, u64 end)
 287 {
 288         u64 extent_start, extent_end, size, total_added = 0;
 289         int ret;
 290
 291         while (start < end) {
 292                 ret = find_first_extent_bit(info->pinned_extents, start,
 293                                             &extent_start, &extent_end,
 294                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 295                 if (ret)
 296                         break;
 297
 298                 if (extent_start <= start) {
 299                         start = extent_end + 1;
 300                 } else if (extent_start > start && extent_start < end) {
 301                         size = extent_start - start;
 302                         total_added += size;
 303                         ret = btrfs_add_free_space(block_group, start,
 304                                                    size);
 305                         BUG_ON(ret);
 306                         start = extent_end + 1;
 307                 } else {
 308                         break;
 309                 }
 310         }
 311
 312         if (start < end) {
 313                 size = end - start;
 314                 total_added += size;
 315                 ret = btrfs_add_free_space(block_group, start, size);
 316                 BUG_ON(ret);
 317         }
 318
 319         return total_added;
 320 }
 321
 322 static int caching_kthread(void *data)
 323 {
 324         struct btrfs_block_group_cache *block_group = data;
 325         struct btrfs_fs_info *fs_info = block_group->fs_info;
 326         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 327         struct btrfs_root *extent_root = fs_info->extent_root;
 328         struct btrfs_path *path;
 329         struct extent_buffer *leaf;
 330         struct btrfs_key key;
 331         u64 total_found = 0;
 332         u64 last = 0;
 333         u32 nritems;
 334         int ret = 0;
 335
 336         path = btrfs_alloc_path();
 337         if (!path)
 338                 return -ENOMEM;
 339
 340         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 341
 342         /*
 343          * We don't want to deadlock with somebody trying to allocate a new
 344          * extent for the extent root while also trying to search the extent
 345          * root to add free space.  So we skip locking and search the commit
 346          * root, since its read-only
 347          */
 348         path->skip_locking = 1;
 349         path->search_commit_root = 1;
 350         path->reada = 2;
 351
 352         key.objectid = last;
 353         key.offset = 0;
 354         key.type = BTRFS_EXTENT_ITEM_KEY;
 355 again:
 356         mutex_lock(&caching_ctl->mutex);
 357         /* need to make sure the commit_root doesn't disappear */
 358         down_read(&fs_info->extent_commit_sem);
 359
 360         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 361         if (ret < 0)
 362                 goto err;
 363
 364         leaf = path->nodes[0];
 365         nritems = btrfs_header_nritems(leaf);
 366
 367         while (1) {
 368                 smp_mb();
 369                 if (fs_info->closing > 1) {
 370                         last = (u64)-1;
 371                         break;
 372                 }
 373
 374                 if (path->slots[0] < nritems) {
 375                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 376                 } else {
 377                         ret = find_next_key(path, 0, &key);
 378                         if (ret)
 379                                 break;
 380
 381                         caching_ctl->progress = last;
 382                         btrfs_release_path(extent_root, path);
 383                         up_read(&fs_info->extent_commit_sem);
 384                         mutex_unlock(&caching_ctl->mutex);
 385                         if (btrfs_transaction_in_commit(fs_info))
 386                                 schedule_timeout(1);
 387                         else
 388                                 cond_resched();
 389                         goto again;
 390                 }
 391
 392                 if (key.objectid < block_group->key.objectid) {
 393                         path->slots[0]++;
 394                         continue;
 395                 }
 396
 397                 if (key.objectid >= block_group->key.objectid +
 398                     block_group->key.offset)
 399                         break;
 400
 401                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 402                         total_found += add_new_free_space(block_group,
 403                                                           fs_info, last,
 404                                                           key.objectid);
 405                         last = key.objectid + key.offset;
 406
 407                         if (total_found > (1024 * 1024 * 2)) {
 408                                 total_found = 0;
 409                                 wake_up(&caching_ctl->wait);
 410                         }
 411                 }
 412                 path->slots[0]++;
 413         }
 414         ret = 0;
 415
 416         total_found += add_new_free_space(block_group, fs_info, last,
 417                                           block_group->key.objectid +
 418                                           block_group->key.offset);
 419         caching_ctl->progress = (u64)-1;
 420
 421         spin_lock(&block_group->lock);
 422         block_group->caching_ctl = NULL;
 423         block_group->cached = BTRFS_CACHE_FINISHED;
 424         spin_unlock(&block_group->lock);
 425
 426 err:
 427         btrfs_free_path(path);
 428         up_read(&fs_info->extent_commit_sem);
 429
 430         free_excluded_extents(extent_root, block_group);
 431
 432         mutex_unlock(&caching_ctl->mutex);
 433         wake_up(&caching_ctl->wait);
 434
 435         put_caching_control(caching_ctl);
 436         atomic_dec(&block_group->space_info->caching_threads);
 437         btrfs_put_block_group(block_group);
 438
 439         return 0;
 440 }
 441
 442 static int cache_block_group(struct btrfs_block_group_cache *cache,
 443                              struct btrfs_trans_handle *trans,
 444                              struct btrfs_root *root,
 445                              int load_cache_only)
 446 {
 447         struct btrfs_fs_info *fs_info = cache->fs_info;
 448         struct btrfs_caching_control *caching_ctl;
 449         struct task_struct *tsk;
 450         int ret = 0;
 451
 452         smp_mb();
 453         if (cache->cached != BTRFS_CACHE_NO)
 454                 return 0;
 455
 456         /*
 457          * We can't do the read from on-disk cache during a commit since we need
 458          * to have the normal tree locking.  Also if we are currently trying to
 459          * allocate blocks for the tree root we can't do the fast caching since
 460          * we likely hold important locks.
 461          */
 462         if (trans && (!trans->transaction->in_commit) &&
 463             (root && root != root->fs_info->tree_root)) {
 464                 spin_lock(&cache->lock);
 465                 if (cache->cached != BTRFS_CACHE_NO) {
 466                         spin_unlock(&cache->lock);
 467                         return 0;
 468                 }
 469                 cache->cached = BTRFS_CACHE_STARTED;
 470                 spin_unlock(&cache->lock);
 471
 472                 ret = load_free_space_cache(fs_info, cache);
 473
 474                 spin_lock(&cache->lock);
 475                 if (ret == 1) {
 476                         cache->cached = BTRFS_CACHE_FINISHED;
 477                         cache->last_byte_to_unpin = (u64)-1;
 478                 } else {
 479                         cache->cached = BTRFS_CACHE_NO;
 480                 }
 481                 spin_unlock(&cache->lock);
 482                 if (ret == 1) {
 483                         free_excluded_extents(fs_info->extent_root, cache);
 484                         return 0;
 485                 }
 486         }
 487
 488         if (load_cache_only)
 489                 return 0;
 490
 491         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 492         BUG_ON(!caching_ctl);
 493
 494         INIT_LIST_HEAD(&caching_ctl->list);
 495         mutex_init(&caching_ctl->mutex);
 496         init_waitqueue_head(&caching_ctl->wait);
 497         caching_ctl->block_group = cache;
 498         caching_ctl->progress = cache->key.objectid;
 499         /* one for caching kthread, one for caching block group list */
 500         atomic_set(&caching_ctl->count, 2);
 501
 502         spin_lock(&cache->lock);
 503         if (cache->cached != BTRFS_CACHE_NO) {
 504                 spin_unlock(&cache->lock);
 505                 kfree(caching_ctl);
 506                 return 0;
 507         }
 508         cache->caching_ctl = caching_ctl;
 509         cache->cached = BTRFS_CACHE_STARTED;
 510         spin_unlock(&cache->lock);
 511
 512         down_write(&fs_info->extent_commit_sem);
 513         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 514         up_write(&fs_info->extent_commit_sem);
 515
 516         atomic_inc(&cache->space_info->caching_threads);
 517         btrfs_get_block_group(cache);
 518
 519         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 520                           cache->key.objectid);
 521         if (IS_ERR(tsk)) {
 522                 ret = PTR_ERR(tsk);
 523                 printk(KERN_ERR "error running thread %d\n", ret);
 524                 BUG();
 525         }
 526
 527         return ret;
 528 }
 529
 530 /*
 531  * return the block group that starts at or after bytenr
 532  */
 533 static struct btrfs_block_group_cache *
 534 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 535 {
 536         struct btrfs_block_group_cache *cache;
 537
 538         cache = block_group_cache_tree_search(info, bytenr, 0);
 539
 540         return cache;
 541 }
 542
 543 /*
 544  * return the block group that contains the given bytenr
 545  */
 546 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 547                                                  struct btrfs_fs_info *info,
 548                                                  u64 bytenr)
 549 {
 550         struct btrfs_block_group_cache *cache;
 551
 552         cache = block_group_cache_tree_search(info, bytenr, 1);
 553
 554         return cache;
 555 }
 556
 557 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 558                                                   u64 flags)
 559 {
 560         struct list_head *head = &info->space_info;
 561         struct btrfs_space_info *found;
 562
 563         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 564                  BTRFS_BLOCK_GROUP_METADATA;
 565
 566         rcu_read_lock();
 567         list_for_each_entry_rcu(found, head, list) {
 568                 if (found->flags & flags) {
 569                         rcu_read_unlock();
 570                         return found;
 571                 }
 572         }
 573         rcu_read_unlock();
 574         return NULL;
 575 }
 576
 577 /*
 578  * after adding space to the filesystem, we need to clear the full flags
 579  * on all the space infos.
 580  */
 581 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 582 {
 583         struct list_head *head = &info->space_info;
 584         struct btrfs_space_info *found;
 585
 586         rcu_read_lock();
 587         list_for_each_entry_rcu(found, head, list)
 588                 found->full = 0;
 589         rcu_read_unlock();
 590 }
 591
 592 static u64 div_factor(u64 num, int factor)
 593 {
 594         if (factor == 10)
 595                 return num;
 596         num *= factor;
 597         do_div(num, 10);
 598         return num;
 599 }
 600
 601 static u64 div_factor_fine(u64 num, int factor)
 602 {
 603         if (factor == 100)
 604                 return num;
 605         num *= factor;
 606         do_div(num, 100);
 607         return num;
 608 }
 609
 610 u64 btrfs_find_block_group(struct btrfs_root *root,
 611                            u64 search_start, u64 search_hint, int owner)
 612 {
 613         struct btrfs_block_group_cache *cache;
 614         u64 used;
 615         u64 last = max(search_hint, search_start);
 616         u64 group_start = 0;
 617         int full_search = 0;
 618         int factor = 9;
 619         int wrapped = 0;
 620 again:
 621         while (1) {
 622                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 623                 if (!cache)
 624                         break;
 625
 626                 spin_lock(&cache->lock);
 627                 last = cache->key.objectid + cache->key.offset;
 628                 used = btrfs_block_group_used(&cache->item);
 629
 630                 if ((full_search || !cache->ro) &&
 631                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 632                         if (used + cache->pinned + cache->reserved <
 633                             div_factor(cache->key.offset, factor)) {
 634                                 group_start = cache->key.objectid;
 635                                 spin_unlock(&cache->lock);
 636                                 btrfs_put_block_group(cache);
 637                                 goto found;
 638                         }
 639                 }
 640                 spin_unlock(&cache->lock);
 641                 btrfs_put_block_group(cache);
 642                 cond_resched();
 643         }
 644         if (!wrapped) {
 645                 last = search_start;
 646                 wrapped = 1;
 647                 goto again;
 648         }
 649         if (!full_search && factor < 10) {
 650                 last = search_start;
 651                 full_search = 1;
 652                 factor = 10;
 653                 goto again;
 654         }
 655 found:
 656         return group_start;
 657 }
 658
 659 /* simple helper to search for an existing extent at a given offset */
 660 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 661 {
 662         int ret;
 663         struct btrfs_key key;
 664         struct btrfs_path *path;
 665
 666         path = btrfs_alloc_path();
 667         BUG_ON(!path);
 668         key.objectid = start;
 669         key.offset = len;
 670         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 671         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 672                                 0, 0);
 673         btrfs_free_path(path);
 674         return ret;
 675 }
 676
 677 /*
 678  * helper function to lookup reference count and flags of extent.
 679  *
 680  * the head node for delayed ref is used to store the sum of all the
 681  * reference count modifications queued up in the rbtree. the head
 682  * node may also store the extent flags to set. This way you can check
 683  * to see what the reference count and extent flags would be if all of
 684  * the delayed refs are not processed.
 685  */
 686 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 687                              struct btrfs_root *root, u64 bytenr,
 688                              u64 num_bytes, u64 *refs, u64 *flags)
 689 {
 690         struct btrfs_delayed_ref_head *head;
 691         struct btrfs_delayed_ref_root *delayed_refs;
 692         struct btrfs_path *path;
 693         struct btrfs_extent_item *ei;
 694         struct extent_buffer *leaf;
 695         struct btrfs_key key;
 696         u32 item_size;
 697         u64 num_refs;
 698         u64 extent_flags;
 699         int ret;
 700
 701         path = btrfs_alloc_path();
 702         if (!path)
 703                 return -ENOMEM;
 704
 705         key.objectid = bytenr;
 706         key.type = BTRFS_EXTENT_ITEM_KEY;
 707         key.offset = num_bytes;
 708         if (!trans) {
 709                 path->skip_locking = 1;
 710                 path->search_commit_root = 1;
 711         }
 712 again:
 713         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 714                                 &key, path, 0, 0);
 715         if (ret < 0)
 716                 goto out_free;
 717
 718         if (ret == 0) {
 719                 leaf = path->nodes[0];
 720                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 721                 if (item_size >= sizeof(*ei)) {
 722                         ei = btrfs_item_ptr(leaf, path->slots[0],
 723                                             struct btrfs_extent_item);
 724                         num_refs = btrfs_extent_refs(leaf, ei);
 725                         extent_flags = btrfs_extent_flags(leaf, ei);
 726                 } else {
 727 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 728                         struct btrfs_extent_item_v0 *ei0;
 729                         BUG_ON(item_size != sizeof(*ei0));
 730                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 731                                              struct btrfs_extent_item_v0);
 732                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 733                         /* FIXME: this isn't correct for data */
 734                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 735 #else
 736                         BUG();
 737 #endif
 738                 }
 739                 BUG_ON(num_refs == 0);
 740         } else {
 741                 num_refs = 0;
 742                 extent_flags = 0;
 743                 ret = 0;
 744         }
 745
 746         if (!trans)
 747                 goto out;
 748
 749         delayed_refs = &trans->transaction->delayed_refs;
 750         spin_lock(&delayed_refs->lock);
 751         head = btrfs_find_delayed_ref_head(trans, bytenr);
 752         if (head) {
 753                 if (!mutex_trylock(&head->mutex)) {
 754                         atomic_inc(&head->node.refs);
 755                         spin_unlock(&delayed_refs->lock);
 756
 757                         btrfs_release_path(root->fs_info->extent_root, path);
 758
 759                         mutex_lock(&head->mutex);
 760                         mutex_unlock(&head->mutex);
 761                         btrfs_put_delayed_ref(&head->node);
 762                         goto again;
 763                 }
 764                 if (head->extent_op && head->extent_op->update_flags)
 765                         extent_flags |= head->extent_op->flags_to_set;
 766                 else
 767                         BUG_ON(num_refs == 0);
 768
 769                 num_refs += head->node.ref_mod;
 770                 mutex_unlock(&head->mutex);
 771         }
 772         spin_unlock(&delayed_refs->lock);
 773 out:
 774         WARN_ON(num_refs == 0);
 775         if (refs)
 776                 *refs = num_refs;
 777         if (flags)
 778                 *flags = extent_flags;
 779 out_free:
 780         btrfs_free_path(path);
 781         return ret;
 782 }
 783
 784 /*
 785  * Back reference rules.  Back refs have three main goals:
 786  *
 787  * 1) differentiate between all holders of references to an extent so that
 788  *    when a reference is dropped we can make sure it was a valid reference
 789  *    before freeing the extent.
 790  *
 791  * 2) Provide enough information to quickly find the holders of an extent
 792  *    if we notice a given block is corrupted or bad.
 793  *
 794  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 795  *    maintenance.  This is actually the same as #2, but with a slightly
 796  *    different use case.
 797  *
 798  * There are two kinds of back refs. The implicit back refs is optimized
 799  * for pointers in non-shared tree blocks. For a given pointer in a block,
 800  * back refs of this kind provide information about the block's owner tree
 801  * and the pointer's key. These information allow us to find the block by
 802  * b-tree searching. The full back refs is for pointers in tree blocks not
 803  * referenced by their owner trees. The location of tree block is recorded
 804  * in the back refs. Actually the full back refs is generic, and can be
 805  * used in all cases the implicit back refs is used. The major shortcoming
 806  * of the full back refs is its overhead. Every time a tree block gets
 807  * COWed, we have to update back refs entry for all pointers in it.
 808  *
 809  * For a newly allocated tree block, we use implicit back refs for
 810  * pointers in it. This means most tree related operations only involve
 811  * implicit back refs. For a tree block created in old transaction, the
 812  * only way to drop a reference to it is COW it. So we can detect the
 813  * event that tree block loses its owner tree's reference and do the
 814  * back refs conversion.
 815  *
 816  * When a tree block is COW'd through a tree, there are four cases:
 817  *
 818  * The reference count of the block is one and the tree is the block's
 819  * owner tree. Nothing to do in this case.
 820  *
 821  * The reference count of the block is one and the tree is not the
 822  * block's owner tree. In this case, full back refs is used for pointers
 823  * in the block. Remove these full back refs, add implicit back refs for
 824  * every pointers in the new block.
 825  *
 826  * The reference count of the block is greater than one and the tree is
 827  * the block's owner tree. In this case, implicit back refs is used for
 828  * pointers in the block. Add full back refs for every pointers in the
 829  * block, increase lower level extents' reference counts. The original
 830  * implicit back refs are entailed to the new block.
 831  *
 832  * The reference count of the block is greater than one and the tree is
 833  * not the block's owner tree. Add implicit back refs for every pointer in
 834  * the new block, increase lower level extents' reference count.
 835  *
 836  * Back Reference Key composing:
 837  *
 838  * The key objectid corresponds to the first byte in the extent,
 839  * The key type is used to differentiate between types of back refs.
 840  * There are different meanings of the key offset for different types
 841  * of back refs.
 842  *
 843  * File extents can be referenced by:
 844  *
 845  * - multiple snapshots, subvolumes, or different generations in one subvol
 846  * - different files inside a single subvolume
 847  * - different offsets inside a file (bookend extents in file.c)
 848  *
 849  * The extent ref structure for the implicit back refs has fields for:
 850  *
 851  * - Objectid of the subvolume root
 852  * - objectid of the file holding the reference
 853  * - original offset in the file
 854  * - how many bookend extents
 855  *
 856  * The key offset for the implicit back refs is hash of the first
 857  * three fields.
 858  *
 859  * The extent ref structure for the full back refs has field for:
 860  *
 861  * - number of pointers in the tree leaf
 862  *
 863  * The key offset for the implicit back refs is the first byte of
 864  * the tree leaf
 865  *
 866  * When a file extent is allocated, The implicit back refs is used.
 867  * the fields are filled in:
 868  *
 869  *     (root_key.objectid, inode objectid, offset in file, 1)
 870  *
 871  * When a file extent is removed file truncation, we find the
 872  * corresponding implicit back refs and check the following fields:
 873  *
 874  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 875  *
 876  * Btree extents can be referenced by:
 877  *
 878  * - Different subvolumes
 879  *
 880  * Both the implicit back refs and the full back refs for tree blocks
 881  * only consist of key. The key offset for the implicit back refs is
 882  * objectid of block's owner tree. The key offset for the full back refs
 883  * is the first byte of parent block.
 884  *
 885  * When implicit back refs is used, information about the lowest key and
 886  * level of the tree block are required. These information are stored in
 887  * tree block info structure.
 888  */
 889
 890 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 891 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 892                                   struct btrfs_root *root,
 893                                   struct btrfs_path *path,
 894                                   u64 owner, u32 extra_size)
 895 {
 896         struct btrfs_extent_item *item;
 897         struct btrfs_extent_item_v0 *ei0;
 898         struct btrfs_extent_ref_v0 *ref0;
 899         struct btrfs_tree_block_info *bi;
 900         struct extent_buffer *leaf;
 901         struct btrfs_key key;
 902         struct btrfs_key found_key;
 903         u32 new_size = sizeof(*item);
 904         u64 refs;
 905         int ret;
 906
 907         leaf = path->nodes[0];
 908         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 909
 910         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 911         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 912                              struct btrfs_extent_item_v0);
 913         refs = btrfs_extent_refs_v0(leaf, ei0);
 914
 915         if (owner == (u64)-1) {
 916                 while (1) {
 917                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 918                                 ret = btrfs_next_leaf(root, path);
 919                                 if (ret < 0)
 920                                         return ret;
 921                                 BUG_ON(ret > 0);
 922                                 leaf = path->nodes[0];
 923                         }
 924                         btrfs_item_key_to_cpu(leaf, &found_key,
 925                                               path->slots[0]);
 926                         BUG_ON(key.objectid != found_key.objectid);
 927                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 928                                 path->slots[0]++;
 929                                 continue;
 930                         }
 931                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 932                                               struct btrfs_extent_ref_v0);
 933                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 934                         break;
 935                 }
 936         }
 937         btrfs_release_path(root, path);
 938
 939         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 940                 new_size += sizeof(*bi);
 941
 942         new_size -= sizeof(*ei0);
 943         ret = btrfs_search_slot(trans, root, &key, path,
 944                                 new_size + extra_size, 1);
 945         if (ret < 0)
 946                 return ret;
 947         BUG_ON(ret);
 948
 949         ret = btrfs_extend_item(trans, root, path, new_size);
 950         BUG_ON(ret);
 951
 952         leaf = path->nodes[0];
 953         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 954         btrfs_set_extent_refs(leaf, item, refs);
 955         /* FIXME: get real generation */
 956         btrfs_set_extent_generation(leaf, item, 0);
 957         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 958                 btrfs_set_extent_flags(leaf, item,
 959                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 960                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 961                 bi = (struct btrfs_tree_block_info *)(item + 1);
 962                 /* FIXME: get first key of the block */
 963                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 964                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 965         } else {
 966                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 967         }
 968         btrfs_mark_buffer_dirty(leaf);
 969         return 0;
 970 }
 971 #endif
 972
 973 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 974 {
 975         u32 high_crc = ~(u32)0;
 976         u32 low_crc = ~(u32)0;
 977         __le64 lenum;
 978
 979         lenum = cpu_to_le64(root_objectid);
 980         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 981         lenum = cpu_to_le64(owner);
 982         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 983         lenum = cpu_to_le64(offset);
 984         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 985
 986         return ((u64)high_crc << 31) ^ (u64)low_crc;
 987 }
 988
 989 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 990                                      struct btrfs_extent_data_ref *ref)
 991 {
 992         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 993                                     btrfs_extent_data_ref_objectid(leaf, ref),
 994                                     btrfs_extent_data_ref_offset(leaf, ref));
 995 }
 996
 997 static int match_extent_data_ref(struct extent_buffer *leaf,
 998                                  struct btrfs_extent_data_ref *ref,
 999                                  u64 root_objectid, u64 owner, u64 offset)
1000 {
1001         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1002             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1003             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1004                 return 0;
1005         return 1;
1006 }
1007
1008 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1009                                            struct btrfs_root *root,
1010                                            struct btrfs_path *path,
1011                                            u64 bytenr, u64 parent,
1012                                            u64 root_objectid,
1013                                            u64 owner, u64 offset)
1014 {
1015         struct btrfs_key key;
1016         struct btrfs_extent_data_ref *ref;
1017         struct extent_buffer *leaf;
1018         u32 nritems;
1019         int ret;
1020         int recow;
1021         int err = -ENOENT;
1022
1023         key.objectid = bytenr;
1024         if (parent) {
1025                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1026                 key.offset = parent;
1027         } else {
1028                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1029                 key.offset = hash_extent_data_ref(root_objectid,
1030                                                   owner, offset);
1031         }
1032 again:
1033         recow = 0;
1034         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1035         if (ret < 0) {
1036                 err = ret;
1037                 goto fail;
1038         }
1039
1040         if (parent) {
1041                 if (!ret)
1042                         return 0;
1043 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1044                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1045                 btrfs_release_path(root, path);
1046                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1047                 if (ret < 0) {
1048                         err = ret;
1049                         goto fail;
1050                 }
1051                 if (!ret)
1052                         return 0;
1053 #endif
1054                 goto fail;
1055         }
1056
1057         leaf = path->nodes[0];
1058         nritems = btrfs_header_nritems(leaf);
1059         while (1) {
1060                 if (path->slots[0] >= nritems) {
1061                         ret = btrfs_next_leaf(root, path);
1062                         if (ret < 0)
1063                                 err = ret;
1064                         if (ret)
1065                                 goto fail;
1066
1067                         leaf = path->nodes[0];
1068                         nritems = btrfs_header_nritems(leaf);
1069                         recow = 1;
1070                 }
1071
1072                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1073                 if (key.objectid != bytenr ||
1074                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1075                         goto fail;
1076
1077                 ref = btrfs_item_ptr(leaf, path->slots[0],
1078                                      struct btrfs_extent_data_ref);
1079
1080                 if (match_extent_data_ref(leaf, ref, root_objectid,
1081                                           owner, offset)) {
1082                         if (recow) {
1083                                 btrfs_release_path(root, path);
1084                                 goto again;
1085                         }
1086                         err = 0;
1087                         break;
1088                 }
1089                 path->slots[0]++;
1090         }
1091 fail:
1092         return err;
1093 }
1094
1095 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1096                                            struct btrfs_root *root,
1097                                            struct btrfs_path *path,
1098                                            u64 bytenr, u64 parent,
1099                                            u64 root_objectid, u64 owner,
1100                                            u64 offset, int refs_to_add)
1101 {
1102         struct btrfs_key key;
1103         struct extent_buffer *leaf;
1104         u32 size;
1105         u32 num_refs;
1106         int ret;
1107
1108         key.objectid = bytenr;
1109         if (parent) {
1110                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1111                 key.offset = parent;
1112                 size = sizeof(struct btrfs_shared_data_ref);
1113         } else {
1114                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1115                 key.offset = hash_extent_data_ref(root_objectid,
1116                                                   owner, offset);
1117                 size = sizeof(struct btrfs_extent_data_ref);
1118         }
1119
1120         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1121         if (ret && ret != -EEXIST)
1122                 goto fail;
1123
1124         leaf = path->nodes[0];
1125         if (parent) {
1126                 struct btrfs_shared_data_ref *ref;
1127                 ref = btrfs_item_ptr(leaf, path->slots[0],
1128                                      struct btrfs_shared_data_ref);
1129                 if (ret == 0) {
1130                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1131                 } else {
1132                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1133                         num_refs += refs_to_add;
1134                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1135                 }
1136         } else {
1137                 struct btrfs_extent_data_ref *ref;
1138                 while (ret == -EEXIST) {
1139                         ref = btrfs_item_ptr(leaf, path->slots[0],
1140                                              struct btrfs_extent_data_ref);
1141                         if (match_extent_data_ref(leaf, ref, root_objectid,
1142                                                   owner, offset))
1143                                 break;
1144                         btrfs_release_path(root, path);
1145                         key.offset++;
1146                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1147                                                       size);
1148                         if (ret && ret != -EEXIST)
1149                                 goto fail;
1150
1151                         leaf = path->nodes[0];
1152                 }
1153                 ref = btrfs_item_ptr(leaf, path->slots[0],
1154                                      struct btrfs_extent_data_ref);
1155                 if (ret == 0) {
1156                         btrfs_set_extent_data_ref_root(leaf, ref,
1157                                                        root_objectid);
1158                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1159                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1160                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1161                 } else {
1162                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1163                         num_refs += refs_to_add;
1164                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1165                 }
1166         }
1167         btrfs_mark_buffer_dirty(leaf);
1168         ret = 0;
1169 fail:
1170         btrfs_release_path(root, path);
1171         return ret;
1172 }
1173
1174 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1175                                            struct btrfs_root *root,
1176                                            struct btrfs_path *path,
1177                                            int refs_to_drop)
1178 {
1179         struct btrfs_key key;
1180         struct btrfs_extent_data_ref *ref1 = NULL;
1181         struct btrfs_shared_data_ref *ref2 = NULL;
1182         struct extent_buffer *leaf;
1183         u32 num_refs = 0;
1184         int ret = 0;
1185
1186         leaf = path->nodes[0];
1187         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1188
1189         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1190                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1191                                       struct btrfs_extent_data_ref);
1192                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1193         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1194                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1195                                       struct btrfs_shared_data_ref);
1196                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1197 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1198         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1199                 struct btrfs_extent_ref_v0 *ref0;
1200                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1201                                       struct btrfs_extent_ref_v0);
1202                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1203 #endif
1204         } else {
1205                 BUG();
1206         }
1207
1208         BUG_ON(num_refs < refs_to_drop);
1209         num_refs -= refs_to_drop;
1210
1211         if (num_refs == 0) {
1212                 ret = btrfs_del_item(trans, root, path);
1213         } else {
1214                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1215                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1216                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1217                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1218 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1219                 else {
1220                         struct btrfs_extent_ref_v0 *ref0;
1221                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1222                                         struct btrfs_extent_ref_v0);
1223                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1224                 }
1225 #endif
1226                 btrfs_mark_buffer_dirty(leaf);
1227         }
1228         return ret;
1229 }
1230
1231 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1232                                           struct btrfs_path *path,
1233                                           struct btrfs_extent_inline_ref *iref)
1234 {
1235         struct btrfs_key key;
1236         struct extent_buffer *leaf;
1237         struct btrfs_extent_data_ref *ref1;
1238         struct btrfs_shared_data_ref *ref2;
1239         u32 num_refs = 0;
1240
1241         leaf = path->nodes[0];
1242         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1243         if (iref) {
1244                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1245                     BTRFS_EXTENT_DATA_REF_KEY) {
1246                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1247                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1248                 } else {
1249                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1250                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1251                 }
1252         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1253                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1254                                       struct btrfs_extent_data_ref);
1255                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1256         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1257                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1258                                       struct btrfs_shared_data_ref);
1259                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1260 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1261         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1262                 struct btrfs_extent_ref_v0 *ref0;
1263                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1264                                       struct btrfs_extent_ref_v0);
1265                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1266 #endif
1267         } else {
1268                 WARN_ON(1);
1269         }
1270         return num_refs;
1271 }
1272
1273 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1274                                           struct btrfs_root *root,
1275                                           struct btrfs_path *path,
1276                                           u64 bytenr, u64 parent,
1277                                           u64 root_objectid)
1278 {
1279         struct btrfs_key key;
1280         int ret;
1281
1282         key.objectid = bytenr;
1283         if (parent) {
1284                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1285                 key.offset = parent;
1286         } else {
1287                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1288                 key.offset = root_objectid;
1289         }
1290
1291         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1292         if (ret > 0)
1293                 ret = -ENOENT;
1294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295         if (ret == -ENOENT && parent) {
1296                 btrfs_release_path(root, path);
1297                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1298                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1299                 if (ret > 0)
1300                         ret = -ENOENT;
1301         }
1302 #endif
1303         return ret;
1304 }
1305
1306 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1307                                           struct btrfs_root *root,
1308                                           struct btrfs_path *path,
1309                                           u64 bytenr, u64 parent,
1310                                           u64 root_objectid)
1311 {
1312         struct btrfs_key key;
1313         int ret;
1314
1315         key.objectid = bytenr;
1316         if (parent) {
1317                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1318                 key.offset = parent;
1319         } else {
1320                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1321                 key.offset = root_objectid;
1322         }
1323
1324         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1325         btrfs_release_path(root, path);
1326         return ret;
1327 }
1328
1329 static inline int extent_ref_type(u64 parent, u64 owner)
1330 {
1331         int type;
1332         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1333                 if (parent > 0)
1334                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1335                 else
1336                         type = BTRFS_TREE_BLOCK_REF_KEY;
1337         } else {
1338                 if (parent > 0)
1339                         type = BTRFS_SHARED_DATA_REF_KEY;
1340                 else
1341                         type = BTRFS_EXTENT_DATA_REF_KEY;
1342         }
1343         return type;
1344 }
1345
1346 static int find_next_key(struct btrfs_path *path, int level,
1347                          struct btrfs_key *key)
1348
1349 {
1350         for (; level < BTRFS_MAX_LEVEL; level++) {
1351                 if (!path->nodes[level])
1352                         break;
1353                 if (path->slots[level] + 1 >=
1354                     btrfs_header_nritems(path->nodes[level]))
1355                         continue;
1356                 if (level == 0)
1357                         btrfs_item_key_to_cpu(path->nodes[level], key,
1358                                               path->slots[level] + 1);
1359                 else
1360                         btrfs_node_key_to_cpu(path->nodes[level], key,
1361                                               path->slots[level] + 1);
1362                 return 0;
1363         }
1364         return 1;
1365 }
1366
1367 /*
1368  * look for inline back ref. if back ref is found, *ref_ret is set
1369  * to the address of inline back ref, and 0 is returned.
1370  *
1371  * if back ref isn't found, *ref_ret is set to the address where it
1372  * should be inserted, and -ENOENT is returned.
1373  *
1374  * if insert is true and there are too many inline back refs, the path
1375  * points to the extent item, and -EAGAIN is returned.
1376  *
1377  * NOTE: inline back refs are ordered in the same way that back ref
1378  *       items in the tree are ordered.
1379  */
1380 static noinline_for_stack
1381 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1382                                  struct btrfs_root *root,
1383                                  struct btrfs_path *path,
1384                                  struct btrfs_extent_inline_ref **ref_ret,
1385                                  u64 bytenr, u64 num_bytes,
1386                                  u64 parent, u64 root_objectid,
1387                                  u64 owner, u64 offset, int insert)
1388 {
1389         struct btrfs_key key;
1390         struct extent_buffer *leaf;
1391         struct btrfs_extent_item *ei;
1392         struct btrfs_extent_inline_ref *iref;
1393         u64 flags;
1394         u64 item_size;
1395         unsigned long ptr;
1396         unsigned long end;
1397         int extra_size;
1398         int type;
1399         int want;
1400         int ret;
1401         int err = 0;
1402
1403         key.objectid = bytenr;
1404         key.type = BTRFS_EXTENT_ITEM_KEY;
1405         key.offset = num_bytes;
1406
1407         want = extent_ref_type(parent, owner);
1408         if (insert) {
1409                 extra_size = btrfs_extent_inline_ref_size(want);
1410                 path->keep_locks = 1;
1411         } else
1412                 extra_size = -1;
1413         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1414         if (ret < 0) {
1415                 err = ret;
1416                 goto out;
1417         }
1418         BUG_ON(ret);
1419
1420         leaf = path->nodes[0];
1421         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1422 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1423         if (item_size < sizeof(*ei)) {
1424                 if (!insert) {
1425                         err = -ENOENT;
1426                         goto out;
1427                 }
1428                 ret = convert_extent_item_v0(trans, root, path, owner,
1429                                              extra_size);
1430                 if (ret < 0) {
1431                         err = ret;
1432                         goto out;
1433                 }
1434                 leaf = path->nodes[0];
1435                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1436         }
1437 #endif
1438         BUG_ON(item_size < sizeof(*ei));
1439
1440         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1441         flags = btrfs_extent_flags(leaf, ei);
1442
1443         ptr = (unsigned long)(ei + 1);
1444         end = (unsigned long)ei + item_size;
1445
1446         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1447                 ptr += sizeof(struct btrfs_tree_block_info);
1448                 BUG_ON(ptr > end);
1449         } else {
1450                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1451         }
1452
1453         err = -ENOENT;
1454         while (1) {
1455                 if (ptr >= end) {
1456                         WARN_ON(ptr > end);
1457                         break;
1458                 }
1459                 iref = (struct btrfs_extent_inline_ref *)ptr;
1460                 type = btrfs_extent_inline_ref_type(leaf, iref);
1461                 if (want < type)
1462                         break;
1463                 if (want > type) {
1464                         ptr += btrfs_extent_inline_ref_size(type);
1465                         continue;
1466                 }
1467
1468                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1469                         struct btrfs_extent_data_ref *dref;
1470                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1471                         if (match_extent_data_ref(leaf, dref, root_objectid,
1472                                                   owner, offset)) {
1473                                 err = 0;
1474                                 break;
1475                         }
1476                         if (hash_extent_data_ref_item(leaf, dref) <
1477                             hash_extent_data_ref(root_objectid, owner, offset))
1478                                 break;
1479                 } else {
1480                         u64 ref_offset;
1481                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1482                         if (parent > 0) {
1483                                 if (parent == ref_offset) {
1484                                         err = 0;
1485                                         break;
1486                                 }
1487                                 if (ref_offset < parent)
1488                                         break;
1489                         } else {
1490                                 if (root_objectid == ref_offset) {
1491                                         err = 0;
1492                                         break;
1493                                 }
1494                                 if (ref_offset < root_objectid)
1495                                         break;
1496                         }
1497                 }
1498                 ptr += btrfs_extent_inline_ref_size(type);
1499         }
1500         if (err == -ENOENT && insert) {
1501                 if (item_size + extra_size >=
1502                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1503                         err = -EAGAIN;
1504                         goto out;
1505                 }
1506                 /*
1507                  * To add new inline back ref, we have to make sure
1508                  * there is no corresponding back ref item.
1509                  * For simplicity, we just do not add new inline back
1510                  * ref if there is any kind of item for this block
1511                  */
1512                 if (find_next_key(path, 0, &key) == 0 &&
1513                     key.objectid == bytenr &&
1514                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1515                         err = -EAGAIN;
1516                         goto out;
1517                 }
1518         }
1519         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1520 out:
1521         if (insert) {
1522                 path->keep_locks = 0;
1523                 btrfs_unlock_up_safe(path, 1);
1524         }
1525         return err;
1526 }
1527
1528 /*
1529  * helper to add new inline back ref
1530  */
1531 static noinline_for_stack
1532 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1533                                 struct btrfs_root *root,
1534                                 struct btrfs_path *path,
1535                                 struct btrfs_extent_inline_ref *iref,
1536                                 u64 parent, u64 root_objectid,
1537                                 u64 owner, u64 offset, int refs_to_add,
1538                                 struct btrfs_delayed_extent_op *extent_op)
1539 {
1540         struct extent_buffer *leaf;
1541         struct btrfs_extent_item *ei;
1542         unsigned long ptr;
1543         unsigned long end;
1544         unsigned long item_offset;
1545         u64 refs;
1546         int size;
1547         int type;
1548         int ret;
1549
1550         leaf = path->nodes[0];
1551         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1552         item_offset = (unsigned long)iref - (unsigned long)ei;
1553
1554         type = extent_ref_type(parent, owner);
1555         size = btrfs_extent_inline_ref_size(type);
1556
1557         ret = btrfs_extend_item(trans, root, path, size);
1558         BUG_ON(ret);
1559
1560         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1561         refs = btrfs_extent_refs(leaf, ei);
1562         refs += refs_to_add;
1563         btrfs_set_extent_refs(leaf, ei, refs);
1564         if (extent_op)
1565                 __run_delayed_extent_op(extent_op, leaf, ei);
1566
1567         ptr = (unsigned long)ei + item_offset;
1568         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1569         if (ptr < end - size)
1570                 memmove_extent_buffer(leaf, ptr + size, ptr,
1571                                       end - size - ptr);
1572
1573         iref = (struct btrfs_extent_inline_ref *)ptr;
1574         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1575         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1576                 struct btrfs_extent_data_ref *dref;
1577                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1578                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1579                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1580                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1581                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1582         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1583                 struct btrfs_shared_data_ref *sref;
1584                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1585                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1586                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1587         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1588                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1589         } else {
1590                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1591         }
1592         btrfs_mark_buffer_dirty(leaf);
1593         return 0;
1594 }
1595
1596 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1597                                  struct btrfs_root *root,
1598                                  struct btrfs_path *path,
1599                                  struct btrfs_extent_inline_ref **ref_ret,
1600                                  u64 bytenr, u64 num_bytes, u64 parent,
1601                                  u64 root_objectid, u64 owner, u64 offset)
1602 {
1603         int ret;
1604
1605         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1606                                            bytenr, num_bytes, parent,
1607                                            root_objectid, owner, offset, 0);
1608         if (ret != -ENOENT)
1609                 return ret;
1610
1611         btrfs_release_path(root, path);
1612         *ref_ret = NULL;
1613
1614         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1615                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1616                                             root_objectid);
1617         } else {
1618                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1619                                              root_objectid, owner, offset);
1620         }
1621         return ret;
1622 }
1623
1624 /*
1625  * helper to update/remove inline back ref
1626  */
1627 static noinline_for_stack
1628 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1629                                  struct btrfs_root *root,
1630                                  struct btrfs_path *path,
1631                                  struct btrfs_extent_inline_ref *iref,
1632                                  int refs_to_mod,
1633                                  struct btrfs_delayed_extent_op *extent_op)
1634 {
1635         struct extent_buffer *leaf;
1636         struct btrfs_extent_item *ei;
1637         struct btrfs_extent_data_ref *dref = NULL;
1638         struct btrfs_shared_data_ref *sref = NULL;
1639         unsigned long ptr;
1640         unsigned long end;
1641         u32 item_size;
1642         int size;
1643         int type;
1644         int ret;
1645         u64 refs;
1646
1647         leaf = path->nodes[0];
1648         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1649         refs = btrfs_extent_refs(leaf, ei);
1650         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1651         refs += refs_to_mod;
1652         btrfs_set_extent_refs(leaf, ei, refs);
1653         if (extent_op)
1654                 __run_delayed_extent_op(extent_op, leaf, ei);
1655
1656         type = btrfs_extent_inline_ref_type(leaf, iref);
1657
1658         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1659                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1660                 refs = btrfs_extent_data_ref_count(leaf, dref);
1661         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1662                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1663                 refs = btrfs_shared_data_ref_count(leaf, sref);
1664         } else {
1665                 refs = 1;
1666                 BUG_ON(refs_to_mod != -1);
1667         }
1668
1669         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1670         refs += refs_to_mod;
1671
1672         if (refs > 0) {
1673                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1674                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1675                 else
1676                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1677         } else {
1678                 size =  btrfs_extent_inline_ref_size(type);
1679                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1680                 ptr = (unsigned long)iref;
1681                 end = (unsigned long)ei + item_size;
1682                 if (ptr + size < end)
1683                         memmove_extent_buffer(leaf, ptr, ptr + size,
1684                                               end - ptr - size);
1685                 item_size -= size;
1686                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1687                 BUG_ON(ret);
1688         }
1689         btrfs_mark_buffer_dirty(leaf);
1690         return 0;
1691 }
1692
1693 static noinline_for_stack
1694 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1695                                  struct btrfs_root *root,
1696                                  struct btrfs_path *path,
1697                                  u64 bytenr, u64 num_bytes, u64 parent,
1698                                  u64 root_objectid, u64 owner,
1699                                  u64 offset, int refs_to_add,
1700                                  struct btrfs_delayed_extent_op *extent_op)
1701 {
1702         struct btrfs_extent_inline_ref *iref;
1703         int ret;
1704
1705         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1706                                            bytenr, num_bytes, parent,
1707                                            root_objectid, owner, offset, 1);
1708         if (ret == 0) {
1709                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1710                 ret = update_inline_extent_backref(trans, root, path, iref,
1711                                                    refs_to_add, extent_op);
1712         } else if (ret == -ENOENT) {
1713                 ret = setup_inline_extent_backref(trans, root, path, iref,
1714                                                   parent, root_objectid,
1715                                                   owner, offset, refs_to_add,
1716                                                   extent_op);
1717         }
1718         return ret;
1719 }
1720
1721 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1722                                  struct btrfs_root *root,
1723                                  struct btrfs_path *path,
1724                                  u64 bytenr, u64 parent, u64 root_objectid,
1725                                  u64 owner, u64 offset, int refs_to_add)
1726 {
1727         int ret;
1728         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1729                 BUG_ON(refs_to_add != 1);
1730                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1731                                             parent, root_objectid);
1732         } else {
1733                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1734                                              parent, root_objectid,
1735                                              owner, offset, refs_to_add);
1736         }
1737         return ret;
1738 }
1739
1740 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1741                                  struct btrfs_root *root,
1742                                  struct btrfs_path *path,
1743                                  struct btrfs_extent_inline_ref *iref,
1744                                  int refs_to_drop, int is_data)
1745 {
1746         int ret;
1747
1748         BUG_ON(!is_data && refs_to_drop != 1);
1749         if (iref) {
1750                 ret = update_inline_extent_backref(trans, root, path, iref,
1751                                                    -refs_to_drop, NULL);
1752         } else if (is_data) {
1753                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1754         } else {
1755                 ret = btrfs_del_item(trans, root, path);
1756         }
1757         return ret;
1758 }
1759
1760 static int btrfs_issue_discard(struct block_device *bdev,
1761                                 u64 start, u64 len)
1762 {
1763         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1764 }
1765
1766 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1767                                 u64 num_bytes, u64 *actual_bytes)
1768 {
1769         int ret;
1770         u64 discarded_bytes = 0;
1771         struct btrfs_multi_bio *multi = NULL;
1772
1773
1774         /* Tell the block device(s) that the sectors can be discarded */
1775         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1776                               bytenr, &num_bytes, &multi, 0);
1777         if (!ret) {
1778                 struct btrfs_bio_stripe *stripe = multi->stripes;
1779                 int i;
1780
1781
1782                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1783                         ret = btrfs_issue_discard(stripe->dev->bdev,
1784                                                   stripe->physical,
1785                                                   stripe->length);
1786                         if (!ret)
1787                                 discarded_bytes += stripe->length;
1788                         else if (ret != -EOPNOTSUPP)
1789                                 break;
1790                 }
1791                 kfree(multi);
1792         }
1793         if (discarded_bytes && ret == -EOPNOTSUPP)
1794                 ret = 0;
1795
1796         if (actual_bytes)
1797                 *actual_bytes = discarded_bytes;
1798
1799
1800         return ret;
1801 }
1802
1803 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1804                          struct btrfs_root *root,
1805                          u64 bytenr, u64 num_bytes, u64 parent,
1806                          u64 root_objectid, u64 owner, u64 offset)
1807 {
1808         int ret;
1809         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1810                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1811
1812         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1813                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1814                                         parent, root_objectid, (int)owner,
1815                                         BTRFS_ADD_DELAYED_REF, NULL);
1816         } else {
1817                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1818                                         parent, root_objectid, owner, offset,
1819                                         BTRFS_ADD_DELAYED_REF, NULL);
1820         }
1821         return ret;
1822 }
1823
1824 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1825                                   struct btrfs_root *root,
1826                                   u64 bytenr, u64 num_bytes,
1827                                   u64 parent, u64 root_objectid,
1828                                   u64 owner, u64 offset, int refs_to_add,
1829                                   struct btrfs_delayed_extent_op *extent_op)
1830 {
1831         struct btrfs_path *path;
1832         struct extent_buffer *leaf;
1833         struct btrfs_extent_item *item;
1834         u64 refs;
1835         int ret;
1836         int err = 0;
1837
1838         path = btrfs_alloc_path();
1839         if (!path)
1840                 return -ENOMEM;
1841
1842         path->reada = 1;
1843         path->leave_spinning = 1;
1844         /* this will setup the path even if it fails to insert the back ref */
1845         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1846                                            path, bytenr, num_bytes, parent,
1847                                            root_objectid, owner, offset,
1848                                            refs_to_add, extent_op);
1849         if (ret == 0)
1850                 goto out;
1851
1852         if (ret != -EAGAIN) {
1853                 err = ret;
1854                 goto out;
1855         }
1856
1857         leaf = path->nodes[0];
1858         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1859         refs = btrfs_extent_refs(leaf, item);
1860         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1861         if (extent_op)
1862                 __run_delayed_extent_op(extent_op, leaf, item);
1863
1864         btrfs_mark_buffer_dirty(leaf);
1865         btrfs_release_path(root->fs_info->extent_root, path);
1866
1867         path->reada = 1;
1868         path->leave_spinning = 1;
1869
1870         /* now insert the actual backref */
1871         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1872                                     path, bytenr, parent, root_objectid,
1873                                     owner, offset, refs_to_add);
1874         BUG_ON(ret);
1875 out:
1876         btrfs_free_path(path);
1877         return err;
1878 }
1879
1880 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1881                                 struct btrfs_root *root,
1882                                 struct btrfs_delayed_ref_node *node,
1883                                 struct btrfs_delayed_extent_op *extent_op,
1884                                 int insert_reserved)
1885 {
1886         int ret = 0;
1887         struct btrfs_delayed_data_ref *ref;
1888         struct btrfs_key ins;
1889         u64 parent = 0;
1890         u64 ref_root = 0;
1891         u64 flags = 0;
1892
1893         ins.objectid = node->bytenr;
1894         ins.offset = node->num_bytes;
1895         ins.type = BTRFS_EXTENT_ITEM_KEY;
1896
1897         ref = btrfs_delayed_node_to_data_ref(node);
1898         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1899                 parent = ref->parent;
1900         else
1901                 ref_root = ref->root;
1902
1903         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1904                 if (extent_op) {
1905                         BUG_ON(extent_op->update_key);
1906                         flags |= extent_op->flags_to_set;
1907                 }
1908                 ret = alloc_reserved_file_extent(trans, root,
1909                                                  parent, ref_root, flags,
1910                                                  ref->objectid, ref->offset,
1911                                                  &ins, node->ref_mod);
1912         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1913                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1914                                              node->num_bytes, parent,
1915                                              ref_root, ref->objectid,
1916                                              ref->offset, node->ref_mod,
1917                                              extent_op);
1918         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1919                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1920                                           node->num_bytes, parent,
1921                                           ref_root, ref->objectid,
1922                                           ref->offset, node->ref_mod,
1923                                           extent_op);
1924         } else {
1925                 BUG();
1926         }
1927         return ret;
1928 }
1929
1930 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1931                                     struct extent_buffer *leaf,
1932                                     struct btrfs_extent_item *ei)
1933 {
1934         u64 flags = btrfs_extent_flags(leaf, ei);
1935         if (extent_op->update_flags) {
1936                 flags |= extent_op->flags_to_set;
1937                 btrfs_set_extent_flags(leaf, ei, flags);
1938         }
1939
1940         if (extent_op->update_key) {
1941                 struct btrfs_tree_block_info *bi;
1942                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1943                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1944                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1945         }
1946 }
1947
1948 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1949                                  struct btrfs_root *root,
1950                                  struct btrfs_delayed_ref_node *node,
1951                                  struct btrfs_delayed_extent_op *extent_op)
1952 {
1953         struct btrfs_key key;
1954         struct btrfs_path *path;
1955         struct btrfs_extent_item *ei;
1956         struct extent_buffer *leaf;
1957         u32 item_size;
1958         int ret;
1959         int err = 0;
1960
1961         path = btrfs_alloc_path();
1962         if (!path)
1963                 return -ENOMEM;
1964
1965         key.objectid = node->bytenr;
1966         key.type = BTRFS_EXTENT_ITEM_KEY;
1967         key.offset = node->num_bytes;
1968
1969         path->reada = 1;
1970         path->leave_spinning = 1;
1971         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1972                                 path, 0, 1);
1973         if (ret < 0) {
1974                 err = ret;
1975                 goto out;
1976         }
1977         if (ret > 0) {
1978                 err = -EIO;
1979                 goto out;
1980         }
1981
1982         leaf = path->nodes[0];
1983         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1984 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1985         if (item_size < sizeof(*ei)) {
1986                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1987                                              path, (u64)-1, 0);
1988                 if (ret < 0) {
1989                         err = ret;
1990                         goto out;
1991                 }
1992                 leaf = path->nodes[0];
1993                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1994         }
1995 #endif
1996         BUG_ON(item_size < sizeof(*ei));
1997         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1998         __run_delayed_extent_op(extent_op, leaf, ei);
1999
2000         btrfs_mark_buffer_dirty(leaf);
2001 out:
2002         btrfs_free_path(path);
2003         return err;
2004 }
2005
2006 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2007                                 struct btrfs_root *root,
2008                                 struct btrfs_delayed_ref_node *node,
2009                                 struct btrfs_delayed_extent_op *extent_op,
2010                                 int insert_reserved)
2011 {
2012         int ret = 0;
2013         struct btrfs_delayed_tree_ref *ref;
2014         struct btrfs_key ins;
2015         u64 parent = 0;
2016         u64 ref_root = 0;
2017
2018         ins.objectid = node->bytenr;
2019         ins.offset = node->num_bytes;
2020         ins.type = BTRFS_EXTENT_ITEM_KEY;
2021
2022         ref = btrfs_delayed_node_to_tree_ref(node);
2023         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2024                 parent = ref->parent;
2025         else
2026                 ref_root = ref->root;
2027
2028         BUG_ON(node->ref_mod != 1);
2029         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2030                 BUG_ON(!extent_op || !extent_op->update_flags ||
2031                        !extent_op->update_key);
2032                 ret = alloc_reserved_tree_block(trans, root,
2033                                                 parent, ref_root,
2034                                                 extent_op->flags_to_set,
2035                                                 &extent_op->key,
2036                                                 ref->level, &ins);
2037         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2038                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2039                                              node->num_bytes, parent, ref_root,
2040                                              ref->level, 0, 1, extent_op);
2041         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2042                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2043                                           node->num_bytes, parent, ref_root,
2044                                           ref->level, 0, 1, extent_op);
2045         } else {
2046                 BUG();
2047         }
2048         return ret;
2049 }
2050
2051 /* helper function to actually process a single delayed ref entry */
2052 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2053                                struct btrfs_root *root,
2054                                struct btrfs_delayed_ref_node *node,
2055                                struct btrfs_delayed_extent_op *extent_op,
2056                                int insert_reserved)
2057 {
2058         int ret;
2059         if (btrfs_delayed_ref_is_head(node)) {
2060                 struct btrfs_delayed_ref_head *head;
2061                 /*
2062                  * we've hit the end of the chain and we were supposed
2063                  * to insert this extent into the tree.  But, it got
2064                  * deleted before we ever needed to insert it, so all
2065                  * we have to do is clean up the accounting
2066                  */
2067                 BUG_ON(extent_op);
2068                 head = btrfs_delayed_node_to_head(node);
2069                 if (insert_reserved) {
2070                         btrfs_pin_extent(root, node->bytenr,
2071                                          node->num_bytes, 1);
2072                         if (head->is_data) {
2073                                 ret = btrfs_del_csums(trans, root,
2074                                                       node->bytenr,
2075                                                       node->num_bytes);
2076                                 BUG_ON(ret);
2077                         }
2078                 }
2079                 mutex_unlock(&head->mutex);
2080                 return 0;
2081         }
2082
2083         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2084             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2085                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2086                                            insert_reserved);
2087         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2088                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2089                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2090                                            insert_reserved);
2091         else
2092                 BUG();
2093         return ret;
2094 }
2095
2096 static noinline struct btrfs_delayed_ref_node *
2097 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2098 {
2099         struct rb_node *node;
2100         struct btrfs_delayed_ref_node *ref;
2101         int action = BTRFS_ADD_DELAYED_REF;
2102 again:
2103         /*
2104          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2105          * this prevents ref count from going down to zero when
2106          * there still are pending delayed ref.
2107          */
2108         node = rb_prev(&head->node.rb_node);
2109         while (1) {
2110                 if (!node)
2111                         break;
2112                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2113                                 rb_node);
2114                 if (ref->bytenr != head->node.bytenr)
2115                         break;
2116                 if (ref->action == action)
2117                         return ref;
2118                 node = rb_prev(node);
2119         }
2120         if (action == BTRFS_ADD_DELAYED_REF) {
2121                 action = BTRFS_DROP_DELAYED_REF;
2122                 goto again;
2123         }
2124         return NULL;
2125 }
2126
2127 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2128                                        struct btrfs_root *root,
2129                                        struct list_head *cluster)
2130 {
2131         struct btrfs_delayed_ref_root *delayed_refs;
2132         struct btrfs_delayed_ref_node *ref;
2133         struct btrfs_delayed_ref_head *locked_ref = NULL;
2134         struct btrfs_delayed_extent_op *extent_op;
2135         int ret;
2136         int count = 0;
2137         int must_insert_reserved = 0;
2138
2139         delayed_refs = &trans->transaction->delayed_refs;
2140         while (1) {
2141                 if (!locked_ref) {
2142                         /* pick a new head ref from the cluster list */
2143                         if (list_empty(cluster))
2144                                 break;
2145
2146                         locked_ref = list_entry(cluster->next,
2147                                      struct btrfs_delayed_ref_head, cluster);
2148
2149                         /* grab the lock that says we are going to process
2150                          * all the refs for this head */
2151                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2152
2153                         /*
2154                          * we may have dropped the spin lock to get the head
2155                          * mutex lock, and that might have given someone else
2156                          * time to free the head.  If that's true, it has been
2157                          * removed from our list and we can move on.
2158                          */
2159                         if (ret == -EAGAIN) {
2160                                 locked_ref = NULL;
2161                                 count++;
2162                                 continue;
2163                         }
2164                 }
2165
2166                 /*
2167                  * record the must insert reserved flag before we
2168                  * drop the spin lock.
2169                  */
2170                 must_insert_reserved = locked_ref->must_insert_reserved;
2171                 locked_ref->must_insert_reserved = 0;
2172
2173                 extent_op = locked_ref->extent_op;
2174                 locked_ref->extent_op = NULL;
2175
2176                 /*
2177                  * locked_ref is the head node, so we have to go one
2178                  * node back for any delayed ref updates
2179                  */
2180                 ref = select_delayed_ref(locked_ref);
2181                 if (!ref) {
2182                         /* All delayed refs have been processed, Go ahead
2183                          * and send the head node to run_one_delayed_ref,
2184                          * so that any accounting fixes can happen
2185                          */
2186                         ref = &locked_ref->node;
2187
2188                         if (extent_op && must_insert_reserved) {
2189                                 kfree(extent_op);
2190                                 extent_op = NULL;
2191                         }
2192
2193                         if (extent_op) {
2194                                 spin_unlock(&delayed_refs->lock);
2195
2196                                 ret = run_delayed_extent_op(trans, root,
2197                                                             ref, extent_op);
2198                                 BUG_ON(ret);
2199                                 kfree(extent_op);
2200
2201                                 cond_resched();
2202                                 spin_lock(&delayed_refs->lock);
2203                                 continue;
2204                         }
2205
2206                         list_del_init(&locked_ref->cluster);
2207                         locked_ref = NULL;
2208                 }
2209
2210                 ref->in_tree = 0;
2211                 rb_erase(&ref->rb_node, &delayed_refs->root);
2212                 delayed_refs->num_entries--;
2213
2214                 spin_unlock(&delayed_refs->lock);
2215
2216                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2217                                           must_insert_reserved);
2218                 BUG_ON(ret);
2219
2220                 btrfs_put_delayed_ref(ref);
2221                 kfree(extent_op);
2222                 count++;
2223
2224                 cond_resched();
2225                 spin_lock(&delayed_refs->lock);
2226         }
2227         return count;
2228 }
2229
2230 /*
2231  * this starts processing the delayed reference count updates and
2232  * extent insertions we have queued up so far.  count can be
2233  * 0, which means to process everything in the tree at the start
2234  * of the run (but not newly added entries), or it can be some target
2235  * number you'd like to process.
2236  */
2237 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2238                            struct btrfs_root *root, unsigned long count)
2239 {
2240         struct rb_node *node;
2241         struct btrfs_delayed_ref_root *delayed_refs;
2242         struct btrfs_delayed_ref_node *ref;
2243         struct list_head cluster;
2244         int ret;
2245         int run_all = count == (unsigned long)-1;
2246         int run_most = 0;
2247
2248         if (root == root->fs_info->extent_root)
2249                 root = root->fs_info->tree_root;
2250
2251         delayed_refs = &trans->transaction->delayed_refs;
2252         INIT_LIST_HEAD(&cluster);
2253 again:
2254         spin_lock(&delayed_refs->lock);
2255         if (count == 0) {
2256                 count = delayed_refs->num_entries * 2;
2257                 run_most = 1;
2258         }
2259         while (1) {
2260                 if (!(run_all || run_most) &&
2261                     delayed_refs->num_heads_ready < 64)
2262                         break;
2263
2264                 /*
2265                  * go find something we can process in the rbtree.  We start at
2266                  * the beginning of the tree, and then build a cluster
2267                  * of refs to process starting at the first one we are able to
2268                  * lock
2269                  */
2270                 ret = btrfs_find_ref_cluster(trans, &cluster,
2271                                              delayed_refs->run_delayed_start);
2272                 if (ret)
2273                         break;
2274
2275                 ret = run_clustered_refs(trans, root, &cluster);
2276                 BUG_ON(ret < 0);
2277
2278                 count -= min_t(unsigned long, ret, count);
2279
2280                 if (count == 0)
2281                         break;
2282         }
2283
2284         if (run_all) {
2285                 node = rb_first(&delayed_refs->root);
2286                 if (!node)
2287                         goto out;
2288                 count = (unsigned long)-1;
2289
2290                 while (node) {
2291                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2292                                        rb_node);
2293                         if (btrfs_delayed_ref_is_head(ref)) {
2294                                 struct btrfs_delayed_ref_head *head;
2295
2296                                 head = btrfs_delayed_node_to_head(ref);
2297                                 atomic_inc(&ref->refs);
2298
2299                                 spin_unlock(&delayed_refs->lock);
2300                                 mutex_lock(&head->mutex);
2301                                 mutex_unlock(&head->mutex);
2302
2303                                 btrfs_put_delayed_ref(ref);
2304                                 cond_resched();
2305                                 goto again;
2306                         }
2307                         node = rb_next(node);
2308                 }
2309                 spin_unlock(&delayed_refs->lock);
2310                 schedule_timeout(1);
2311                 goto again;
2312         }
2313 out:
2314         spin_unlock(&delayed_refs->lock);
2315         return 0;
2316 }
2317
2318 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2319                                 struct btrfs_root *root,
2320                                 u64 bytenr, u64 num_bytes, u64 flags,
2321                                 int is_data)
2322 {
2323         struct btrfs_delayed_extent_op *extent_op;
2324         int ret;
2325
2326         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2327         if (!extent_op)
2328                 return -ENOMEM;
2329
2330         extent_op->flags_to_set = flags;
2331         extent_op->update_flags = 1;
2332         extent_op->update_key = 0;
2333         extent_op->is_data = is_data ? 1 : 0;
2334
2335         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2336         if (ret)
2337                 kfree(extent_op);
2338         return ret;
2339 }
2340
2341 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2342                                       struct btrfs_root *root,
2343                                       struct btrfs_path *path,
2344                                       u64 objectid, u64 offset, u64 bytenr)
2345 {
2346         struct btrfs_delayed_ref_head *head;
2347         struct btrfs_delayed_ref_node *ref;
2348         struct btrfs_delayed_data_ref *data_ref;
2349         struct btrfs_delayed_ref_root *delayed_refs;
2350         struct rb_node *node;
2351         int ret = 0;
2352
2353         ret = -ENOENT;
2354         delayed_refs = &trans->transaction->delayed_refs;
2355         spin_lock(&delayed_refs->lock);
2356         head = btrfs_find_delayed_ref_head(trans, bytenr);
2357         if (!head)
2358                 goto out;
2359
2360         if (!mutex_trylock(&head->mutex)) {
2361                 atomic_inc(&head->node.refs);
2362                 spin_unlock(&delayed_refs->lock);
2363
2364                 btrfs_release_path(root->fs_info->extent_root, path);
2365
2366                 mutex_lock(&head->mutex);
2367                 mutex_unlock(&head->mutex);
2368                 btrfs_put_delayed_ref(&head->node);
2369                 return -EAGAIN;
2370         }
2371
2372         node = rb_prev(&head->node.rb_node);
2373         if (!node)
2374                 goto out_unlock;
2375
2376         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2377
2378         if (ref->bytenr != bytenr)
2379                 goto out_unlock;
2380
2381         ret = 1;
2382         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2383                 goto out_unlock;
2384
2385         data_ref = btrfs_delayed_node_to_data_ref(ref);
2386
2387         node = rb_prev(node);
2388         if (node) {
2389                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2390                 if (ref->bytenr == bytenr)
2391                         goto out_unlock;
2392         }
2393
2394         if (data_ref->root != root->root_key.objectid ||
2395             data_ref->objectid != objectid || data_ref->offset != offset)
2396                 goto out_unlock;
2397
2398         ret = 0;
2399 out_unlock:
2400         mutex_unlock(&head->mutex);
2401 out:
2402         spin_unlock(&delayed_refs->lock);
2403         return ret;
2404 }
2405
2406 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2407                                         struct btrfs_root *root,
2408                                         struct btrfs_path *path,
2409                                         u64 objectid, u64 offset, u64 bytenr)
2410 {
2411         struct btrfs_root *extent_root = root->fs_info->extent_root;
2412         struct extent_buffer *leaf;
2413         struct btrfs_extent_data_ref *ref;
2414         struct btrfs_extent_inline_ref *iref;
2415         struct btrfs_extent_item *ei;
2416         struct btrfs_key key;
2417         u32 item_size;
2418         int ret;
2419
2420         key.objectid = bytenr;
2421         key.offset = (u64)-1;
2422         key.type = BTRFS_EXTENT_ITEM_KEY;
2423
2424         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2425         if (ret < 0)
2426                 goto out;
2427         BUG_ON(ret == 0);
2428
2429         ret = -ENOENT;
2430         if (path->slots[0] == 0)
2431                 goto out;
2432
2433         path->slots[0]--;
2434         leaf = path->nodes[0];
2435         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2436
2437         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2438                 goto out;
2439
2440         ret = 1;
2441         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2442 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2443         if (item_size < sizeof(*ei)) {
2444                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2445                 goto out;
2446         }
2447 #endif
2448         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2449
2450         if (item_size != sizeof(*ei) +
2451             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2452                 goto out;
2453
2454         if (btrfs_extent_generation(leaf, ei) <=
2455             btrfs_root_last_snapshot(&root->root_item))
2456                 goto out;
2457
2458         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2459         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2460             BTRFS_EXTENT_DATA_REF_KEY)
2461                 goto out;
2462
2463         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2464         if (btrfs_extent_refs(leaf, ei) !=
2465             btrfs_extent_data_ref_count(leaf, ref) ||
2466             btrfs_extent_data_ref_root(leaf, ref) !=
2467             root->root_key.objectid ||
2468             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2469             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2470                 goto out;
2471
2472         ret = 0;
2473 out:
2474         return ret;
2475 }
2476
2477 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2478                           struct btrfs_root *root,
2479                           u64 objectid, u64 offset, u64 bytenr)
2480 {
2481         struct btrfs_path *path;
2482         int ret;
2483         int ret2;
2484
2485         path = btrfs_alloc_path();
2486         if (!path)
2487                 return -ENOENT;
2488
2489         do {
2490                 ret = check_committed_ref(trans, root, path, objectid,
2491                                           offset, bytenr);
2492                 if (ret && ret != -ENOENT)
2493                         goto out;
2494
2495                 ret2 = check_delayed_ref(trans, root, path, objectid,
2496                                          offset, bytenr);
2497         } while (ret2 == -EAGAIN);
2498
2499         if (ret2 && ret2 != -ENOENT) {
2500                 ret = ret2;
2501                 goto out;
2502         }
2503
2504         if (ret != -ENOENT || ret2 != -ENOENT)
2505                 ret = 0;
2506 out:
2507         btrfs_free_path(path);
2508         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2509                 WARN_ON(ret > 0);
2510         return ret;
2511 }
2512
2513 #if 0
2514 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2515                     struct extent_buffer *buf, u32 nr_extents)
2516 {
2517         struct btrfs_key key;
2518         struct btrfs_file_extent_item *fi;
2519         u64 root_gen;
2520         u32 nritems;
2521         int i;
2522         int level;
2523         int ret = 0;
2524         int shared = 0;
2525
2526         if (!root->ref_cows)
2527                 return 0;
2528
2529         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2530                 shared = 0;
2531                 root_gen = root->root_key.offset;
2532         } else {
2533                 shared = 1;
2534                 root_gen = trans->transid - 1;
2535         }
2536
2537         level = btrfs_header_level(buf);
2538         nritems = btrfs_header_nritems(buf);
2539
2540         if (level == 0) {
2541                 struct btrfs_leaf_ref *ref;
2542                 struct btrfs_extent_info *info;
2543
2544                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2545                 if (!ref) {
2546                         ret = -ENOMEM;
2547                         goto out;
2548                 }
2549
2550                 ref->root_gen = root_gen;
2551                 ref->bytenr = buf->start;
2552                 ref->owner = btrfs_header_owner(buf);
2553                 ref->generation = btrfs_header_generation(buf);
2554                 ref->nritems = nr_extents;
2555                 info = ref->extents;
2556
2557                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2558                         u64 disk_bytenr;
2559                         btrfs_item_key_to_cpu(buf, &key, i);
2560                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2561                                 continue;
2562                         fi = btrfs_item_ptr(buf, i,
2563                                             struct btrfs_file_extent_item);
2564                         if (btrfs_file_extent_type(buf, fi) ==
2565                             BTRFS_FILE_EXTENT_INLINE)
2566                                 continue;
2567                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2568                         if (disk_bytenr == 0)
2569                                 continue;
2570
2571                         info->bytenr = disk_bytenr;
2572                         info->num_bytes =
2573                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2574                         info->objectid = key.objectid;
2575                         info->offset = key.offset;
2576                         info++;
2577                 }
2578
2579                 ret = btrfs_add_leaf_ref(root, ref, shared);
2580                 if (ret == -EEXIST && shared) {
2581                         struct btrfs_leaf_ref *old;
2582                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2583                         BUG_ON(!old);
2584                         btrfs_remove_leaf_ref(root, old);
2585                         btrfs_free_leaf_ref(root, old);
2586                         ret = btrfs_add_leaf_ref(root, ref, shared);
2587                 }
2588                 WARN_ON(ret);
2589                 btrfs_free_leaf_ref(root, ref);
2590         }
2591 out:
2592         return ret;
2593 }
2594
2595 /* when a block goes through cow, we update the reference counts of
2596  * everything that block points to.  The internal pointers of the block
2597  * can be in just about any order, and it is likely to have clusters of
2598  * things that are close together and clusters of things that are not.
2599  *
2600  * To help reduce the seeks that come with updating all of these reference
2601  * counts, sort them by byte number before actual updates are done.
2602  *
2603  * struct refsort is used to match byte number to slot in the btree block.
2604  * we sort based on the byte number and then use the slot to actually
2605  * find the item.
2606  *
2607  * struct refsort is smaller than strcut btrfs_item and smaller than
2608  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2609  * for a btree block, there's no way for a kmalloc of refsorts for a
2610  * single node to be bigger than a page.
2611  */
2612 struct refsort {
2613         u64 bytenr;
2614         u32 slot;
2615 };
2616
2617 /*
2618  * for passing into sort()
2619  */
2620 static int refsort_cmp(const void *a_void, const void *b_void)
2621 {
2622         const struct refsort *a = a_void;
2623         const struct refsort *b = b_void;
2624
2625         if (a->bytenr < b->bytenr)
2626                 return -1;
2627         if (a->bytenr > b->bytenr)
2628                 return 1;
2629         return 0;
2630 }
2631 #endif
2632
2633 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2634                            struct btrfs_root *root,
2635                            struct extent_buffer *buf,
2636                            int full_backref, int inc)
2637 {
2638         u64 bytenr;
2639         u64 num_bytes;
2640         u64 parent;
2641         u64 ref_root;
2642         u32 nritems;
2643         struct btrfs_key key;
2644         struct btrfs_file_extent_item *fi;
2645         int i;
2646         int level;
2647         int ret = 0;
2648         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2649                             u64, u64, u64, u64, u64, u64);
2650
2651         ref_root = btrfs_header_owner(buf);
2652         nritems = btrfs_header_nritems(buf);
2653         level = btrfs_header_level(buf);
2654
2655         if (!root->ref_cows && level == 0)
2656                 return 0;
2657
2658         if (inc)
2659                 process_func = btrfs_inc_extent_ref;
2660         else
2661                 process_func = btrfs_free_extent;
2662
2663         if (full_backref)
2664                 parent = buf->start;
2665         else
2666                 parent = 0;
2667
2668         for (i = 0; i < nritems; i++) {
2669                 if (level == 0) {
2670                         btrfs_item_key_to_cpu(buf, &key, i);
2671                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2672                                 continue;
2673                         fi = btrfs_item_ptr(buf, i,
2674                                             struct btrfs_file_extent_item);
2675                         if (btrfs_file_extent_type(buf, fi) ==
2676                             BTRFS_FILE_EXTENT_INLINE)
2677                                 continue;
2678                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2679                         if (bytenr == 0)
2680                                 continue;
2681
2682                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2683                         key.offset -= btrfs_file_extent_offset(buf, fi);
2684                         ret = process_func(trans, root, bytenr, num_bytes,
2685                                            parent, ref_root, key.objectid,
2686                                            key.offset);
2687                         if (ret)
2688                                 goto fail;
2689                 } else {
2690                         bytenr = btrfs_node_blockptr(buf, i);
2691                         num_bytes = btrfs_level_size(root, level - 1);
2692                         ret = process_func(trans, root, bytenr, num_bytes,
2693                                            parent, ref_root, level - 1, 0);
2694                         if (ret)
2695                                 goto fail;
2696                 }
2697         }
2698         return 0;
2699 fail:
2700         BUG();
2701         return ret;
2702 }
2703
2704 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2705                   struct extent_buffer *buf, int full_backref)
2706 {
2707         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2708 }
2709
2710 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2711                   struct extent_buffer *buf, int full_backref)
2712 {
2713         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2714 }
2715
2716 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2717                                  struct btrfs_root *root,
2718                                  struct btrfs_path *path,
2719                                  struct btrfs_block_group_cache *cache)
2720 {
2721         int ret;
2722         struct btrfs_root *extent_root = root->fs_info->extent_root;
2723         unsigned long bi;
2724         struct extent_buffer *leaf;
2725
2726         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2727         if (ret < 0)
2728                 goto fail;
2729         BUG_ON(ret);
2730
2731         leaf = path->nodes[0];
2732         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2733         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2734         btrfs_mark_buffer_dirty(leaf);
2735         btrfs_release_path(extent_root, path);
2736 fail:
2737         if (ret)
2738                 return ret;
2739         return 0;
2740
2741 }
2742
2743 static struct btrfs_block_group_cache *
2744 next_block_group(struct btrfs_root *root,
2745                  struct btrfs_block_group_cache *cache)
2746 {
2747         struct rb_node *node;
2748         spin_lock(&root->fs_info->block_group_cache_lock);
2749         node = rb_next(&cache->cache_node);
2750         btrfs_put_block_group(cache);
2751         if (node) {
2752                 cache = rb_entry(node, struct btrfs_block_group_cache,
2753                                  cache_node);
2754                 btrfs_get_block_group(cache);
2755         } else
2756                 cache = NULL;
2757         spin_unlock(&root->fs_info->block_group_cache_lock);
2758         return cache;
2759 }
2760
2761 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2762                             struct btrfs_trans_handle *trans,
2763                             struct btrfs_path *path)
2764 {
2765         struct btrfs_root *root = block_group->fs_info->tree_root;
2766         struct inode *inode = NULL;
2767         u64 alloc_hint = 0;
2768         int dcs = BTRFS_DC_ERROR;
2769         int num_pages = 0;
2770         int retries = 0;
2771         int ret = 0;
2772
2773         /*
2774          * If this block group is smaller than 100 megs don't bother caching the
2775          * block group.
2776          */
2777         if (block_group->key.offset < (100 * 1024 * 1024)) {
2778                 spin_lock(&block_group->lock);
2779                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2780                 spin_unlock(&block_group->lock);
2781                 return 0;
2782         }
2783
2784 again:
2785         inode = lookup_free_space_inode(root, block_group, path);
2786         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2787                 ret = PTR_ERR(inode);
2788                 btrfs_release_path(root, path);
2789                 goto out;
2790         }
2791
2792         if (IS_ERR(inode)) {
2793                 BUG_ON(retries);
2794                 retries++;
2795
2796                 if (block_group->ro)
2797                         goto out_free;
2798
2799                 ret = create_free_space_inode(root, trans, block_group, path);
2800                 if (ret)
2801                         goto out_free;
2802                 goto again;
2803         }
2804
2805         /*
2806          * We want to set the generation to 0, that way if anything goes wrong
2807          * from here on out we know not to trust this cache when we load up next
2808          * time.
2809          */
2810         BTRFS_I(inode)->generation = 0;
2811         ret = btrfs_update_inode(trans, root, inode);
2812         WARN_ON(ret);
2813
2814         if (i_size_read(inode) > 0) {
2815                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2816                                                       inode);
2817                 if (ret)
2818                         goto out_put;
2819         }
2820
2821         spin_lock(&block_group->lock);
2822         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2823                 /* We're not cached, don't bother trying to write stuff out */
2824                 dcs = BTRFS_DC_WRITTEN;
2825                 spin_unlock(&block_group->lock);
2826                 goto out_put;
2827         }
2828         spin_unlock(&block_group->lock);
2829
2830         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2831         if (!num_pages)
2832                 num_pages = 1;
2833
2834         /*
2835          * Just to make absolutely sure we have enough space, we're going to
2836          * preallocate 12 pages worth of space for each block group.  In
2837          * practice we ought to use at most 8, but we need extra space so we can
2838          * add our header and have a terminator between the extents and the
2839          * bitmaps.
2840          */
2841         num_pages *= 16;
2842         num_pages *= PAGE_CACHE_SIZE;
2843
2844         ret = btrfs_check_data_free_space(inode, num_pages);
2845         if (ret)
2846                 goto out_put;
2847
2848         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2849                                               num_pages, num_pages,
2850                                               &alloc_hint);
2851         if (!ret)
2852                 dcs = BTRFS_DC_SETUP;
2853         btrfs_free_reserved_data_space(inode, num_pages);
2854 out_put:
2855         iput(inode);
2856 out_free:
2857         btrfs_release_path(root, path);
2858 out:
2859         spin_lock(&block_group->lock);
2860         block_group->disk_cache_state = dcs;
2861         spin_unlock(&block_group->lock);
2862
2863         return ret;
2864 }
2865
2866 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2867                                    struct btrfs_root *root)
2868 {
2869         struct btrfs_block_group_cache *cache;
2870         int err = 0;
2871         struct btrfs_path *path;
2872         u64 last = 0;
2873
2874         path = btrfs_alloc_path();
2875         if (!path)
2876                 return -ENOMEM;
2877
2878 again:
2879         while (1) {
2880                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2881                 while (cache) {
2882                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2883                                 break;
2884                         cache = next_block_group(root, cache);
2885                 }
2886                 if (!cache) {
2887                         if (last == 0)
2888                                 break;
2889                         last = 0;
2890                         continue;
2891                 }
2892                 err = cache_save_setup(cache, trans, path);
2893                 last = cache->key.objectid + cache->key.offset;
2894                 btrfs_put_block_group(cache);
2895         }
2896
2897         while (1) {
2898                 if (last == 0) {
2899                         err = btrfs_run_delayed_refs(trans, root,
2900                                                      (unsigned long)-1);
2901                         BUG_ON(err);
2902                 }
2903
2904                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2905                 while (cache) {
2906                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2907                                 btrfs_put_block_group(cache);
2908                                 goto again;
2909                         }
2910
2911                         if (cache->dirty)
2912                                 break;
2913                         cache = next_block_group(root, cache);
2914                 }
2915                 if (!cache) {
2916                         if (last == 0)
2917                                 break;
2918                         last = 0;
2919                         continue;
2920                 }
2921
2922                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2923                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2924                 cache->dirty = 0;
2925                 last = cache->key.objectid + cache->key.offset;
2926
2927                 err = write_one_cache_group(trans, root, path, cache);
2928                 BUG_ON(err);
2929                 btrfs_put_block_group(cache);
2930         }
2931
2932         while (1) {
2933                 /*
2934                  * I don't think this is needed since we're just marking our
2935                  * preallocated extent as written, but just in case it can't
2936                  * hurt.
2937                  */
2938                 if (last == 0) {
2939                         err = btrfs_run_delayed_refs(trans, root,
2940                                                      (unsigned long)-1);
2941                         BUG_ON(err);
2942                 }
2943
2944                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2945                 while (cache) {
2946                         /*
2947                          * Really this shouldn't happen, but it could if we
2948                          * couldn't write the entire preallocated extent and
2949                          * splitting the extent resulted in a new block.
2950                          */
2951                         if (cache->dirty) {
2952                                 btrfs_put_block_group(cache);
2953                                 goto again;
2954                         }
2955                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2956                                 break;
2957                         cache = next_block_group(root, cache);
2958                 }
2959                 if (!cache) {
2960                         if (last == 0)
2961                                 break;
2962                         last = 0;
2963                         continue;
2964                 }
2965
2966                 btrfs_write_out_cache(root, trans, cache, path);
2967
2968                 /*
2969                  * If we didn't have an error then the cache state is still
2970                  * NEED_WRITE, so we can set it to WRITTEN.
2971                  */
2972                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2973                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2974                 last = cache->key.objectid + cache->key.offset;
2975                 btrfs_put_block_group(cache);
2976         }
2977
2978         btrfs_free_path(path);
2979         return 0;
2980 }
2981
2982 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2983 {
2984         struct btrfs_block_group_cache *block_group;
2985         int readonly = 0;
2986
2987         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2988         if (!block_group || block_group->ro)
2989                 readonly = 1;
2990         if (block_group)
2991                 btrfs_put_block_group(block_group);
2992         return readonly;
2993 }
2994
2995 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2996                              u64 total_bytes, u64 bytes_used,
2997                              struct btrfs_space_info **space_info)
2998 {
2999         struct btrfs_space_info *found;
3000         int i;
3001         int factor;
3002
3003         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3004                      BTRFS_BLOCK_GROUP_RAID10))
3005                 factor = 2;
3006         else
3007                 factor = 1;
3008
3009         found = __find_space_info(info, flags);
3010         if (found) {
3011                 spin_lock(&found->lock);
3012                 found->total_bytes += total_bytes;
3013                 found->disk_total += total_bytes * factor;
3014                 found->bytes_used += bytes_used;
3015                 found->disk_used += bytes_used * factor;
3016                 found->full = 0;
3017                 spin_unlock(&found->lock);
3018                 *space_info = found;
3019                 return 0;
3020         }
3021         found = kzalloc(sizeof(*found), GFP_NOFS);
3022         if (!found)
3023                 return -ENOMEM;
3024
3025         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3026                 INIT_LIST_HEAD(&found->block_groups[i]);
3027         init_rwsem(&found->groups_sem);
3028         spin_lock_init(&found->lock);
3029         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3030                                 BTRFS_BLOCK_GROUP_SYSTEM |
3031                                 BTRFS_BLOCK_GROUP_METADATA);
3032         found->total_bytes = total_bytes;
3033         found->disk_total = total_bytes * factor;
3034         found->bytes_used = bytes_used;
3035         found->disk_used = bytes_used * factor;
3036         found->bytes_pinned = 0;
3037         found->bytes_reserved = 0;
3038         found->bytes_readonly = 0;
3039         found->bytes_may_use = 0;
3040         found->full = 0;
3041         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3042         *space_info = found;
3043         list_add_rcu(&found->list, &info->space_info);
3044         atomic_set(&found->caching_threads, 0);
3045         return 0;
3046 }
3047
3048 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3049 {
3050         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3051                                    BTRFS_BLOCK_GROUP_RAID1 |
3052                                    BTRFS_BLOCK_GROUP_RAID10 |
3053                                    BTRFS_BLOCK_GROUP_DUP);
3054         if (extra_flags) {
3055                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3056                         fs_info->avail_data_alloc_bits |= extra_flags;
3057                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3058                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3059                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3060                         fs_info->avail_system_alloc_bits |= extra_flags;
3061         }
3062 }
3063
3064 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3065 {
3066         /*
3067          * we add in the count of missing devices because we want
3068          * to make sure that any RAID levels on a degraded FS
3069          * continue to be honored.
3070          */
3071         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3072                 root->fs_info->fs_devices->missing_devices;
3073
3074         if (num_devices == 1)
3075                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3076         if (num_devices < 4)
3077                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3078
3079         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3080             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3081                       BTRFS_BLOCK_GROUP_RAID10))) {
3082                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3083         }
3084
3085         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3086             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3087                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3088         }
3089
3090         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3091             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3092              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3093              (flags & BTRFS_BLOCK_GROUP_DUP)))
3094                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3095         return flags;
3096 }
3097
3098 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3099 {
3100         if (flags & BTRFS_BLOCK_GROUP_DATA)
3101                 flags |= root->fs_info->avail_data_alloc_bits &
3102                          root->fs_info->data_alloc_profile;
3103         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3104                 flags |= root->fs_info->avail_system_alloc_bits &
3105                          root->fs_info->system_alloc_profile;
3106         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3107                 flags |= root->fs_info->avail_metadata_alloc_bits &
3108                          root->fs_info->metadata_alloc_profile;
3109         return btrfs_reduce_alloc_profile(root, flags);
3110 }
3111
3112 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3113 {
3114         u64 flags;
3115
3116         if (data)
3117                 flags = BTRFS_BLOCK_GROUP_DATA;
3118         else if (root == root->fs_info->chunk_root)
3119                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3120         else
3121                 flags = BTRFS_BLOCK_GROUP_METADATA;
3122
3123         return get_alloc_profile(root, flags);
3124 }
3125
3126 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3127 {
3128         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3129                                                        BTRFS_BLOCK_GROUP_DATA);
3130 }
3131
3132 /*
3133  * This will check the space that the inode allocates from to make sure we have
3134  * enough space for bytes.
3135  */
3136 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3137 {
3138         struct btrfs_space_info *data_sinfo;
3139         struct btrfs_root *root = BTRFS_I(inode)->root;
3140         u64 used;
3141         int ret = 0, committed = 0, alloc_chunk = 1;
3142
3143         /* make sure bytes are sectorsize aligned */
3144         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3145
3146         if (root == root->fs_info->tree_root) {
3147                 alloc_chunk = 0;
3148                 committed = 1;
3149         }
3150
3151         data_sinfo = BTRFS_I(inode)->space_info;
3152         if (!data_sinfo)
3153                 goto alloc;
3154
3155 again:
3156         /* make sure we have enough space to handle the data first */
3157         spin_lock(&data_sinfo->lock);
3158         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3159                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3160                 data_sinfo->bytes_may_use;
3161
3162         if (used + bytes > data_sinfo->total_bytes) {
3163                 struct btrfs_trans_handle *trans;
3164
3165                 /*
3166                  * if we don't have enough free bytes in this space then we need
3167                  * to alloc a new chunk.
3168                  */
3169                 if (!data_sinfo->full && alloc_chunk) {
3170                         u64 alloc_target;
3171
3172                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3173                         spin_unlock(&data_sinfo->lock);
3174 alloc:
3175                         alloc_target = btrfs_get_alloc_profile(root, 1);
3176                         trans = btrfs_join_transaction(root, 1);
3177                         if (IS_ERR(trans))
3178                                 return PTR_ERR(trans);
3179
3180                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3181                                              bytes + 2 * 1024 * 1024,
3182                                              alloc_target,
3183                                              CHUNK_ALLOC_NO_FORCE);
3184                         btrfs_end_transaction(trans, root);
3185                         if (ret < 0) {
3186                                 if (ret != -ENOSPC)
3187                                         return ret;
3188                                 else
3189                                         goto commit_trans;
3190                         }
3191
3192                         if (!data_sinfo) {
3193                                 btrfs_set_inode_space_info(root, inode);
3194                                 data_sinfo = BTRFS_I(inode)->space_info;
3195                         }
3196                         goto again;
3197                 }
3198                 spin_unlock(&data_sinfo->lock);
3199
3200                 /* commit the current transaction and try again */
3201 commit_trans:
3202                 if (!committed && !root->fs_info->open_ioctl_trans) {
3203                         committed = 1;
3204                         trans = btrfs_join_transaction(root, 1);
3205                         if (IS_ERR(trans))
3206                                 return PTR_ERR(trans);
3207                         ret = btrfs_commit_transaction(trans, root);
3208                         if (ret)
3209                                 return ret;
3210                         goto again;
3211                 }
3212
3213 #if 0 /* I hope we never need this code again, just in case */
3214                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3215                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3216                        "%llu bytes_readonly, %llu may use %llu total\n",
3217                        (unsigned long long)bytes,
3218                        (unsigned long long)data_sinfo->bytes_used,
3219                        (unsigned long long)data_sinfo->bytes_reserved,
3220                        (unsigned long long)data_sinfo->bytes_pinned,
3221                        (unsigned long long)data_sinfo->bytes_readonly,
3222                        (unsigned long long)data_sinfo->bytes_may_use,
3223                        (unsigned long long)data_sinfo->total_bytes);
3224 #endif
3225                 return -ENOSPC;
3226         }
3227         data_sinfo->bytes_may_use += bytes;
3228         BTRFS_I(inode)->reserved_bytes += bytes;
3229         spin_unlock(&data_sinfo->lock);
3230
3231         return 0;
3232 }
3233
3234 /*
3235  * called when we are clearing an delalloc extent from the
3236  * inode's io_tree or there was an error for whatever reason
3237  * after calling btrfs_check_data_free_space
3238  */
3239 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3240 {
3241         struct btrfs_root *root = BTRFS_I(inode)->root;
3242         struct btrfs_space_info *data_sinfo;
3243
3244         /* make sure bytes are sectorsize aligned */
3245         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3246
3247         data_sinfo = BTRFS_I(inode)->space_info;
3248         spin_lock(&data_sinfo->lock);
3249         data_sinfo->bytes_may_use -= bytes;
3250         BTRFS_I(inode)->reserved_bytes -= bytes;
3251         spin_unlock(&data_sinfo->lock);
3252 }
3253
3254 static void force_metadata_allocation(struct btrfs_fs_info *info)
3255 {
3256         struct list_head *head = &info->space_info;
3257         struct btrfs_space_info *found;
3258
3259         rcu_read_lock();
3260         list_for_each_entry_rcu(found, head, list) {
3261                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3262                         found->force_alloc = CHUNK_ALLOC_FORCE;
3263         }
3264         rcu_read_unlock();
3265 }
3266
3267 static int should_alloc_chunk(struct btrfs_root *root,
3268                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3269                               int force)
3270 {
3271         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3272         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3273         u64 thresh;
3274
3275         if (force == CHUNK_ALLOC_FORCE)
3276                 return 1;
3277
3278         /*
3279          * in limited mode, we want to have some free space up to
3280          * about 1% of the FS size.
3281          */
3282         if (force == CHUNK_ALLOC_LIMITED) {
3283                 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3284                 thresh = max_t(u64, 64 * 1024 * 1024,
3285                                div_factor_fine(thresh, 1));
3286
3287                 if (num_bytes - num_allocated < thresh)
3288                         return 1;
3289         }
3290
3291         /*
3292          * we have two similar checks here, one based on percentage
3293          * and once based on a hard number of 256MB.  The idea
3294          * is that if we have a good amount of free
3295          * room, don't allocate a chunk.  A good mount is
3296          * less than 80% utilized of the chunks we have allocated,
3297          * or more than 256MB free
3298          */
3299         if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3300                 return 0;
3301
3302         if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3303                 return 0;
3304
3305         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3306
3307         /* 256MB or 5% of the FS */
3308         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3309
3310         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3311                 return 0;
3312         return 1;
3313 }
3314
3315 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3316                           struct btrfs_root *extent_root, u64 alloc_bytes,
3317                           u64 flags, int force)
3318 {
3319         struct btrfs_space_info *space_info;
3320         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3321         int ret = 0;
3322
3323         mutex_lock(&fs_info->chunk_mutex);
3324
3325         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3326
3327         space_info = __find_space_info(extent_root->fs_info, flags);
3328         if (!space_info) {
3329                 ret = update_space_info(extent_root->fs_info, flags,
3330                                         0, 0, &space_info);
3331                 BUG_ON(ret);
3332         }
3333         BUG_ON(!space_info);
3334
3335         spin_lock(&space_info->lock);
3336         if (space_info->force_alloc)
3337                 force = space_info->force_alloc;
3338         if (space_info->full) {
3339                 spin_unlock(&space_info->lock);
3340                 goto out;
3341         }
3342
3343         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3344                 spin_unlock(&space_info->lock);
3345                 goto out;
3346         }
3347
3348         spin_unlock(&space_info->lock);
3349
3350         /*
3351          * If we have mixed data/metadata chunks we want to make sure we keep
3352          * allocating mixed chunks instead of individual chunks.
3353          */
3354         if (btrfs_mixed_space_info(space_info))
3355                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3356
3357         /*
3358          * if we're doing a data chunk, go ahead and make sure that
3359          * we keep a reasonable number of metadata chunks allocated in the
3360          * FS as well.
3361          */
3362         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3363                 fs_info->data_chunk_allocations++;
3364                 if (!(fs_info->data_chunk_allocations %
3365                       fs_info->metadata_ratio))
3366                         force_metadata_allocation(fs_info);
3367         }
3368
3369         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3370         spin_lock(&space_info->lock);
3371         if (ret)
3372                 space_info->full = 1;
3373         else
3374                 ret = 1;
3375         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3376         spin_unlock(&space_info->lock);
3377 out:
3378         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3379         return ret;
3380 }
3381
3382 /*
3383  * shrink metadata reservation for delalloc
3384  */
3385 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3386                            struct btrfs_root *root, u64 to_reclaim, int sync)
3387 {
3388         struct btrfs_block_rsv *block_rsv;
3389         struct btrfs_space_info *space_info;
3390         u64 reserved;
3391         u64 max_reclaim;
3392         u64 reclaimed = 0;
3393         long time_left;
3394         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3395         int loops = 0;
3396         unsigned long progress;
3397
3398         block_rsv = &root->fs_info->delalloc_block_rsv;
3399         space_info = block_rsv->space_info;
3400
3401         smp_mb();
3402         reserved = space_info->bytes_reserved;
3403         progress = space_info->reservation_progress;
3404
3405         if (reserved == 0)
3406                 return 0;
3407
3408         max_reclaim = min(reserved, to_reclaim);
3409
3410         while (loops < 1024) {
3411                 /* have the flusher threads jump in and do some IO */
3412                 smp_mb();
3413                 nr_pages = min_t(unsigned long, nr_pages,
3414                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3415                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3416
3417                 spin_lock(&space_info->lock);
3418                 if (reserved > space_info->bytes_reserved)
3419                         reclaimed += reserved - space_info->bytes_reserved;
3420                 reserved = space_info->bytes_reserved;
3421                 spin_unlock(&space_info->lock);
3422
3423                 loops++;
3424
3425                 if (reserved == 0 || reclaimed >= max_reclaim)
3426                         break;
3427
3428                 if (trans && trans->transaction->blocked)
3429                         return -EAGAIN;
3430
3431                 time_left = schedule_timeout_interruptible(1);
3432
3433                 /* We were interrupted, exit */
3434                 if (time_left)
3435                         break;
3436
3437                 /* we've kicked the IO a few times, if anything has been freed,
3438                  * exit.  There is no sense in looping here for a long time
3439                  * when we really need to commit the transaction, or there are
3440                  * just too many writers without enough free space
3441                  */
3442
3443                 if (loops > 3) {
3444                         smp_mb();
3445                         if (progress != space_info->reservation_progress)
3446                                 break;
3447                 }
3448
3449         }
3450         return reclaimed >= to_reclaim;
3451 }
3452
3453 /*
3454  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3455  * idea is if this is the first call (retries == 0) then we will add to our
3456  * reserved count if we can't make the allocation in order to hold our place
3457  * while we go and try and free up space.  That way for retries > 1 we don't try
3458  * and add space, we just check to see if the amount of unused space is >= the
3459  * total space, meaning that our reservation is valid.
3460  *
3461  * However if we don't intend to retry this reservation, pass -1 as retries so
3462  * that it short circuits this logic.
3463  */
3464 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3465                                   struct btrfs_root *root,
3466                                   struct btrfs_block_rsv *block_rsv,
3467                                   u64 orig_bytes, int flush)
3468 {
3469         struct btrfs_space_info *space_info = block_rsv->space_info;
3470         u64 unused;
3471         u64 num_bytes = orig_bytes;
3472         int retries = 0;
3473         int ret = 0;
3474         bool reserved = false;
3475         bool committed = false;
3476
3477 again:
3478         ret = -ENOSPC;
3479         if (reserved)
3480                 num_bytes = 0;
3481
3482         spin_lock(&space_info->lock);
3483         unused = space_info->bytes_used + space_info->bytes_reserved +
3484                  space_info->bytes_pinned + space_info->bytes_readonly +
3485                  space_info->bytes_may_use;
3486
3487         /*
3488          * The idea here is that we've not already over-reserved the block group
3489          * then we can go ahead and save our reservation first and then start
3490          * flushing if we need to.  Otherwise if we've already overcommitted
3491          * lets start flushing stuff first and then come back and try to make
3492          * our reservation.
3493          */
3494         if (unused <= space_info->total_bytes) {
3495                 unused = space_info->total_bytes - unused;
3496                 if (unused >= num_bytes) {
3497                         if (!reserved)
3498                                 space_info->bytes_reserved += orig_bytes;
3499                         ret = 0;
3500                 } else {
3501                         /*
3502                          * Ok set num_bytes to orig_bytes since we aren't
3503                          * overocmmitted, this way we only try and reclaim what
3504                          * we need.
3505                          */
3506                         num_bytes = orig_bytes;
3507                 }
3508         } else {
3509                 /*
3510                  * Ok we're over committed, set num_bytes to the overcommitted
3511                  * amount plus the amount of bytes that we need for this
3512                  * reservation.
3513                  */
3514                 num_bytes = unused - space_info->total_bytes +
3515                         (orig_bytes * (retries + 1));
3516         }
3517
3518         /*
3519          * Couldn't make our reservation, save our place so while we're trying
3520          * to reclaim space we can actually use it instead of somebody else
3521          * stealing it from us.
3522          */
3523         if (ret && !reserved) {
3524                 space_info->bytes_reserved += orig_bytes;
3525                 reserved = true;
3526         }
3527
3528         spin_unlock(&space_info->lock);
3529
3530         if (!ret)
3531                 return 0;
3532
3533         if (!flush)
3534                 goto out;
3535
3536         /*
3537          * We do synchronous shrinking since we don't actually unreserve
3538          * metadata until after the IO is completed.
3539          */
3540         ret = shrink_delalloc(trans, root, num_bytes, 1);
3541         if (ret > 0)
3542                 return 0;
3543         else if (ret < 0)
3544                 goto out;
3545
3546         /*
3547          * So if we were overcommitted it's possible that somebody else flushed
3548          * out enough space and we simply didn't have enough space to reclaim,
3549          * so go back around and try again.
3550          */
3551         if (retries < 2) {
3552                 retries++;
3553                 goto again;
3554         }
3555
3556         spin_lock(&space_info->lock);
3557         /*
3558          * Not enough space to be reclaimed, don't bother committing the
3559          * transaction.
3560          */
3561         if (space_info->bytes_pinned < orig_bytes)
3562                 ret = -ENOSPC;
3563         spin_unlock(&space_info->lock);
3564         if (ret)
3565                 goto out;
3566
3567         ret = -EAGAIN;
3568         if (trans || committed)
3569                 goto out;
3570
3571         ret = -ENOSPC;
3572         trans = btrfs_join_transaction(root, 1);
3573         if (IS_ERR(trans))
3574                 goto out;
3575         ret = btrfs_commit_transaction(trans, root);
3576         if (!ret) {
3577                 trans = NULL;
3578                 committed = true;
3579                 goto again;
3580         }
3581
3582 out:
3583         if (reserved) {
3584                 spin_lock(&space_info->lock);
3585                 space_info->bytes_reserved -= orig_bytes;
3586                 spin_unlock(&space_info->lock);
3587         }
3588
3589         return ret;
3590 }
3591
3592 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3593                                              struct btrfs_root *root)
3594 {
3595         struct btrfs_block_rsv *block_rsv;
3596         if (root->ref_cows)
3597                 block_rsv = trans->block_rsv;
3598         else
3599                 block_rsv = root->block_rsv;
3600
3601         if (!block_rsv)
3602                 block_rsv = &root->fs_info->empty_block_rsv;
3603
3604         return block_rsv;
3605 }
3606
3607 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3608                                u64 num_bytes)
3609 {
3610         int ret = -ENOSPC;
3611         spin_lock(&block_rsv->lock);
3612         if (block_rsv->reserved >= num_bytes) {
3613                 block_rsv->reserved -= num_bytes;
3614                 if (block_rsv->reserved < block_rsv->size)
3615                         block_rsv->full = 0;
3616                 ret = 0;
3617         }
3618         spin_unlock(&block_rsv->lock);
3619         return ret;
3620 }
3621
3622 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3623                                 u64 num_bytes, int update_size)
3624 {
3625         spin_lock(&block_rsv->lock);
3626         block_rsv->reserved += num_bytes;
3627         if (update_size)
3628                 block_rsv->size += num_bytes;
3629         else if (block_rsv->reserved >= block_rsv->size)
3630                 block_rsv->full = 1;
3631         spin_unlock(&block_rsv->lock);
3632 }
3633
3634 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3635                              struct btrfs_block_rsv *dest, u64 num_bytes)
3636 {
3637         struct btrfs_space_info *space_info = block_rsv->space_info;
3638
3639         spin_lock(&block_rsv->lock);
3640         if (num_bytes == (u64)-1)
3641                 num_bytes = block_rsv->size;
3642         block_rsv->size -= num_bytes;
3643         if (block_rsv->reserved >= block_rsv->size) {
3644                 num_bytes = block_rsv->reserved - block_rsv->size;
3645                 block_rsv->reserved = block_rsv->size;
3646                 block_rsv->full = 1;
3647         } else {
3648                 num_bytes = 0;
3649         }
3650         spin_unlock(&block_rsv->lock);
3651
3652         if (num_bytes > 0) {
3653                 if (dest) {
3654                         spin_lock(&dest->lock);
3655                         if (!dest->full) {
3656                                 u64 bytes_to_add;
3657
3658                                 bytes_to_add = dest->size - dest->reserved;
3659                                 bytes_to_add = min(num_bytes, bytes_to_add);
3660                                 dest->reserved += bytes_to_add;
3661                                 if (dest->reserved >= dest->size)
3662                                         dest->full = 1;
3663                                 num_bytes -= bytes_to_add;
3664                         }
3665                         spin_unlock(&dest->lock);
3666                 }
3667                 if (num_bytes) {
3668                         spin_lock(&space_info->lock);
3669                         space_info->bytes_reserved -= num_bytes;
3670                         space_info->reservation_progress++;
3671                         spin_unlock(&space_info->lock);
3672                 }
3673         }
3674 }
3675
3676 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3677                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3678 {
3679         int ret;
3680
3681         ret = block_rsv_use_bytes(src, num_bytes);
3682         if (ret)
3683                 return ret;
3684
3685         block_rsv_add_bytes(dst, num_bytes, 1);
3686         return 0;
3687 }
3688
3689 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3690 {
3691         memset(rsv, 0, sizeof(*rsv));
3692         spin_lock_init(&rsv->lock);
3693         atomic_set(&rsv->usage, 1);
3694         rsv->priority = 6;
3695         INIT_LIST_HEAD(&rsv->list);
3696 }
3697
3698 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3699 {
3700         struct btrfs_block_rsv *block_rsv;
3701         struct btrfs_fs_info *fs_info = root->fs_info;
3702
3703         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3704         if (!block_rsv)
3705                 return NULL;
3706
3707         btrfs_init_block_rsv(block_rsv);
3708         block_rsv->space_info = __find_space_info(fs_info,
3709                                                   BTRFS_BLOCK_GROUP_METADATA);
3710         return block_rsv;
3711 }
3712
3713 void btrfs_free_block_rsv(struct btrfs_root *root,
3714                           struct btrfs_block_rsv *rsv)
3715 {
3716         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3717                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3718                 if (!rsv->durable)
3719                         kfree(rsv);
3720         }
3721 }
3722
3723 /*
3724  * make the block_rsv struct be able to capture freed space.
3725  * the captured space will re-add to the the block_rsv struct
3726  * after transaction commit
3727  */
3728 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3729                                  struct btrfs_block_rsv *block_rsv)
3730 {
3731         block_rsv->durable = 1;
3732         mutex_lock(&fs_info->durable_block_rsv_mutex);
3733         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3734         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3735 }
3736
3737 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3738                         struct btrfs_root *root,
3739                         struct btrfs_block_rsv *block_rsv,
3740                         u64 num_bytes)
3741 {
3742         int ret;
3743
3744         if (num_bytes == 0)
3745                 return 0;
3746
3747         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3748         if (!ret) {
3749                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3750                 return 0;
3751         }
3752
3753         return ret;
3754 }
3755
3756 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3757                           struct btrfs_root *root,
3758                           struct btrfs_block_rsv *block_rsv,
3759                           u64 min_reserved, int min_factor)
3760 {
3761         u64 num_bytes = 0;
3762         int commit_trans = 0;
3763         int ret = -ENOSPC;
3764
3765         if (!block_rsv)
3766                 return 0;
3767
3768         spin_lock(&block_rsv->lock);
3769         if (min_factor > 0)
3770                 num_bytes = div_factor(block_rsv->size, min_factor);
3771         if (min_reserved > num_bytes)
3772                 num_bytes = min_reserved;
3773
3774         if (block_rsv->reserved >= num_bytes) {
3775                 ret = 0;
3776         } else {
3777                 num_bytes -= block_rsv->reserved;
3778                 if (block_rsv->durable &&
3779                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3780                         commit_trans = 1;
3781         }
3782         spin_unlock(&block_rsv->lock);
3783         if (!ret)
3784                 return 0;
3785
3786         if (block_rsv->refill_used) {
3787                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3788                                              num_bytes, 0);
3789                 if (!ret) {
3790                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3791                         return 0;
3792                 }
3793         }
3794
3795         if (commit_trans) {
3796                 if (trans)
3797                         return -EAGAIN;
3798
3799                 trans = btrfs_join_transaction(root, 1);
3800                 BUG_ON(IS_ERR(trans));
3801                 ret = btrfs_commit_transaction(trans, root);
3802                 return 0;
3803         }
3804
3805         return -ENOSPC;
3806 }
3807
3808 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3809                             struct btrfs_block_rsv *dst_rsv,
3810                             u64 num_bytes)
3811 {
3812         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3813 }
3814
3815 void btrfs_block_rsv_release(struct btrfs_root *root,
3816                              struct btrfs_block_rsv *block_rsv,
3817                              u64 num_bytes)
3818 {
3819         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3820         if (global_rsv->full || global_rsv == block_rsv ||
3821             block_rsv->space_info != global_rsv->space_info)
3822                 global_rsv = NULL;
3823         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3824 }
3825
3826 /*
3827  * helper to calculate size of global block reservation.
3828  * the desired value is sum of space used by extent tree,
3829  * checksum tree and root tree
3830  */
3831 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3832 {
3833         struct btrfs_space_info *sinfo;
3834         u64 num_bytes;
3835         u64 meta_used;
3836         u64 data_used;
3837         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3838 #if 0
3839         /*
3840          * per tree used space accounting can be inaccuracy, so we
3841          * can't rely on it.
3842          */
3843         spin_lock(&fs_info->extent_root->accounting_lock);
3844         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3845         spin_unlock(&fs_info->extent_root->accounting_lock);
3846
3847         spin_lock(&fs_info->csum_root->accounting_lock);
3848         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3849         spin_unlock(&fs_info->csum_root->accounting_lock);
3850
3851         spin_lock(&fs_info->tree_root->accounting_lock);
3852         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3853         spin_unlock(&fs_info->tree_root->accounting_lock);
3854 #endif
3855         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3856         spin_lock(&sinfo->lock);
3857         data_used = sinfo->bytes_used;
3858         spin_unlock(&sinfo->lock);
3859
3860         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3861         spin_lock(&sinfo->lock);
3862         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3863                 data_used = 0;
3864         meta_used = sinfo->bytes_used;
3865         spin_unlock(&sinfo->lock);
3866
3867         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3868                     csum_size * 2;
3869         num_bytes += div64_u64(data_used + meta_used, 50);
3870
3871         if (num_bytes * 3 > meta_used)
3872                 num_bytes = div64_u64(meta_used, 3);
3873
3874         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3875 }
3876
3877 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3878 {
3879         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3880         struct btrfs_space_info *sinfo = block_rsv->space_info;
3881         u64 num_bytes;
3882
3883         num_bytes = calc_global_metadata_size(fs_info);
3884
3885         spin_lock(&block_rsv->lock);
3886         spin_lock(&sinfo->lock);
3887
3888         block_rsv->size = num_bytes;
3889
3890         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3891                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3892                     sinfo->bytes_may_use;
3893
3894         if (sinfo->total_bytes > num_bytes) {
3895                 num_bytes = sinfo->total_bytes - num_bytes;
3896                 block_rsv->reserved += num_bytes;
3897                 sinfo->bytes_reserved += num_bytes;
3898         }
3899
3900         if (block_rsv->reserved >= block_rsv->size) {
3901                 num_bytes = block_rsv->reserved - block_rsv->size;
3902                 sinfo->bytes_reserved -= num_bytes;
3903                 sinfo->reservation_progress++;
3904                 block_rsv->reserved = block_rsv->size;
3905                 block_rsv->full = 1;
3906         }
3907 #if 0
3908         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3909                 block_rsv->size, block_rsv->reserved);
3910 #endif
3911         spin_unlock(&sinfo->lock);
3912         spin_unlock(&block_rsv->lock);
3913 }
3914
3915 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3916 {
3917         struct btrfs_space_info *space_info;
3918
3919         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3920         fs_info->chunk_block_rsv.space_info = space_info;
3921         fs_info->chunk_block_rsv.priority = 10;
3922
3923         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3924         fs_info->global_block_rsv.space_info = space_info;
3925         fs_info->global_block_rsv.priority = 10;
3926         fs_info->global_block_rsv.refill_used = 1;
3927         fs_info->delalloc_block_rsv.space_info = space_info;
3928         fs_info->trans_block_rsv.space_info = space_info;
3929         fs_info->empty_block_rsv.space_info = space_info;
3930         fs_info->empty_block_rsv.priority = 10;
3931
3932         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3933         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3934         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3935         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3936         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3937
3938         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3939
3940         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3941
3942         update_global_block_rsv(fs_info);
3943 }
3944
3945 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3946 {
3947         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3948         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3949         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3950         WARN_ON(fs_info->trans_block_rsv.size > 0);
3951         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3952         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3953         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3954 }
3955
3956 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3957 {
3958         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3959                 3 * num_items;
3960 }
3961
3962 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3963                                  struct btrfs_root *root,
3964                                  int num_items)
3965 {
3966         u64 num_bytes;
3967         int ret;
3968
3969         if (num_items == 0 || root->fs_info->chunk_root == root)
3970                 return 0;
3971
3972         num_bytes = calc_trans_metadata_size(root, num_items);
3973         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3974                                   num_bytes);
3975         if (!ret) {
3976                 trans->bytes_reserved += num_bytes;
3977                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3978         }
3979         return ret;
3980 }
3981
3982 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3983                                   struct btrfs_root *root)
3984 {
3985         if (!trans->bytes_reserved)
3986                 return;
3987
3988         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3989         btrfs_block_rsv_release(root, trans->block_rsv,
3990                                 trans->bytes_reserved);
3991         trans->bytes_reserved = 0;
3992 }
3993
3994 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3995                                   struct inode *inode)
3996 {
3997         struct btrfs_root *root = BTRFS_I(inode)->root;
3998         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3999         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4000
4001         /*
4002          * one for deleting orphan item, one for updating inode and
4003          * two for calling btrfs_truncate_inode_items.
4004          *
4005          * btrfs_truncate_inode_items is a delete operation, it frees
4006          * more space than it uses in most cases. So two units of
4007          * metadata space should be enough for calling it many times.
4008          * If all of the metadata space is used, we can commit
4009          * transaction and use space it freed.
4010          */
4011         u64 num_bytes = calc_trans_metadata_size(root, 4);
4012         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4013 }
4014
4015 void btrfs_orphan_release_metadata(struct inode *inode)
4016 {
4017         struct btrfs_root *root = BTRFS_I(inode)->root;
4018         u64 num_bytes = calc_trans_metadata_size(root, 4);
4019         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4020 }
4021
4022 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4023                                 struct btrfs_pending_snapshot *pending)
4024 {
4025         struct btrfs_root *root = pending->root;
4026         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4027         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4028         /*
4029          * two for root back/forward refs, two for directory entries
4030          * and one for root of the snapshot.
4031          */
4032         u64 num_bytes = calc_trans_metadata_size(root, 5);
4033         dst_rsv->space_info = src_rsv->space_info;
4034         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4035 }
4036
4037 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
4038 {
4039         return num_bytes >>= 3;
4040 }
4041
4042 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4043 {
4044         struct btrfs_root *root = BTRFS_I(inode)->root;
4045         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4046         u64 to_reserve;
4047         int nr_extents;
4048         int reserved_extents;
4049         int ret;
4050
4051         if (btrfs_transaction_in_commit(root->fs_info))
4052                 schedule_timeout(1);
4053
4054         num_bytes = ALIGN(num_bytes, root->sectorsize);
4055
4056         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4057         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4058
4059         if (nr_extents > reserved_extents) {
4060                 nr_extents -= reserved_extents;
4061                 to_reserve = calc_trans_metadata_size(root, nr_extents);
4062         } else {
4063                 nr_extents = 0;
4064                 to_reserve = 0;
4065         }
4066
4067         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4068         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4069         if (ret)
4070                 return ret;
4071
4072         atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
4073         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4074
4075         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4076
4077         if (block_rsv->size > 512 * 1024 * 1024)
4078                 shrink_delalloc(NULL, root, to_reserve, 0);
4079
4080         return 0;
4081 }
4082
4083 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4084 {
4085         struct btrfs_root *root = BTRFS_I(inode)->root;
4086         u64 to_free;
4087         int nr_extents;
4088         int reserved_extents;
4089
4090         num_bytes = ALIGN(num_bytes, root->sectorsize);
4091         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4092         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4093
4094         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4095         do {
4096                 int old, new;
4097
4098                 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4099                 if (nr_extents >= reserved_extents) {
4100                         nr_extents = 0;
4101                         break;
4102                 }
4103                 old = reserved_extents;
4104                 nr_extents = reserved_extents - nr_extents;
4105                 new = reserved_extents - nr_extents;
4106                 old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
4107                                      reserved_extents, new);
4108                 if (likely(old == reserved_extents))
4109                         break;
4110                 reserved_extents = old;
4111         } while (1);
4112
4113         to_free = calc_csum_metadata_size(inode, num_bytes);
4114         if (nr_extents > 0)
4115                 to_free += calc_trans_metadata_size(root, nr_extents);
4116
4117         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4118                                 to_free);
4119 }
4120
4121 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4122 {
4123         int ret;
4124
4125         ret = btrfs_check_data_free_space(inode, num_bytes);
4126         if (ret)
4127                 return ret;
4128
4129         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4130         if (ret) {
4131                 btrfs_free_reserved_data_space(inode, num_bytes);
4132                 return ret;
4133         }
4134
4135         return 0;
4136 }
4137
4138 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4139 {
4140         btrfs_delalloc_release_metadata(inode, num_bytes);
4141         btrfs_free_reserved_data_space(inode, num_bytes);
4142 }
4143
4144 static int update_block_group(struct btrfs_trans_handle *trans,
4145                               struct btrfs_root *root,
4146                               u64 bytenr, u64 num_bytes, int alloc)
4147 {
4148         struct btrfs_block_group_cache *cache = NULL;
4149         struct btrfs_fs_info *info = root->fs_info;
4150         u64 total = num_bytes;
4151         u64 old_val;
4152         u64 byte_in_group;
4153         int factor;
4154
4155         /* block accounting for super block */
4156         spin_lock(&info->delalloc_lock);
4157         old_val = btrfs_super_bytes_used(&info->super_copy);
4158         if (alloc)
4159                 old_val += num_bytes;
4160         else
4161                 old_val -= num_bytes;
4162         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4163         spin_unlock(&info->delalloc_lock);
4164
4165         while (total) {
4166                 cache = btrfs_lookup_block_group(info, bytenr);
4167                 if (!cache)
4168                         return -1;
4169                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4170                                     BTRFS_BLOCK_GROUP_RAID1 |
4171                                     BTRFS_BLOCK_GROUP_RAID10))
4172                         factor = 2;
4173                 else
4174                         factor = 1;
4175                 /*
4176                  * If this block group has free space cache written out, we
4177                  * need to make sure to load it if we are removing space.  This
4178                  * is because we need the unpinning stage to actually add the
4179                  * space back to the block group, otherwise we will leak space.
4180                  */
4181                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4182                         cache_block_group(cache, trans, NULL, 1);
4183
4184                 byte_in_group = bytenr - cache->key.objectid;
4185                 WARN_ON(byte_in_group > cache->key.offset);
4186
4187                 spin_lock(&cache->space_info->lock);
4188                 spin_lock(&cache->lock);
4189
4190                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4191                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4192                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4193
4194                 cache->dirty = 1;
4195                 old_val = btrfs_block_group_used(&cache->item);
4196                 num_bytes = min(total, cache->key.offset - byte_in_group);
4197                 if (alloc) {
4198                         old_val += num_bytes;
4199                         btrfs_set_block_group_used(&cache->item, old_val);
4200                         cache->reserved -= num_bytes;
4201                         cache->space_info->bytes_reserved -= num_bytes;
4202                         cache->space_info->reservation_progress++;
4203                         cache->space_info->bytes_used += num_bytes;
4204                         cache->space_info->disk_used += num_bytes * factor;
4205                         spin_unlock(&cache->lock);
4206                         spin_unlock(&cache->space_info->lock);
4207                 } else {
4208                         old_val -= num_bytes;
4209                         btrfs_set_block_group_used(&cache->item, old_val);
4210                         cache->pinned += num_bytes;
4211                         cache->space_info->bytes_pinned += num_bytes;
4212                         cache->space_info->bytes_used -= num_bytes;
4213                         cache->space_info->disk_used -= num_bytes * factor;
4214                         spin_unlock(&cache->lock);
4215                         spin_unlock(&cache->space_info->lock);
4216
4217                         set_extent_dirty(info->pinned_extents,
4218                                          bytenr, bytenr + num_bytes - 1,
4219                                          GFP_NOFS | __GFP_NOFAIL);
4220                 }
4221                 btrfs_put_block_group(cache);
4222                 total -= num_bytes;
4223                 bytenr += num_bytes;
4224         }
4225         return 0;
4226 }
4227
4228 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4229 {
4230         struct btrfs_block_group_cache *cache;
4231         u64 bytenr;
4232
4233         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4234         if (!cache)
4235                 return 0;
4236
4237         bytenr = cache->key.objectid;
4238         btrfs_put_block_group(cache);
4239
4240         return bytenr;
4241 }
4242
4243 static int pin_down_extent(struct btrfs_root *root,
4244                            struct btrfs_block_group_cache *cache,
4245                            u64 bytenr, u64 num_bytes, int reserved)
4246 {
4247         spin_lock(&cache->space_info->lock);
4248         spin_lock(&cache->lock);
4249         cache->pinned += num_bytes;
4250         cache->space_info->bytes_pinned += num_bytes;
4251         if (reserved) {
4252                 cache->reserved -= num_bytes;
4253                 cache->space_info->bytes_reserved -= num_bytes;
4254                 cache->space_info->reservation_progress++;
4255         }
4256         spin_unlock(&cache->lock);
4257         spin_unlock(&cache->space_info->lock);
4258
4259         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4260                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4261         return 0;
4262 }
4263
4264 /*
4265  * this function must be called within transaction
4266  */
4267 int btrfs_pin_extent(struct btrfs_root *root,
4268                      u64 bytenr, u64 num_bytes, int reserved)
4269 {
4270         struct btrfs_block_group_cache *cache;
4271
4272         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4273         BUG_ON(!cache);
4274
4275         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4276
4277         btrfs_put_block_group(cache);
4278         return 0;
4279 }
4280
4281 /*
4282  * update size of reserved extents. this function may return -EAGAIN
4283  * if 'reserve' is true or 'sinfo' is false.
4284  */
4285 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4286                                 u64 num_bytes, int reserve, int sinfo)
4287 {
4288         int ret = 0;
4289         if (sinfo) {
4290                 struct btrfs_space_info *space_info = cache->space_info;
4291                 spin_lock(&space_info->lock);
4292                 spin_lock(&cache->lock);
4293                 if (reserve) {
4294                         if (cache->ro) {
4295                                 ret = -EAGAIN;
4296                         } else {
4297                                 cache->reserved += num_bytes;
4298                                 space_info->bytes_reserved += num_bytes;
4299                         }
4300                 } else {
4301                         if (cache->ro)
4302                                 space_info->bytes_readonly += num_bytes;
4303                         cache->reserved -= num_bytes;
4304                         space_info->bytes_reserved -= num_bytes;
4305                         space_info->reservation_progress++;
4306                 }
4307                 spin_unlock(&cache->lock);
4308                 spin_unlock(&space_info->lock);
4309         } else {
4310                 spin_lock(&cache->lock);
4311                 if (cache->ro) {
4312                         ret = -EAGAIN;
4313                 } else {
4314                         if (reserve)
4315                                 cache->reserved += num_bytes;
4316                         else
4317                                 cache->reserved -= num_bytes;
4318                 }
4319                 spin_unlock(&cache->lock);
4320         }
4321         return ret;
4322 }
4323
4324 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4325                                 struct btrfs_root *root)
4326 {
4327         struct btrfs_fs_info *fs_info = root->fs_info;
4328         struct btrfs_caching_control *next;
4329         struct btrfs_caching_control *caching_ctl;
4330         struct btrfs_block_group_cache *cache;
4331
4332         down_write(&fs_info->extent_commit_sem);
4333
4334         list_for_each_entry_safe(caching_ctl, next,
4335                                  &fs_info->caching_block_groups, list) {
4336                 cache = caching_ctl->block_group;
4337                 if (block_group_cache_done(cache)) {
4338                         cache->last_byte_to_unpin = (u64)-1;
4339                         list_del_init(&caching_ctl->list);
4340                         put_caching_control(caching_ctl);
4341                 } else {
4342                         cache->last_byte_to_unpin = caching_ctl->progress;
4343                 }
4344         }
4345
4346         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4347                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4348         else
4349                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4350
4351         up_write(&fs_info->extent_commit_sem);
4352
4353         update_global_block_rsv(fs_info);
4354         return 0;
4355 }
4356
4357 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4358 {
4359         struct btrfs_fs_info *fs_info = root->fs_info;
4360         struct btrfs_block_group_cache *cache = NULL;
4361         u64 len;
4362
4363         while (start <= end) {
4364                 if (!cache ||
4365                     start >= cache->key.objectid + cache->key.offset) {
4366                         if (cache)
4367                                 btrfs_put_block_group(cache);
4368                         cache = btrfs_lookup_block_group(fs_info, start);
4369                         BUG_ON(!cache);
4370                 }
4371
4372                 len = cache->key.objectid + cache->key.offset - start;
4373                 len = min(len, end + 1 - start);
4374
4375                 if (start < cache->last_byte_to_unpin) {
4376                         len = min(len, cache->last_byte_to_unpin - start);
4377                         btrfs_add_free_space(cache, start, len);
4378                 }
4379
4380                 start += len;
4381
4382                 spin_lock(&cache->space_info->lock);
4383                 spin_lock(&cache->lock);
4384                 cache->pinned -= len;
4385                 cache->space_info->bytes_pinned -= len;
4386                 if (cache->ro) {
4387                         cache->space_info->bytes_readonly += len;
4388                 } else if (cache->reserved_pinned > 0) {
4389                         len = min(len, cache->reserved_pinned);
4390                         cache->reserved_pinned -= len;
4391                         cache->space_info->bytes_reserved += len;
4392                 }
4393                 spin_unlock(&cache->lock);
4394                 spin_unlock(&cache->space_info->lock);
4395         }
4396
4397         if (cache)
4398                 btrfs_put_block_group(cache);
4399         return 0;
4400 }
4401
4402 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4403                                struct btrfs_root *root)
4404 {
4405         struct btrfs_fs_info *fs_info = root->fs_info;
4406         struct extent_io_tree *unpin;
4407         struct btrfs_block_rsv *block_rsv;
4408         struct btrfs_block_rsv *next_rsv;
4409         u64 start;
4410         u64 end;
4411         int idx;
4412         int ret;
4413
4414         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4415                 unpin = &fs_info->freed_extents[1];
4416         else
4417                 unpin = &fs_info->freed_extents[0];
4418
4419         while (1) {
4420                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4421                                             EXTENT_DIRTY);
4422                 if (ret)
4423                         break;
4424
4425                 if (btrfs_test_opt(root, DISCARD))
4426                         ret = btrfs_discard_extent(root, start,
4427                                                    end + 1 - start, NULL);
4428
4429                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4430                 unpin_extent_range(root, start, end);
4431                 cond_resched();
4432         }
4433
4434         mutex_lock(&fs_info->durable_block_rsv_mutex);
4435         list_for_each_entry_safe(block_rsv, next_rsv,
4436                                  &fs_info->durable_block_rsv_list, list) {
4437
4438                 idx = trans->transid & 0x1;
4439                 if (block_rsv->freed[idx] > 0) {
4440                         block_rsv_add_bytes(block_rsv,
4441                                             block_rsv->freed[idx], 0);
4442                         block_rsv->freed[idx] = 0;
4443                 }
4444                 if (atomic_read(&block_rsv->usage) == 0) {
4445                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4446
4447                         if (block_rsv->freed[0] == 0 &&
4448                             block_rsv->freed[1] == 0) {
4449                                 list_del_init(&block_rsv->list);
4450                                 kfree(block_rsv);
4451                         }
4452                 } else {
4453                         btrfs_block_rsv_release(root, block_rsv, 0);
4454                 }
4455         }
4456         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4457
4458         return 0;
4459 }
4460
4461 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4462                                 struct btrfs_root *root,
4463                                 u64 bytenr, u64 num_bytes, u64 parent,
4464                                 u64 root_objectid, u64 owner_objectid,
4465                                 u64 owner_offset, int refs_to_drop,
4466                                 struct btrfs_delayed_extent_op *extent_op)
4467 {
4468         struct btrfs_key key;
4469         struct btrfs_path *path;
4470         struct btrfs_fs_info *info = root->fs_info;
4471         struct btrfs_root *extent_root = info->extent_root;
4472         struct extent_buffer *leaf;
4473         struct btrfs_extent_item *ei;
4474         struct btrfs_extent_inline_ref *iref;
4475         int ret;
4476         int is_data;
4477         int extent_slot = 0;
4478         int found_extent = 0;
4479         int num_to_del = 1;
4480         u32 item_size;
4481         u64 refs;
4482
4483         path = btrfs_alloc_path();
4484         if (!path)
4485                 return -ENOMEM;
4486
4487         path->reada = 1;
4488         path->leave_spinning = 1;
4489
4490         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4491         BUG_ON(!is_data && refs_to_drop != 1);
4492
4493         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4494                                     bytenr, num_bytes, parent,
4495                                     root_objectid, owner_objectid,
4496                                     owner_offset);
4497         if (ret == 0) {
4498                 extent_slot = path->slots[0];
4499                 while (extent_slot >= 0) {
4500                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4501                                               extent_slot);
4502                         if (key.objectid != bytenr)
4503                                 break;
4504                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4505                             key.offset == num_bytes) {
4506                                 found_extent = 1;
4507                                 break;
4508                         }
4509                         if (path->slots[0] - extent_slot > 5)
4510                                 break;
4511                         extent_slot--;
4512                 }
4513 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4514                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4515                 if (found_extent && item_size < sizeof(*ei))
4516                         found_extent = 0;
4517 #endif
4518                 if (!found_extent) {
4519                         BUG_ON(iref);
4520                         ret = remove_extent_backref(trans, extent_root, path,
4521                                                     NULL, refs_to_drop,
4522                                                     is_data);
4523                         BUG_ON(ret);
4524                         btrfs_release_path(extent_root, path);
4525                         path->leave_spinning = 1;
4526
4527                         key.objectid = bytenr;
4528                         key.type = BTRFS_EXTENT_ITEM_KEY;
4529                         key.offset = num_bytes;
4530
4531                         ret = btrfs_search_slot(trans, extent_root,
4532                                                 &key, path, -1, 1);
4533                         if (ret) {
4534                                 printk(KERN_ERR "umm, got %d back from search"
4535                                        ", was looking for %llu\n", ret,
4536                                        (unsigned long long)bytenr);
4537                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4538                         }
4539                         BUG_ON(ret);
4540                         extent_slot = path->slots[0];
4541                 }
4542         } else {
4543                 btrfs_print_leaf(extent_root, path->nodes[0]);
4544                 WARN_ON(1);
4545                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4546                        "parent %llu root %llu  owner %llu offset %llu\n",
4547                        (unsigned long long)bytenr,
4548                        (unsigned long long)parent,
4549                        (unsigned long long)root_objectid,
4550                        (unsigned long long)owner_objectid,
4551                        (unsigned long long)owner_offset);
4552         }
4553
4554         leaf = path->nodes[0];
4555         item_size = btrfs_item_size_nr(leaf, extent_slot);
4556 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4557         if (item_size < sizeof(*ei)) {
4558                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4559                 ret = convert_extent_item_v0(trans, extent_root, path,
4560                                              owner_objectid, 0);
4561                 BUG_ON(ret < 0);
4562
4563                 btrfs_release_path(extent_root, path);
4564                 path->leave_spinning = 1;
4565
4566                 key.objectid = bytenr;
4567                 key.type = BTRFS_EXTENT_ITEM_KEY;
4568                 key.offset = num_bytes;
4569
4570                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4571                                         -1, 1);
4572                 if (ret) {
4573                         printk(KERN_ERR "umm, got %d back from search"
4574                                ", was looking for %llu\n", ret,
4575                                (unsigned long long)bytenr);
4576                         btrfs_print_leaf(extent_root, path->nodes[0]);
4577                 }
4578                 BUG_ON(ret);
4579                 extent_slot = path->slots[0];
4580                 leaf = path->nodes[0];
4581                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4582         }
4583 #endif
4584         BUG_ON(item_size < sizeof(*ei));
4585         ei = btrfs_item_ptr(leaf, extent_slot,
4586                             struct btrfs_extent_item);
4587         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4588                 struct btrfs_tree_block_info *bi;
4589                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4590                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4591                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4592         }
4593
4594         refs = btrfs_extent_refs(leaf, ei);
4595         BUG_ON(refs < refs_to_drop);
4596         refs -= refs_to_drop;
4597
4598         if (refs > 0) {
4599                 if (extent_op)
4600                         __run_delayed_extent_op(extent_op, leaf, ei);
4601                 /*
4602                  * In the case of inline back ref, reference count will
4603                  * be updated by remove_extent_backref
4604                  */
4605                 if (iref) {
4606                         BUG_ON(!found_extent);
4607                 } else {
4608                         btrfs_set_extent_refs(leaf, ei, refs);
4609                         btrfs_mark_buffer_dirty(leaf);
4610                 }
4611                 if (found_extent) {
4612                         ret = remove_extent_backref(trans, extent_root, path,
4613                                                     iref, refs_to_drop,
4614                                                     is_data);
4615                         BUG_ON(ret);
4616                 }
4617         } else {
4618                 if (found_extent) {
4619                         BUG_ON(is_data && refs_to_drop !=
4620                                extent_data_ref_count(root, path, iref));
4621                         if (iref) {
4622                                 BUG_ON(path->slots[0] != extent_slot);
4623                         } else {
4624                                 BUG_ON(path->slots[0] != extent_slot + 1);
4625                                 path->slots[0] = extent_slot;
4626                                 num_to_del = 2;
4627                         }
4628                 }
4629
4630                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4631                                       num_to_del);
4632                 BUG_ON(ret);
4633                 btrfs_release_path(extent_root, path);
4634
4635                 if (is_data) {
4636                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4637                         BUG_ON(ret);
4638                 } else {
4639                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4640                              bytenr >> PAGE_CACHE_SHIFT,
4641                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4642                 }
4643
4644                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4645                 BUG_ON(ret);
4646         }
4647         btrfs_free_path(path);
4648         return ret;
4649 }
4650
4651 /*
4652  * when we free an block, it is possible (and likely) that we free the last
4653  * delayed ref for that extent as well.  This searches the delayed ref tree for
4654  * a given extent, and if there are no other delayed refs to be processed, it
4655  * removes it from the tree.
4656  */
4657 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4658                                       struct btrfs_root *root, u64 bytenr)
4659 {
4660         struct btrfs_delayed_ref_head *head;
4661         struct btrfs_delayed_ref_root *delayed_refs;
4662         struct btrfs_delayed_ref_node *ref;
4663         struct rb_node *node;
4664         int ret = 0;
4665
4666         delayed_refs = &trans->transaction->delayed_refs;
4667         spin_lock(&delayed_refs->lock);
4668         head = btrfs_find_delayed_ref_head(trans, bytenr);
4669         if (!head)
4670                 goto out;
4671
4672         node = rb_prev(&head->node.rb_node);
4673         if (!node)
4674                 goto out;
4675
4676         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4677
4678         /* there are still entries for this ref, we can't drop it */
4679         if (ref->bytenr == bytenr)
4680                 goto out;
4681
4682         if (head->extent_op) {
4683                 if (!head->must_insert_reserved)
4684                         goto out;
4685                 kfree(head->extent_op);
4686                 head->extent_op = NULL;
4687         }
4688
4689         /*
4690          * waiting for the lock here would deadlock.  If someone else has it
4691          * locked they are already in the process of dropping it anyway
4692          */
4693         if (!mutex_trylock(&head->mutex))
4694                 goto out;
4695
4696         /*
4697          * at this point we have a head with no other entries.  Go
4698          * ahead and process it.
4699          */
4700         head->node.in_tree = 0;
4701         rb_erase(&head->node.rb_node, &delayed_refs->root);
4702
4703         delayed_refs->num_entries--;
4704
4705         /*
4706          * we don't take a ref on the node because we're removing it from the
4707          * tree, so we just steal the ref the tree was holding.
4708          */
4709         delayed_refs->num_heads--;
4710         if (list_empty(&head->cluster))
4711                 delayed_refs->num_heads_ready--;
4712
4713         list_del_init(&head->cluster);
4714         spin_unlock(&delayed_refs->lock);
4715
4716         BUG_ON(head->extent_op);
4717         if (head->must_insert_reserved)
4718                 ret = 1;
4719
4720         mutex_unlock(&head->mutex);
4721         btrfs_put_delayed_ref(&head->node);
4722         return ret;
4723 out:
4724         spin_unlock(&delayed_refs->lock);
4725         return 0;
4726 }
4727
4728 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4729                            struct btrfs_root *root,
4730                            struct extent_buffer *buf,
4731                            u64 parent, int last_ref)
4732 {
4733         struct btrfs_block_rsv *block_rsv;
4734         struct btrfs_block_group_cache *cache = NULL;
4735         int ret;
4736
4737         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4738                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4739                                                 parent, root->root_key.objectid,
4740                                                 btrfs_header_level(buf),
4741                                                 BTRFS_DROP_DELAYED_REF, NULL);
4742                 BUG_ON(ret);
4743         }
4744
4745         if (!last_ref)
4746                 return;
4747
4748         block_rsv = get_block_rsv(trans, root);
4749         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4750         if (block_rsv->space_info != cache->space_info)
4751                 goto out;
4752
4753         if (btrfs_header_generation(buf) == trans->transid) {
4754                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4755                         ret = check_ref_cleanup(trans, root, buf->start);
4756                         if (!ret)
4757                                 goto pin;
4758                 }
4759
4760                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4761                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4762                         goto pin;
4763                 }
4764
4765                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4766
4767                 btrfs_add_free_space(cache, buf->start, buf->len);
4768                 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
4769                 if (ret == -EAGAIN) {
4770                         /* block group became read-only */
4771                         btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
4772                         goto out;
4773                 }
4774
4775                 ret = 1;
4776                 spin_lock(&block_rsv->lock);
4777                 if (block_rsv->reserved < block_rsv->size) {
4778                         block_rsv->reserved += buf->len;
4779                         ret = 0;
4780                 }
4781                 spin_unlock(&block_rsv->lock);
4782
4783                 if (ret) {
4784                         spin_lock(&cache->space_info->lock);
4785                         cache->space_info->bytes_reserved -= buf->len;
4786                         cache->space_info->reservation_progress++;
4787                         spin_unlock(&cache->space_info->lock);
4788                 }
4789                 goto out;
4790         }
4791 pin:
4792         if (block_rsv->durable && !cache->ro) {
4793                 ret = 0;
4794                 spin_lock(&cache->lock);
4795                 if (!cache->ro) {
4796                         cache->reserved_pinned += buf->len;
4797                         ret = 1;
4798                 }
4799                 spin_unlock(&cache->lock);
4800
4801                 if (ret) {
4802                         spin_lock(&block_rsv->lock);
4803                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4804                         spin_unlock(&block_rsv->lock);
4805                 }
4806         }
4807 out:
4808         /*
4809          * Deleting the buffer, clear the corrupt flag since it doesn't matter
4810          * anymore.
4811          */
4812         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4813         btrfs_put_block_group(cache);
4814 }
4815
4816 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4817                       struct btrfs_root *root,
4818                       u64 bytenr, u64 num_bytes, u64 parent,
4819                       u64 root_objectid, u64 owner, u64 offset)
4820 {
4821         int ret;
4822
4823         /*
4824          * tree log blocks never actually go into the extent allocation
4825          * tree, just update pinning info and exit early.
4826          */
4827         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4828                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4829                 /* unlocks the pinned mutex */
4830                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4831                 ret = 0;
4832         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4833                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4834                                         parent, root_objectid, (int)owner,
4835                                         BTRFS_DROP_DELAYED_REF, NULL);
4836                 BUG_ON(ret);
4837         } else {
4838                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4839                                         parent, root_objectid, owner,
4840                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4841                 BUG_ON(ret);
4842         }
4843         return ret;
4844 }
4845
4846 static u64 stripe_align(struct btrfs_root *root, u64 val)
4847 {
4848         u64 mask = ((u64)root->stripesize - 1);
4849         u64 ret = (val + mask) & ~mask;
4850         return ret;
4851 }
4852
4853 /*
4854  * when we wait for progress in the block group caching, its because
4855  * our allocation attempt failed at least once.  So, we must sleep
4856  * and let some progress happen before we try again.
4857  *
4858  * This function will sleep at least once waiting for new free space to
4859  * show up, and then it will check the block group free space numbers
4860  * for our min num_bytes.  Another option is to have it go ahead
4861  * and look in the rbtree for a free extent of a given size, but this
4862  * is a good start.
4863  */
4864 static noinline int
4865 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4866                                 u64 num_bytes)
4867 {
4868         struct btrfs_caching_control *caching_ctl;
4869         DEFINE_WAIT(wait);
4870
4871         caching_ctl = get_caching_control(cache);
4872         if (!caching_ctl)
4873                 return 0;
4874
4875         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4876                    (cache->free_space >= num_bytes));
4877
4878         put_caching_control(caching_ctl);
4879         return 0;
4880 }
4881
4882 static noinline int
4883 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4884 {
4885         struct btrfs_caching_control *caching_ctl;
4886         DEFINE_WAIT(wait);
4887
4888         caching_ctl = get_caching_control(cache);
4889         if (!caching_ctl)
4890                 return 0;
4891
4892         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4893
4894         put_caching_control(caching_ctl);
4895         return 0;
4896 }
4897
4898 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4899 {
4900         int index;
4901         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4902                 index = 0;
4903         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4904                 index = 1;
4905         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4906                 index = 2;
4907         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4908                 index = 3;
4909         else
4910                 index = 4;
4911         return index;
4912 }
4913
4914 enum btrfs_loop_type {
4915         LOOP_FIND_IDEAL = 0,
4916         LOOP_CACHING_NOWAIT = 1,
4917         LOOP_CACHING_WAIT = 2,
4918         LOOP_ALLOC_CHUNK = 3,
4919         LOOP_NO_EMPTY_SIZE = 4,
4920 };
4921
4922 /*
4923  * walks the btree of allocated extents and find a hole of a given size.
4924  * The key ins is changed to record the hole:
4925  * ins->objectid == block start
4926  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4927  * ins->offset == number of blocks
4928  * Any available blocks before search_start are skipped.
4929  */
4930 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4931                                      struct btrfs_root *orig_root,
4932                                      u64 num_bytes, u64 empty_size,
4933                                      u64 search_start, u64 search_end,
4934                                      u64 hint_byte, struct btrfs_key *ins,
4935                                      int data)
4936 {
4937         int ret = 0;
4938         struct btrfs_root *root = orig_root->fs_info->extent_root;
4939         struct btrfs_free_cluster *last_ptr = NULL;
4940         struct btrfs_block_group_cache *block_group = NULL;
4941         int empty_cluster = 2 * 1024 * 1024;
4942         int allowed_chunk_alloc = 0;
4943         int done_chunk_alloc = 0;
4944         struct btrfs_space_info *space_info;
4945         int last_ptr_loop = 0;
4946         int loop = 0;
4947         int index = 0;
4948         bool found_uncached_bg = false;
4949         bool failed_cluster_refill = false;
4950         bool failed_alloc = false;
4951         bool use_cluster = true;
4952         u64 ideal_cache_percent = 0;
4953         u64 ideal_cache_offset = 0;
4954
4955         WARN_ON(num_bytes < root->sectorsize);
4956         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4957         ins->objectid = 0;
4958         ins->offset = 0;
4959
4960         space_info = __find_space_info(root->fs_info, data);
4961         if (!space_info) {
4962                 printk(KERN_ERR "No space info for %d\n", data);
4963                 return -ENOSPC;
4964         }
4965
4966         /*
4967          * If the space info is for both data and metadata it means we have a
4968          * small filesystem and we can't use the clustering stuff.
4969          */
4970         if (btrfs_mixed_space_info(space_info))
4971                 use_cluster = false;
4972
4973         if (orig_root->ref_cows || empty_size)
4974                 allowed_chunk_alloc = 1;
4975
4976         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4977                 last_ptr = &root->fs_info->meta_alloc_cluster;
4978                 if (!btrfs_test_opt(root, SSD))
4979                         empty_cluster = 64 * 1024;
4980         }
4981
4982         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4983             btrfs_test_opt(root, SSD)) {
4984                 last_ptr = &root->fs_info->data_alloc_cluster;
4985         }
4986
4987         if (last_ptr) {
4988                 spin_lock(&last_ptr->lock);
4989                 if (last_ptr->block_group)
4990                         hint_byte = last_ptr->window_start;
4991                 spin_unlock(&last_ptr->lock);
4992         }
4993
4994         search_start = max(search_start, first_logical_byte(root, 0));
4995         search_start = max(search_start, hint_byte);
4996
4997         if (!last_ptr)
4998                 empty_cluster = 0;
4999
5000         if (search_start == hint_byte) {
5001 ideal_cache:
5002                 block_group = btrfs_lookup_block_group(root->fs_info,
5003                                                        search_start);
5004                 /*
5005                  * we don't want to use the block group if it doesn't match our
5006                  * allocation bits, or if its not cached.
5007                  *
5008                  * However if we are re-searching with an ideal block group
5009                  * picked out then we don't care that the block group is cached.
5010                  */
5011                 if (block_group && block_group_bits(block_group, data) &&
5012                     (block_group->cached != BTRFS_CACHE_NO ||
5013                      search_start == ideal_cache_offset)) {
5014                         down_read(&space_info->groups_sem);
5015                         if (list_empty(&block_group->list) ||
5016                             block_group->ro) {
5017                                 /*
5018                                  * someone is removing this block group,
5019                                  * we can't jump into the have_block_group
5020                                  * target because our list pointers are not
5021                                  * valid
5022                                  */
5023                                 btrfs_put_block_group(block_group);
5024                                 up_read(&space_info->groups_sem);
5025                         } else {
5026                                 index = get_block_group_index(block_group);
5027                                 goto have_block_group;
5028                         }
5029                 } else if (block_group) {
5030                         btrfs_put_block_group(block_group);
5031                 }
5032         }
5033 search:
5034         down_read(&space_info->groups_sem);
5035         list_for_each_entry(block_group, &space_info->block_groups[index],
5036                             list) {
5037                 u64 offset;
5038                 int cached;
5039
5040                 btrfs_get_block_group(block_group);
5041                 search_start = block_group->key.objectid;
5042
5043                 /*
5044                  * this can happen if we end up cycling through all the
5045                  * raid types, but we want to make sure we only allocate
5046                  * for the proper type.
5047                  */
5048                 if (!block_group_bits(block_group, data)) {
5049                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5050                                 BTRFS_BLOCK_GROUP_RAID1 |
5051                                 BTRFS_BLOCK_GROUP_RAID10;
5052
5053                         /*
5054                          * if they asked for extra copies and this block group
5055                          * doesn't provide them, bail.  This does allow us to
5056                          * fill raid0 from raid1.
5057                          */
5058                         if ((data & extra) && !(block_group->flags & extra))
5059                                 goto loop;
5060                 }
5061
5062 have_block_group:
5063                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5064                         u64 free_percent;
5065
5066                         ret = cache_block_group(block_group, trans,
5067                                                 orig_root, 1);
5068                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5069                                 goto have_block_group;
5070
5071                         free_percent = btrfs_block_group_used(&block_group->item);
5072                         free_percent *= 100;
5073                         free_percent = div64_u64(free_percent,
5074                                                  block_group->key.offset);
5075                         free_percent = 100 - free_percent;
5076                         if (free_percent > ideal_cache_percent &&
5077                             likely(!block_group->ro)) {
5078                                 ideal_cache_offset = block_group->key.objectid;
5079                                 ideal_cache_percent = free_percent;
5080                         }
5081
5082                         /*
5083                          * We only want to start kthread caching if we are at
5084                          * the point where we will wait for caching to make
5085                          * progress, or if our ideal search is over and we've
5086                          * found somebody to start caching.
5087                          */
5088                         if (loop > LOOP_CACHING_NOWAIT ||
5089                             (loop > LOOP_FIND_IDEAL &&
5090                              atomic_read(&space_info->caching_threads) < 2)) {
5091                                 ret = cache_block_group(block_group, trans,
5092                                                         orig_root, 0);
5093                                 BUG_ON(ret);
5094                         }
5095                         found_uncached_bg = true;
5096
5097                         /*
5098                          * If loop is set for cached only, try the next block
5099                          * group.
5100                          */
5101                         if (loop == LOOP_FIND_IDEAL)
5102                                 goto loop;
5103                 }
5104
5105                 cached = block_group_cache_done(block_group);
5106                 if (unlikely(!cached))
5107                         found_uncached_bg = true;
5108
5109                 if (unlikely(block_group->ro))
5110                         goto loop;
5111
5112                 /*
5113                  * Ok we want to try and use the cluster allocator, so lets look
5114                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5115                  * have tried the cluster allocator plenty of times at this
5116                  * point and not have found anything, so we are likely way too
5117                  * fragmented for the clustering stuff to find anything, so lets
5118                  * just skip it and let the allocator find whatever block it can
5119                  * find
5120                  */
5121                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5122                         /*
5123                          * the refill lock keeps out other
5124                          * people trying to start a new cluster
5125                          */
5126                         spin_lock(&last_ptr->refill_lock);
5127                         if (last_ptr->block_group &&
5128                             (last_ptr->block_group->ro ||
5129                             !block_group_bits(last_ptr->block_group, data))) {
5130                                 offset = 0;
5131                                 goto refill_cluster;
5132                         }
5133
5134                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5135                                                  num_bytes, search_start);
5136                         if (offset) {
5137                                 /* we have a block, we're done */
5138                                 spin_unlock(&last_ptr->refill_lock);
5139                                 goto checks;
5140                         }
5141
5142                         spin_lock(&last_ptr->lock);
5143                         /*
5144                          * whoops, this cluster doesn't actually point to
5145                          * this block group.  Get a ref on the block
5146                          * group is does point to and try again
5147                          */
5148                         if (!last_ptr_loop && last_ptr->block_group &&
5149                             last_ptr->block_group != block_group) {
5150
5151                                 btrfs_put_block_group(block_group);
5152                                 block_group = last_ptr->block_group;
5153                                 btrfs_get_block_group(block_group);
5154                                 spin_unlock(&last_ptr->lock);
5155                                 spin_unlock(&last_ptr->refill_lock);
5156
5157                                 last_ptr_loop = 1;
5158                                 search_start = block_group->key.objectid;
5159                                 /*
5160                                  * we know this block group is properly
5161                                  * in the list because
5162                                  * btrfs_remove_block_group, drops the
5163                                  * cluster before it removes the block
5164                                  * group from the list
5165                                  */
5166                                 goto have_block_group;
5167                         }
5168                         spin_unlock(&last_ptr->lock);
5169 refill_cluster:
5170                         /*
5171                          * this cluster didn't work out, free it and
5172                          * start over
5173                          */
5174                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5175
5176                         last_ptr_loop = 0;
5177
5178                         /* allocate a cluster in this block group */
5179                         ret = btrfs_find_space_cluster(trans, root,
5180                                                block_group, last_ptr,
5181                                                offset, num_bytes,
5182                                                empty_cluster + empty_size);
5183                         if (ret == 0) {
5184                                 /*
5185                                  * now pull our allocation out of this
5186                                  * cluster
5187                                  */
5188                                 offset = btrfs_alloc_from_cluster(block_group,
5189                                                   last_ptr, num_bytes,
5190                                                   search_start);
5191                                 if (offset) {
5192                                         /* we found one, proceed */
5193                                         spin_unlock(&last_ptr->refill_lock);
5194                                         goto checks;
5195                                 }
5196                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5197                                    && !failed_cluster_refill) {
5198                                 spin_unlock(&last_ptr->refill_lock);
5199
5200                                 failed_cluster_refill = true;
5201                                 wait_block_group_cache_progress(block_group,
5202                                        num_bytes + empty_cluster + empty_size);
5203                                 goto have_block_group;
5204                         }
5205
5206                         /*
5207                          * at this point we either didn't find a cluster
5208                          * or we weren't able to allocate a block from our
5209                          * cluster.  Free the cluster we've been trying
5210                          * to use, and go to the next block group
5211                          */
5212                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5213                         spin_unlock(&last_ptr->refill_lock);
5214                         goto loop;
5215                 }
5216
5217                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5218                                                     num_bytes, empty_size);
5219                 /*
5220                  * If we didn't find a chunk, and we haven't failed on this
5221                  * block group before, and this block group is in the middle of
5222                  * caching and we are ok with waiting, then go ahead and wait
5223                  * for progress to be made, and set failed_alloc to true.
5224                  *
5225                  * If failed_alloc is true then we've already waited on this
5226                  * block group once and should move on to the next block group.
5227                  */
5228                 if (!offset && !failed_alloc && !cached &&
5229                     loop > LOOP_CACHING_NOWAIT) {
5230                         wait_block_group_cache_progress(block_group,
5231                                                 num_bytes + empty_size);
5232                         failed_alloc = true;
5233                         goto have_block_group;
5234                 } else if (!offset) {
5235                         goto loop;
5236                 }
5237 checks:
5238                 search_start = stripe_align(root, offset);
5239                 /* move on to the next group */
5240                 if (search_start + num_bytes >= search_end) {
5241                         btrfs_add_free_space(block_group, offset, num_bytes);
5242                         goto loop;
5243                 }
5244
5245                 /* move on to the next group */
5246                 if (search_start + num_bytes >
5247                     block_group->key.objectid + block_group->key.offset) {
5248                         btrfs_add_free_space(block_group, offset, num_bytes);
5249                         goto loop;
5250                 }
5251
5252                 ins->objectid = search_start;
5253                 ins->offset = num_bytes;
5254
5255                 if (offset < search_start)
5256                         btrfs_add_free_space(block_group, offset,
5257                                              search_start - offset);
5258                 BUG_ON(offset > search_start);
5259
5260                 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
5261                                             (data & BTRFS_BLOCK_GROUP_DATA));
5262                 if (ret == -EAGAIN) {
5263                         btrfs_add_free_space(block_group, offset, num_bytes);
5264                         goto loop;
5265                 }
5266
5267                 /* we are all good, lets return */
5268                 ins->objectid = search_start;
5269                 ins->offset = num_bytes;
5270
5271                 if (offset < search_start)
5272                         btrfs_add_free_space(block_group, offset,
5273                                              search_start - offset);
5274                 BUG_ON(offset > search_start);
5275                 break;
5276 loop:
5277                 failed_cluster_refill = false;
5278                 failed_alloc = false;
5279                 BUG_ON(index != get_block_group_index(block_group));
5280                 btrfs_put_block_group(block_group);
5281         }
5282         up_read(&space_info->groups_sem);
5283
5284         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5285                 goto search;
5286
5287         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5288          *                      for them to make caching progress.  Also
5289          *                      determine the best possible bg to cache
5290          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5291          *                      caching kthreads as we move along
5292          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5293          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5294          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5295          *                      again
5296          */
5297         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5298             (found_uncached_bg || empty_size || empty_cluster ||
5299              allowed_chunk_alloc)) {
5300                 index = 0;
5301                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5302                         found_uncached_bg = false;
5303                         loop++;
5304                         if (!ideal_cache_percent &&
5305                             atomic_read(&space_info->caching_threads))
5306                                 goto search;
5307
5308                         /*
5309                          * 1 of the following 2 things have happened so far
5310                          *
5311                          * 1) We found an ideal block group for caching that
5312                          * is mostly full and will cache quickly, so we might
5313                          * as well wait for it.
5314                          *
5315                          * 2) We searched for cached only and we didn't find
5316                          * anything, and we didn't start any caching kthreads
5317                          * either, so chances are we will loop through and
5318                          * start a couple caching kthreads, and then come back
5319                          * around and just wait for them.  This will be slower
5320                          * because we will have 2 caching kthreads reading at
5321                          * the same time when we could have just started one
5322                          * and waited for it to get far enough to give us an
5323                          * allocation, so go ahead and go to the wait caching
5324                          * loop.
5325                          */
5326                         loop = LOOP_CACHING_WAIT;
5327                         search_start = ideal_cache_offset;
5328                         ideal_cache_percent = 0;
5329                         goto ideal_cache;
5330                 } else if (loop == LOOP_FIND_IDEAL) {
5331                         /*
5332                          * Didn't find a uncached bg, wait on anything we find
5333                          * next.
5334                          */
5335                         loop = LOOP_CACHING_WAIT;
5336                         goto search;
5337                 }
5338
5339                 if (loop < LOOP_CACHING_WAIT) {
5340                         loop++;
5341                         goto search;
5342                 }
5343
5344                 if (loop == LOOP_ALLOC_CHUNK) {
5345                         empty_size = 0;
5346                         empty_cluster = 0;
5347                 }
5348
5349                 if (allowed_chunk_alloc) {
5350                         ret = do_chunk_alloc(trans, root, num_bytes +
5351                                              2 * 1024 * 1024, data,
5352                                              CHUNK_ALLOC_LIMITED);
5353                         allowed_chunk_alloc = 0;
5354                         done_chunk_alloc = 1;
5355                 } else if (!done_chunk_alloc &&
5356                            space_info->force_alloc == CHUNK_ALLOC_NO_FORCE) {
5357                         space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5358                 }
5359
5360                 if (loop < LOOP_NO_EMPTY_SIZE) {
5361                         loop++;
5362                         goto search;
5363                 }
5364                 ret = -ENOSPC;
5365         } else if (!ins->objectid) {
5366                 ret = -ENOSPC;
5367         }
5368
5369         /* we found what we needed */
5370         if (ins->objectid) {
5371                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5372                         trans->block_group = block_group->key.objectid;
5373
5374                 btrfs_put_block_group(block_group);
5375                 ret = 0;
5376         }
5377
5378         return ret;
5379 }
5380
5381 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5382                             int dump_block_groups)
5383 {
5384         struct btrfs_block_group_cache *cache;
5385         int index = 0;
5386
5387         spin_lock(&info->lock);
5388         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5389                (unsigned long long)(info->total_bytes - info->bytes_used -
5390                                     info->bytes_pinned - info->bytes_reserved -
5391                                     info->bytes_readonly),
5392                (info->full) ? "" : "not ");
5393         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5394                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5395                (unsigned long long)info->total_bytes,
5396                (unsigned long long)info->bytes_used,
5397                (unsigned long long)info->bytes_pinned,
5398                (unsigned long long)info->bytes_reserved,
5399                (unsigned long long)info->bytes_may_use,
5400                (unsigned long long)info->bytes_readonly);
5401         spin_unlock(&info->lock);
5402
5403         if (!dump_block_groups)
5404                 return;
5405
5406         down_read(&info->groups_sem);
5407 again:
5408         list_for_each_entry(cache, &info->block_groups[index], list) {
5409                 spin_lock(&cache->lock);
5410                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5411                        "%llu pinned %llu reserved\n",
5412                        (unsigned long long)cache->key.objectid,
5413                        (unsigned long long)cache->key.offset,
5414                        (unsigned long long)btrfs_block_group_used(&cache->item),
5415                        (unsigned long long)cache->pinned,
5416                        (unsigned long long)cache->reserved);
5417                 btrfs_dump_free_space(cache, bytes);
5418                 spin_unlock(&cache->lock);
5419         }
5420         if (++index < BTRFS_NR_RAID_TYPES)
5421                 goto again;
5422         up_read(&info->groups_sem);
5423 }
5424
5425 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5426                          struct btrfs_root *root,
5427                          u64 num_bytes, u64 min_alloc_size,
5428                          u64 empty_size, u64 hint_byte,
5429                          u64 search_end, struct btrfs_key *ins,
5430                          u64 data)
5431 {
5432         int ret;
5433         u64 search_start = 0;
5434
5435         data = btrfs_get_alloc_profile(root, data);
5436 again:
5437         /*
5438          * the only place that sets empty_size is btrfs_realloc_node, which
5439          * is not called recursively on allocations
5440          */
5441         if (empty_size || root->ref_cows)
5442                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5443                                      num_bytes + 2 * 1024 * 1024, data,
5444                                      CHUNK_ALLOC_NO_FORCE);
5445
5446         WARN_ON(num_bytes < root->sectorsize);
5447         ret = find_free_extent(trans, root, num_bytes, empty_size,
5448                                search_start, search_end, hint_byte,
5449                                ins, data);
5450
5451         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5452                 num_bytes = num_bytes >> 1;
5453                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5454                 num_bytes = max(num_bytes, min_alloc_size);
5455                 do_chunk_alloc(trans, root->fs_info->extent_root,
5456                                num_bytes, data, CHUNK_ALLOC_FORCE);
5457                 goto again;
5458         }
5459         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5460                 struct btrfs_space_info *sinfo;
5461
5462                 sinfo = __find_space_info(root->fs_info, data);
5463                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5464                        "wanted %llu\n", (unsigned long long)data,
5465                        (unsigned long long)num_bytes);
5466                 dump_space_info(sinfo, num_bytes, 1);
5467         }
5468
5469         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5470
5471         return ret;
5472 }
5473
5474 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5475 {
5476         struct btrfs_block_group_cache *cache;
5477         int ret = 0;
5478
5479         cache = btrfs_lookup_block_group(root->fs_info, start);
5480         if (!cache) {
5481                 printk(KERN_ERR "Unable to find block group for %llu\n",
5482                        (unsigned long long)start);
5483                 return -ENOSPC;
5484         }
5485
5486         if (btrfs_test_opt(root, DISCARD))
5487                 ret = btrfs_discard_extent(root, start, len, NULL);
5488
5489         btrfs_add_free_space(cache, start, len);
5490         btrfs_update_reserved_bytes(cache, len, 0, 1);
5491         btrfs_put_block_group(cache);
5492
5493         trace_btrfs_reserved_extent_free(root, start, len);
5494
5495         return ret;
5496 }
5497
5498 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5499                                       struct btrfs_root *root,
5500                                       u64 parent, u64 root_objectid,
5501                                       u64 flags, u64 owner, u64 offset,
5502                                       struct btrfs_key *ins, int ref_mod)
5503 {
5504         int ret;
5505         struct btrfs_fs_info *fs_info = root->fs_info;
5506         struct btrfs_extent_item *extent_item;
5507         struct btrfs_extent_inline_ref *iref;
5508         struct btrfs_path *path;
5509         struct extent_buffer *leaf;
5510         int type;
5511         u32 size;
5512
5513         if (parent > 0)
5514                 type = BTRFS_SHARED_DATA_REF_KEY;
5515         else
5516                 type = BTRFS_EXTENT_DATA_REF_KEY;
5517
5518         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5519
5520         path = btrfs_alloc_path();
5521         if (!path)
5522                 return -ENOMEM;
5523
5524         path->leave_spinning = 1;
5525         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5526                                       ins, size);
5527         BUG_ON(ret);
5528
5529         leaf = path->nodes[0];
5530         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5531                                      struct btrfs_extent_item);
5532         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5533         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5534         btrfs_set_extent_flags(leaf, extent_item,
5535                                flags | BTRFS_EXTENT_FLAG_DATA);
5536
5537         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5538         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5539         if (parent > 0) {
5540                 struct btrfs_shared_data_ref *ref;
5541                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5542                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5543                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5544         } else {
5545                 struct btrfs_extent_data_ref *ref;
5546                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5547                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5548                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5549                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5550                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5551         }
5552
5553         btrfs_mark_buffer_dirty(path->nodes[0]);
5554         btrfs_free_path(path);
5555
5556         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5557         if (ret) {
5558                 printk(KERN_ERR "btrfs update block group failed for %llu "
5559                        "%llu\n", (unsigned long long)ins->objectid,
5560                        (unsigned long long)ins->offset);
5561                 BUG();
5562         }
5563         return ret;
5564 }
5565
5566 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5567                                      struct btrfs_root *root,
5568                                      u64 parent, u64 root_objectid,
5569                                      u64 flags, struct btrfs_disk_key *key,
5570                                      int level, struct btrfs_key *ins)
5571 {
5572         int ret;
5573         struct btrfs_fs_info *fs_info = root->fs_info;
5574         struct btrfs_extent_item *extent_item;
5575         struct btrfs_tree_block_info *block_info;
5576         struct btrfs_extent_inline_ref *iref;
5577         struct btrfs_path *path;
5578         struct extent_buffer *leaf;
5579         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5580
5581         path = btrfs_alloc_path();
5582         BUG_ON(!path);
5583
5584         path->leave_spinning = 1;
5585         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5586                                       ins, size);
5587         BUG_ON(ret);
5588
5589         leaf = path->nodes[0];
5590         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5591                                      struct btrfs_extent_item);
5592         btrfs_set_extent_refs(leaf, extent_item, 1);
5593         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5594         btrfs_set_extent_flags(leaf, extent_item,
5595                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5596         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5597
5598         btrfs_set_tree_block_key(leaf, block_info, key);
5599         btrfs_set_tree_block_level(leaf, block_info, level);
5600
5601         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5602         if (parent > 0) {
5603                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5604                 btrfs_set_extent_inline_ref_type(leaf, iref,
5605                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5606                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5607         } else {
5608                 btrfs_set_extent_inline_ref_type(leaf, iref,
5609                                                  BTRFS_TREE_BLOCK_REF_KEY);
5610                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5611         }
5612
5613         btrfs_mark_buffer_dirty(leaf);
5614         btrfs_free_path(path);
5615
5616         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5617         if (ret) {
5618                 printk(KERN_ERR "btrfs update block group failed for %llu "
5619                        "%llu\n", (unsigned long long)ins->objectid,
5620                        (unsigned long long)ins->offset);
5621                 BUG();
5622         }
5623         return ret;
5624 }
5625
5626 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5627                                      struct btrfs_root *root,
5628                                      u64 root_objectid, u64 owner,
5629                                      u64 offset, struct btrfs_key *ins)
5630 {
5631         int ret;
5632
5633         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5634
5635         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5636                                          0, root_objectid, owner, offset,
5637                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5638         return ret;
5639 }
5640
5641 /*
5642  * this is used by the tree logging recovery code.  It records that
5643  * an extent has been allocated and makes sure to clear the free
5644  * space cache bits as well
5645  */
5646 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5647                                    struct btrfs_root *root,
5648                                    u64 root_objectid, u64 owner, u64 offset,
5649                                    struct btrfs_key *ins)
5650 {
5651         int ret;
5652         struct btrfs_block_group_cache *block_group;
5653         struct btrfs_caching_control *caching_ctl;
5654         u64 start = ins->objectid;
5655         u64 num_bytes = ins->offset;
5656
5657         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5658         cache_block_group(block_group, trans, NULL, 0);
5659         caching_ctl = get_caching_control(block_group);
5660
5661         if (!caching_ctl) {
5662                 BUG_ON(!block_group_cache_done(block_group));
5663                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5664                 BUG_ON(ret);
5665         } else {
5666                 mutex_lock(&caching_ctl->mutex);
5667
5668                 if (start >= caching_ctl->progress) {
5669                         ret = add_excluded_extent(root, start, num_bytes);
5670                         BUG_ON(ret);
5671                 } else if (start + num_bytes <= caching_ctl->progress) {
5672                         ret = btrfs_remove_free_space(block_group,
5673                                                       start, num_bytes);
5674                         BUG_ON(ret);
5675                 } else {
5676                         num_bytes = caching_ctl->progress - start;
5677                         ret = btrfs_remove_free_space(block_group,
5678                                                       start, num_bytes);
5679                         BUG_ON(ret);
5680
5681                         start = caching_ctl->progress;
5682                         num_bytes = ins->objectid + ins->offset -
5683                                     caching_ctl->progress;
5684                         ret = add_excluded_extent(root, start, num_bytes);
5685                         BUG_ON(ret);
5686                 }
5687
5688                 mutex_unlock(&caching_ctl->mutex);
5689                 put_caching_control(caching_ctl);
5690         }
5691
5692         ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
5693         BUG_ON(ret);
5694         btrfs_put_block_group(block_group);
5695         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5696                                          0, owner, offset, ins, 1);
5697         return ret;
5698 }
5699
5700 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5701                                             struct btrfs_root *root,
5702                                             u64 bytenr, u32 blocksize,
5703                                             int level)
5704 {
5705         struct extent_buffer *buf;
5706
5707         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5708         if (!buf)
5709                 return ERR_PTR(-ENOMEM);
5710         btrfs_set_header_generation(buf, trans->transid);
5711         btrfs_set_buffer_lockdep_class(buf, level);
5712         btrfs_tree_lock(buf);
5713         clean_tree_block(trans, root, buf);
5714
5715         btrfs_set_lock_blocking(buf);
5716         btrfs_set_buffer_uptodate(buf);
5717
5718         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5719                 /*
5720                  * we allow two log transactions at a time, use different
5721                  * EXENT bit to differentiate dirty pages.
5722                  */
5723                 if (root->log_transid % 2 == 0)
5724                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5725                                         buf->start + buf->len - 1, GFP_NOFS);
5726                 else
5727                         set_extent_new(&root->dirty_log_pages, buf->start,
5728                                         buf->start + buf->len - 1, GFP_NOFS);
5729         } else {
5730                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5731                          buf->start + buf->len - 1, GFP_NOFS);
5732         }
5733         trans->blocks_used++;
5734         /* this returns a buffer locked for blocking */
5735         return buf;
5736 }
5737
5738 static struct btrfs_block_rsv *
5739 use_block_rsv(struct btrfs_trans_handle *trans,
5740               struct btrfs_root *root, u32 blocksize)
5741 {
5742         struct btrfs_block_rsv *block_rsv;
5743         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5744         int ret;
5745
5746         block_rsv = get_block_rsv(trans, root);
5747
5748         if (block_rsv->size == 0) {
5749                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5750                                              blocksize, 0);
5751                 /*
5752                  * If we couldn't reserve metadata bytes try and use some from
5753                  * the global reserve.
5754                  */
5755                 if (ret && block_rsv != global_rsv) {
5756                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5757                         if (!ret)
5758                                 return global_rsv;
5759                         return ERR_PTR(ret);
5760                 } else if (ret) {
5761                         return ERR_PTR(ret);
5762                 }
5763                 return block_rsv;
5764         }
5765
5766         ret = block_rsv_use_bytes(block_rsv, blocksize);
5767         if (!ret)
5768                 return block_rsv;
5769         if (ret) {
5770                 WARN_ON(1);
5771                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5772                                              0);
5773                 if (!ret) {
5774                         spin_lock(&block_rsv->lock);
5775                         block_rsv->size += blocksize;
5776                         spin_unlock(&block_rsv->lock);
5777                         return block_rsv;
5778                 } else if (ret && block_rsv != global_rsv) {
5779                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5780                         if (!ret)
5781                                 return global_rsv;
5782                 }
5783         }
5784
5785         return ERR_PTR(-ENOSPC);
5786 }
5787
5788 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5789 {
5790         block_rsv_add_bytes(block_rsv, blocksize, 0);
5791         block_rsv_release_bytes(block_rsv, NULL, 0);
5792 }
5793
5794 /*
5795  * finds a free extent and does all the dirty work required for allocation
5796  * returns the key for the extent through ins, and a tree buffer for
5797  * the first block of the extent through buf.
5798  *
5799  * returns the tree buffer or NULL.
5800  */
5801 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5802                                         struct btrfs_root *root, u32 blocksize,
5803                                         u64 parent, u64 root_objectid,
5804                                         struct btrfs_disk_key *key, int level,
5805                                         u64 hint, u64 empty_size)
5806 {
5807         struct btrfs_key ins;
5808         struct btrfs_block_rsv *block_rsv;
5809         struct extent_buffer *buf;
5810         u64 flags = 0;
5811         int ret;
5812
5813
5814         block_rsv = use_block_rsv(trans, root, blocksize);
5815         if (IS_ERR(block_rsv))
5816                 return ERR_CAST(block_rsv);
5817
5818         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5819                                    empty_size, hint, (u64)-1, &ins, 0);
5820         if (ret) {
5821                 unuse_block_rsv(block_rsv, blocksize);
5822                 return ERR_PTR(ret);
5823         }
5824
5825         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5826                                     blocksize, level);
5827         BUG_ON(IS_ERR(buf));
5828
5829         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5830                 if (parent == 0)
5831                         parent = ins.objectid;
5832                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5833         } else
5834                 BUG_ON(parent > 0);
5835
5836         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5837                 struct btrfs_delayed_extent_op *extent_op;
5838                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5839                 BUG_ON(!extent_op);
5840                 if (key)
5841                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5842                 else
5843                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5844                 extent_op->flags_to_set = flags;
5845                 extent_op->update_key = 1;
5846                 extent_op->update_flags = 1;
5847                 extent_op->is_data = 0;
5848
5849                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5850                                         ins.offset, parent, root_objectid,
5851                                         level, BTRFS_ADD_DELAYED_EXTENT,
5852                                         extent_op);
5853                 BUG_ON(ret);
5854         }
5855         return buf;
5856 }
5857
5858 struct walk_control {
5859         u64 refs[BTRFS_MAX_LEVEL];
5860         u64 flags[BTRFS_MAX_LEVEL];
5861         struct btrfs_key update_progress;
5862         int stage;
5863         int level;
5864         int shared_level;
5865         int update_ref;
5866         int keep_locks;
5867         int reada_slot;
5868         int reada_count;
5869 };
5870
5871 #define DROP_REFERENCE  1
5872 #define UPDATE_BACKREF  2
5873
5874 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5875                                      struct btrfs_root *root,
5876                                      struct walk_control *wc,
5877                                      struct btrfs_path *path)
5878 {
5879         u64 bytenr;
5880         u64 generation;
5881         u64 refs;
5882         u64 flags;
5883         u32 nritems;
5884         u32 blocksize;
5885         struct btrfs_key key;
5886         struct extent_buffer *eb;
5887         int ret;
5888         int slot;
5889         int nread = 0;
5890
5891         if (path->slots[wc->level] < wc->reada_slot) {
5892                 wc->reada_count = wc->reada_count * 2 / 3;
5893                 wc->reada_count = max(wc->reada_count, 2);
5894         } else {
5895                 wc->reada_count = wc->reada_count * 3 / 2;
5896                 wc->reada_count = min_t(int, wc->reada_count,
5897                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5898         }
5899
5900         eb = path->nodes[wc->level];
5901         nritems = btrfs_header_nritems(eb);
5902         blocksize = btrfs_level_size(root, wc->level - 1);
5903
5904         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5905                 if (nread >= wc->reada_count)
5906                         break;
5907
5908                 cond_resched();
5909                 bytenr = btrfs_node_blockptr(eb, slot);
5910                 generation = btrfs_node_ptr_generation(eb, slot);
5911
5912                 if (slot == path->slots[wc->level])
5913                         goto reada;
5914
5915                 if (wc->stage == UPDATE_BACKREF &&
5916                     generation <= root->root_key.offset)
5917                         continue;
5918
5919                 /* We don't lock the tree block, it's OK to be racy here */
5920                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5921                                                &refs, &flags);
5922                 BUG_ON(ret);
5923                 BUG_ON(refs == 0);
5924
5925                 if (wc->stage == DROP_REFERENCE) {
5926                         if (refs == 1)
5927                                 goto reada;
5928
5929                         if (wc->level == 1 &&
5930                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5931                                 continue;
5932                         if (!wc->update_ref ||
5933                             generation <= root->root_key.offset)
5934                                 continue;
5935                         btrfs_node_key_to_cpu(eb, &key, slot);
5936                         ret = btrfs_comp_cpu_keys(&key,
5937                                                   &wc->update_progress);
5938                         if (ret < 0)
5939                                 continue;
5940                 } else {
5941                         if (wc->level == 1 &&
5942                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5943                                 continue;
5944                 }
5945 reada:
5946                 ret = readahead_tree_block(root, bytenr, blocksize,
5947                                            generation);
5948                 if (ret)
5949                         break;
5950                 nread++;
5951         }
5952         wc->reada_slot = slot;
5953 }
5954
5955 /*
5956  * hepler to process tree block while walking down the tree.
5957  *
5958  * when wc->stage == UPDATE_BACKREF, this function updates
5959  * back refs for pointers in the block.
5960  *
5961  * NOTE: return value 1 means we should stop walking down.
5962  */
5963 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5964                                    struct btrfs_root *root,
5965                                    struct btrfs_path *path,
5966                                    struct walk_control *wc, int lookup_info)
5967 {
5968         int level = wc->level;
5969         struct extent_buffer *eb = path->nodes[level];
5970         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5971         int ret;
5972
5973         if (wc->stage == UPDATE_BACKREF &&
5974             btrfs_header_owner(eb) != root->root_key.objectid)
5975                 return 1;
5976
5977         /*
5978          * when reference count of tree block is 1, it won't increase
5979          * again. once full backref flag is set, we never clear it.
5980          */
5981         if (lookup_info &&
5982             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5983              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5984                 BUG_ON(!path->locks[level]);
5985                 ret = btrfs_lookup_extent_info(trans, root,
5986                                                eb->start, eb->len,
5987                                                &wc->refs[level],
5988                                                &wc->flags[level]);
5989                 BUG_ON(ret);
5990                 BUG_ON(wc->refs[level] == 0);
5991         }
5992
5993         if (wc->stage == DROP_REFERENCE) {
5994                 if (wc->refs[level] > 1)
5995                         return 1;
5996
5997                 if (path->locks[level] && !wc->keep_locks) {
5998                         btrfs_tree_unlock(eb);
5999                         path->locks[level] = 0;
6000                 }
6001                 return 0;
6002         }
6003
6004         /* wc->stage == UPDATE_BACKREF */
6005         if (!(wc->flags[level] & flag)) {
6006                 BUG_ON(!path->locks[level]);
6007                 ret = btrfs_inc_ref(trans, root, eb, 1);
6008                 BUG_ON(ret);
6009                 ret = btrfs_dec_ref(trans, root, eb, 0);
6010                 BUG_ON(ret);
6011                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6012                                                   eb->len, flag, 0);
6013                 BUG_ON(ret);
6014                 wc->flags[level] |= flag;
6015         }
6016
6017         /*
6018          * the block is shared by multiple trees, so it's not good to
6019          * keep the tree lock
6020          */
6021         if (path->locks[level] && level > 0) {
6022                 btrfs_tree_unlock(eb);
6023                 path->locks[level] = 0;
6024         }
6025         return 0;
6026 }
6027
6028 /*
6029  * hepler to process tree block pointer.
6030  *
6031  * when wc->stage == DROP_REFERENCE, this function checks
6032  * reference count of the block pointed to. if the block
6033  * is shared and we need update back refs for the subtree
6034  * rooted at the block, this function changes wc->stage to
6035  * UPDATE_BACKREF. if the block is shared and there is no
6036  * need to update back, this function drops the reference
6037  * to the block.
6038  *
6039  * NOTE: return value 1 means we should stop walking down.
6040  */
6041 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6042                                  struct btrfs_root *root,
6043                                  struct btrfs_path *path,
6044                                  struct walk_control *wc, int *lookup_info)
6045 {
6046         u64 bytenr;
6047         u64 generation;
6048         u64 parent;
6049         u32 blocksize;
6050         struct btrfs_key key;
6051         struct extent_buffer *next;
6052         int level = wc->level;
6053         int reada = 0;
6054         int ret = 0;
6055
6056         generation = btrfs_node_ptr_generation(path->nodes[level],
6057                                                path->slots[level]);
6058         /*
6059          * if the lower level block was created before the snapshot
6060          * was created, we know there is no need to update back refs
6061          * for the subtree
6062          */
6063         if (wc->stage == UPDATE_BACKREF &&
6064             generation <= root->root_key.offset) {
6065                 *lookup_info = 1;
6066                 return 1;
6067         }
6068
6069         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6070         blocksize = btrfs_level_size(root, level - 1);
6071
6072         next = btrfs_find_tree_block(root, bytenr, blocksize);
6073         if (!next) {
6074                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6075                 if (!next)
6076                         return -ENOMEM;
6077                 reada = 1;
6078         }
6079         btrfs_tree_lock(next);
6080         btrfs_set_lock_blocking(next);
6081
6082         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6083                                        &wc->refs[level - 1],
6084                                        &wc->flags[level - 1]);
6085         BUG_ON(ret);
6086         BUG_ON(wc->refs[level - 1] == 0);
6087         *lookup_info = 0;
6088
6089         if (wc->stage == DROP_REFERENCE) {
6090                 if (wc->refs[level - 1] > 1) {
6091                         if (level == 1 &&
6092                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6093                                 goto skip;
6094
6095                         if (!wc->update_ref ||
6096                             generation <= root->root_key.offset)
6097                                 goto skip;
6098
6099                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6100                                               path->slots[level]);
6101                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6102                         if (ret < 0)
6103                                 goto skip;
6104
6105                         wc->stage = UPDATE_BACKREF;
6106                         wc->shared_level = level - 1;
6107                 }
6108         } else {
6109                 if (level == 1 &&
6110                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6111                         goto skip;
6112         }
6113
6114         if (!btrfs_buffer_uptodate(next, generation)) {
6115                 btrfs_tree_unlock(next);
6116                 free_extent_buffer(next);
6117                 next = NULL;
6118                 *lookup_info = 1;
6119         }
6120
6121         if (!next) {
6122                 if (reada && level == 1)
6123                         reada_walk_down(trans, root, wc, path);
6124                 next = read_tree_block(root, bytenr, blocksize, generation);
6125                 if (!next)
6126                         return -EIO;
6127                 btrfs_tree_lock(next);
6128                 btrfs_set_lock_blocking(next);
6129         }
6130
6131         level--;
6132         BUG_ON(level != btrfs_header_level(next));
6133         path->nodes[level] = next;
6134         path->slots[level] = 0;
6135         path->locks[level] = 1;
6136         wc->level = level;
6137         if (wc->level == 1)
6138                 wc->reada_slot = 0;
6139         return 0;
6140 skip:
6141         wc->refs[level - 1] = 0;
6142         wc->flags[level - 1] = 0;
6143         if (wc->stage == DROP_REFERENCE) {
6144                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6145                         parent = path->nodes[level]->start;
6146                 } else {
6147                         BUG_ON(root->root_key.objectid !=
6148                                btrfs_header_owner(path->nodes[level]));
6149                         parent = 0;
6150                 }
6151
6152                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6153                                         root->root_key.objectid, level - 1, 0);
6154                 BUG_ON(ret);
6155         }
6156         btrfs_tree_unlock(next);
6157         free_extent_buffer(next);
6158         *lookup_info = 1;
6159         return 1;
6160 }
6161
6162 /*
6163  * hepler to process tree block while walking up the tree.
6164  *
6165  * when wc->stage == DROP_REFERENCE, this function drops
6166  * reference count on the block.
6167  *
6168  * when wc->stage == UPDATE_BACKREF, this function changes
6169  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6170  * to UPDATE_BACKREF previously while processing the block.
6171  *
6172  * NOTE: return value 1 means we should stop walking up.
6173  */
6174 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6175                                  struct btrfs_root *root,
6176                                  struct btrfs_path *path,
6177                                  struct walk_control *wc)
6178 {
6179         int ret;
6180         int level = wc->level;
6181         struct extent_buffer *eb = path->nodes[level];
6182         u64 parent = 0;
6183
6184         if (wc->stage == UPDATE_BACKREF) {
6185                 BUG_ON(wc->shared_level < level);
6186                 if (level < wc->shared_level)
6187                         goto out;
6188
6189                 ret = find_next_key(path, level + 1, &wc->update_progress);
6190                 if (ret > 0)
6191                         wc->update_ref = 0;
6192
6193                 wc->stage = DROP_REFERENCE;
6194                 wc->shared_level = -1;
6195                 path->slots[level] = 0;
6196
6197                 /*
6198                  * check reference count again if the block isn't locked.
6199                  * we should start walking down the tree again if reference
6200                  * count is one.
6201                  */
6202                 if (!path->locks[level]) {
6203                         BUG_ON(level == 0);
6204                         btrfs_tree_lock(eb);
6205                         btrfs_set_lock_blocking(eb);
6206                         path->locks[level] = 1;
6207
6208                         ret = btrfs_lookup_extent_info(trans, root,
6209                                                        eb->start, eb->len,
6210                                                        &wc->refs[level],
6211                                                        &wc->flags[level]);
6212                         BUG_ON(ret);
6213                         BUG_ON(wc->refs[level] == 0);
6214                         if (wc->refs[level] == 1) {
6215                                 btrfs_tree_unlock(eb);
6216                                 path->locks[level] = 0;
6217                                 return 1;
6218                         }
6219                 }
6220         }
6221
6222         /* wc->stage == DROP_REFERENCE */
6223         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6224
6225         if (wc->refs[level] == 1) {
6226                 if (level == 0) {
6227                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6228                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6229                         else
6230                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6231                         BUG_ON(ret);
6232                 }
6233                 /* make block locked assertion in clean_tree_block happy */
6234                 if (!path->locks[level] &&
6235                     btrfs_header_generation(eb) == trans->transid) {
6236                         btrfs_tree_lock(eb);
6237                         btrfs_set_lock_blocking(eb);
6238                         path->locks[level] = 1;
6239                 }
6240                 clean_tree_block(trans, root, eb);
6241         }
6242
6243         if (eb == root->node) {
6244                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6245                         parent = eb->start;
6246                 else
6247                         BUG_ON(root->root_key.objectid !=
6248                                btrfs_header_owner(eb));
6249         } else {
6250                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6251                         parent = path->nodes[level + 1]->start;
6252                 else
6253                         BUG_ON(root->root_key.objectid !=
6254                                btrfs_header_owner(path->nodes[level + 1]));
6255         }
6256
6257         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6258 out:
6259         wc->refs[level] = 0;
6260         wc->flags[level] = 0;
6261         return 0;
6262 }
6263
6264 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6265                                    struct btrfs_root *root,
6266                                    struct btrfs_path *path,
6267                                    struct walk_control *wc)
6268 {
6269         int level = wc->level;
6270         int lookup_info = 1;
6271         int ret;
6272
6273         while (level >= 0) {
6274                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6275                 if (ret > 0)
6276                         break;
6277
6278                 if (level == 0)
6279                         break;
6280
6281                 if (path->slots[level] >=
6282                     btrfs_header_nritems(path->nodes[level]))
6283                         break;
6284
6285                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6286                 if (ret > 0) {
6287                         path->slots[level]++;
6288                         continue;
6289                 } else if (ret < 0)
6290                         return ret;
6291                 level = wc->level;
6292         }
6293         return 0;
6294 }
6295
6296 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6297                                  struct btrfs_root *root,
6298                                  struct btrfs_path *path,
6299                                  struct walk_control *wc, int max_level)
6300 {
6301         int level = wc->level;
6302         int ret;
6303
6304         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6305         while (level < max_level && path->nodes[level]) {
6306                 wc->level = level;
6307                 if (path->slots[level] + 1 <
6308                     btrfs_header_nritems(path->nodes[level])) {
6309                         path->slots[level]++;
6310                         return 0;
6311                 } else {
6312                         ret = walk_up_proc(trans, root, path, wc);
6313                         if (ret > 0)
6314                                 return 0;
6315
6316                         if (path->locks[level]) {
6317                                 btrfs_tree_unlock(path->nodes[level]);
6318                                 path->locks[level] = 0;
6319                         }
6320                         free_extent_buffer(path->nodes[level]);
6321                         path->nodes[level] = NULL;
6322                         level++;
6323                 }
6324         }
6325         return 1;
6326 }
6327
6328 /*
6329  * drop a subvolume tree.
6330  *
6331  * this function traverses the tree freeing any blocks that only
6332  * referenced by the tree.
6333  *
6334  * when a shared tree block is found. this function decreases its
6335  * reference count by one. if update_ref is true, this function
6336  * also make sure backrefs for the shared block and all lower level
6337  * blocks are properly updated.
6338  */
6339 int btrfs_drop_snapshot(struct btrfs_root *root,
6340                         struct btrfs_block_rsv *block_rsv, int update_ref)
6341 {
6342         struct btrfs_path *path;
6343         struct btrfs_trans_handle *trans;
6344         struct btrfs_root *tree_root = root->fs_info->tree_root;
6345         struct btrfs_root_item *root_item = &root->root_item;
6346         struct walk_control *wc;
6347         struct btrfs_key key;
6348         int err = 0;
6349         int ret;
6350         int level;
6351
6352         path = btrfs_alloc_path();
6353         BUG_ON(!path);
6354
6355         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6356         BUG_ON(!wc);
6357
6358         trans = btrfs_start_transaction(tree_root, 0);
6359         BUG_ON(IS_ERR(trans));
6360
6361         if (block_rsv)
6362                 trans->block_rsv = block_rsv;
6363
6364         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6365                 level = btrfs_header_level(root->node);
6366                 path->nodes[level] = btrfs_lock_root_node(root);
6367                 btrfs_set_lock_blocking(path->nodes[level]);
6368                 path->slots[level] = 0;
6369                 path->locks[level] = 1;
6370                 memset(&wc->update_progress, 0,
6371                        sizeof(wc->update_progress));
6372         } else {
6373                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6374                 memcpy(&wc->update_progress, &key,
6375                        sizeof(wc->update_progress));
6376
6377                 level = root_item->drop_level;
6378                 BUG_ON(level == 0);
6379                 path->lowest_level = level;
6380                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6381                 path->lowest_level = 0;
6382                 if (ret < 0) {
6383                         err = ret;
6384                         goto out;
6385                 }
6386                 WARN_ON(ret > 0);
6387
6388                 /*
6389                  * unlock our path, this is safe because only this
6390                  * function is allowed to delete this snapshot
6391                  */
6392                 btrfs_unlock_up_safe(path, 0);
6393
6394                 level = btrfs_header_level(root->node);
6395                 while (1) {
6396                         btrfs_tree_lock(path->nodes[level]);
6397                         btrfs_set_lock_blocking(path->nodes[level]);
6398
6399                         ret = btrfs_lookup_extent_info(trans, root,
6400                                                 path->nodes[level]->start,
6401                                                 path->nodes[level]->len,
6402                                                 &wc->refs[level],
6403                                                 &wc->flags[level]);
6404                         BUG_ON(ret);
6405                         BUG_ON(wc->refs[level] == 0);
6406
6407                         if (level == root_item->drop_level)
6408                                 break;
6409
6410                         btrfs_tree_unlock(path->nodes[level]);
6411                         WARN_ON(wc->refs[level] != 1);
6412                         level--;
6413                 }
6414         }
6415
6416         wc->level = level;
6417         wc->shared_level = -1;
6418         wc->stage = DROP_REFERENCE;
6419         wc->update_ref = update_ref;
6420         wc->keep_locks = 0;
6421         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6422
6423         while (1) {
6424                 ret = walk_down_tree(trans, root, path, wc);
6425                 if (ret < 0) {
6426                         err = ret;
6427                         break;
6428                 }
6429
6430                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6431                 if (ret < 0) {
6432                         err = ret;
6433                         break;
6434                 }
6435
6436                 if (ret > 0) {
6437                         BUG_ON(wc->stage != DROP_REFERENCE);
6438                         break;
6439                 }
6440
6441                 if (wc->stage == DROP_REFERENCE) {
6442                         level = wc->level;
6443                         btrfs_node_key(path->nodes[level],
6444                                        &root_item->drop_progress,
6445                                        path->slots[level]);
6446                         root_item->drop_level = level;
6447                 }
6448
6449                 BUG_ON(wc->level == 0);
6450                 if (btrfs_should_end_transaction(trans, tree_root)) {
6451                         ret = btrfs_update_root(trans, tree_root,
6452                                                 &root->root_key,
6453                                                 root_item);
6454                         BUG_ON(ret);
6455
6456                         btrfs_end_transaction_throttle(trans, tree_root);
6457                         trans = btrfs_start_transaction(tree_root, 0);
6458                         BUG_ON(IS_ERR(trans));
6459                         if (block_rsv)
6460                                 trans->block_rsv = block_rsv;
6461                 }
6462         }
6463         btrfs_release_path(root, path);
6464         BUG_ON(err);
6465
6466         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6467         BUG_ON(ret);
6468
6469         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6470                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6471                                            NULL, NULL);
6472                 BUG_ON(ret < 0);
6473                 if (ret > 0) {
6474                         /* if we fail to delete the orphan item this time
6475                          * around, it'll get picked up the next time.
6476                          *
6477                          * The most common failure here is just -ENOENT.
6478                          */
6479                         btrfs_del_orphan_item(trans, tree_root,
6480                                               root->root_key.objectid);
6481                 }
6482         }
6483
6484         if (root->in_radix) {
6485                 btrfs_free_fs_root(tree_root->fs_info, root);
6486         } else {
6487                 free_extent_buffer(root->node);
6488                 free_extent_buffer(root->commit_root);
6489                 kfree(root);
6490         }
6491 out:
6492         btrfs_end_transaction_throttle(trans, tree_root);
6493         kfree(wc);
6494         btrfs_free_path(path);
6495         return err;
6496 }
6497
6498 /*
6499  * drop subtree rooted at tree block 'node'.
6500  *
6501  * NOTE: this function will unlock and release tree block 'node'
6502  */
6503 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6504                         struct btrfs_root *root,
6505                         struct extent_buffer *node,
6506                         struct extent_buffer *parent)
6507 {
6508         struct btrfs_path *path;
6509         struct walk_control *wc;
6510         int level;
6511         int parent_level;
6512         int ret = 0;
6513         int wret;
6514
6515         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6516
6517         path = btrfs_alloc_path();
6518         if (!path)
6519                 return -ENOMEM;
6520
6521         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6522         if (!wc) {
6523                 btrfs_free_path(path);
6524                 return -ENOMEM;
6525         }
6526
6527         btrfs_assert_tree_locked(parent);
6528         parent_level = btrfs_header_level(parent);
6529         extent_buffer_get(parent);
6530         path->nodes[parent_level] = parent;
6531         path->slots[parent_level] = btrfs_header_nritems(parent);
6532
6533         btrfs_assert_tree_locked(node);
6534         level = btrfs_header_level(node);
6535         path->nodes[level] = node;
6536         path->slots[level] = 0;
6537         path->locks[level] = 1;
6538
6539         wc->refs[parent_level] = 1;
6540         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6541         wc->level = level;
6542         wc->shared_level = -1;
6543         wc->stage = DROP_REFERENCE;
6544         wc->update_ref = 0;
6545         wc->keep_locks = 1;
6546         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6547
6548         while (1) {
6549                 wret = walk_down_tree(trans, root, path, wc);
6550                 if (wret < 0) {
6551                         ret = wret;
6552                         break;
6553                 }
6554
6555                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6556                 if (wret < 0)
6557                         ret = wret;
6558                 if (wret != 0)
6559                         break;
6560         }
6561
6562         kfree(wc);
6563         btrfs_free_path(path);
6564         return ret;
6565 }
6566
6567 #if 0
6568 static unsigned long calc_ra(unsigned long start, unsigned long last,
6569                              unsigned long nr)
6570 {
6571         return min(last, start + nr - 1);
6572 }
6573
6574 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6575                                          u64 len)
6576 {
6577         u64 page_start;
6578         u64 page_end;
6579         unsigned long first_index;
6580         unsigned long last_index;
6581         unsigned long i;
6582         struct page *page;
6583         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6584         struct file_ra_state *ra;
6585         struct btrfs_ordered_extent *ordered;
6586         unsigned int total_read = 0;
6587         unsigned int total_dirty = 0;
6588         int ret = 0;
6589
6590         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6591         if (!ra)
6592                 return -ENOMEM;
6593
6594         mutex_lock(&inode->i_mutex);
6595         first_index = start >> PAGE_CACHE_SHIFT;
6596         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6597
6598         /* make sure the dirty trick played by the caller work */
6599         ret = invalidate_inode_pages2_range(inode->i_mapping,
6600                                             first_index, last_index);
6601         if (ret)
6602                 goto out_unlock;
6603
6604         file_ra_state_init(ra, inode->i_mapping);
6605
6606         for (i = first_index ; i <= last_index; i++) {
6607                 if (total_read % ra->ra_pages == 0) {
6608                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6609                                        calc_ra(i, last_index, ra->ra_pages));
6610                 }
6611                 total_read++;
6612 again:
6613                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6614                         BUG_ON(1);
6615                 page = grab_cache_page(inode->i_mapping, i);
6616                 if (!page) {
6617                         ret = -ENOMEM;
6618                         goto out_unlock;
6619                 }
6620                 if (!PageUptodate(page)) {
6621                         btrfs_readpage(NULL, page);
6622                         lock_page(page);
6623                         if (!PageUptodate(page)) {
6624                                 unlock_page(page);
6625                                 page_cache_release(page);
6626                                 ret = -EIO;
6627                                 goto out_unlock;
6628                         }
6629                 }
6630                 wait_on_page_writeback(page);
6631
6632                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6633                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6634                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6635
6636                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6637                 if (ordered) {
6638                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6639                         unlock_page(page);
6640                         page_cache_release(page);
6641                         btrfs_start_ordered_extent(inode, ordered, 1);
6642                         btrfs_put_ordered_extent(ordered);
6643                         goto again;
6644                 }
6645                 set_page_extent_mapped(page);
6646
6647                 if (i == first_index)
6648                         set_extent_bits(io_tree, page_start, page_end,
6649                                         EXTENT_BOUNDARY, GFP_NOFS);
6650                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6651
6652                 set_page_dirty(page);
6653                 total_dirty++;
6654
6655                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6656                 unlock_page(page);
6657                 page_cache_release(page);
6658         }
6659
6660 out_unlock:
6661         kfree(ra);
6662         mutex_unlock(&inode->i_mutex);
6663         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6664         return ret;
6665 }
6666
6667 static noinline int relocate_data_extent(struct inode *reloc_inode,
6668                                          struct btrfs_key *extent_key,
6669                                          u64 offset)
6670 {
6671         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6672         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6673         struct extent_map *em;
6674         u64 start = extent_key->objectid - offset;
6675         u64 end = start + extent_key->offset - 1;
6676
6677         em = alloc_extent_map(GFP_NOFS);
6678         BUG_ON(!em);
6679
6680         em->start = start;
6681         em->len = extent_key->offset;
6682         em->block_len = extent_key->offset;
6683         em->block_start = extent_key->objectid;
6684         em->bdev = root->fs_info->fs_devices->latest_bdev;
6685         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6686
6687         /* setup extent map to cheat btrfs_readpage */
6688         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6689         while (1) {
6690                 int ret;
6691                 write_lock(&em_tree->lock);
6692                 ret = add_extent_mapping(em_tree, em);
6693                 write_unlock(&em_tree->lock);
6694                 if (ret != -EEXIST) {
6695                         free_extent_map(em);
6696                         break;
6697                 }
6698                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6699         }
6700         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6701
6702         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6703 }
6704
6705 struct btrfs_ref_path {
6706         u64 extent_start;
6707         u64 nodes[BTRFS_MAX_LEVEL];
6708         u64 root_objectid;
6709         u64 root_generation;
6710         u64 owner_objectid;
6711         u32 num_refs;
6712         int lowest_level;
6713         int current_level;
6714         int shared_level;
6715
6716         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6717         u64 new_nodes[BTRFS_MAX_LEVEL];
6718 };
6719
6720 struct disk_extent {
6721         u64 ram_bytes;
6722         u64 disk_bytenr;
6723         u64 disk_num_bytes;
6724         u64 offset;
6725         u64 num_bytes;
6726         u8 compression;
6727         u8 encryption;
6728         u16 other_encoding;
6729 };
6730
6731 static int is_cowonly_root(u64 root_objectid)
6732 {
6733         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6734             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6735             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6736             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6737             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6738             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6739                 return 1;
6740         return 0;
6741 }
6742
6743 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6744                                     struct btrfs_root *extent_root,
6745                                     struct btrfs_ref_path *ref_path,
6746                                     int first_time)
6747 {
6748         struct extent_buffer *leaf;
6749         struct btrfs_path *path;
6750         struct btrfs_extent_ref *ref;
6751         struct btrfs_key key;
6752         struct btrfs_key found_key;
6753         u64 bytenr;
6754         u32 nritems;
6755         int level;
6756         int ret = 1;
6757
6758         path = btrfs_alloc_path();
6759         if (!path)
6760                 return -ENOMEM;
6761
6762         if (first_time) {
6763                 ref_path->lowest_level = -1;
6764                 ref_path->current_level = -1;
6765                 ref_path->shared_level = -1;
6766                 goto walk_up;
6767         }
6768 walk_down:
6769         level = ref_path->current_level - 1;
6770         while (level >= -1) {
6771                 u64 parent;
6772                 if (level < ref_path->lowest_level)
6773                         break;
6774
6775                 if (level >= 0)
6776                         bytenr = ref_path->nodes[level];
6777                 else
6778                         bytenr = ref_path->extent_start;
6779                 BUG_ON(bytenr == 0);
6780
6781                 parent = ref_path->nodes[level + 1];
6782                 ref_path->nodes[level + 1] = 0;
6783                 ref_path->current_level = level;
6784                 BUG_ON(parent == 0);
6785
6786                 key.objectid = bytenr;
6787                 key.offset = parent + 1;
6788                 key.type = BTRFS_EXTENT_REF_KEY;
6789
6790                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6791                 if (ret < 0)
6792                         goto out;
6793                 BUG_ON(ret == 0);
6794
6795                 leaf = path->nodes[0];
6796                 nritems = btrfs_header_nritems(leaf);
6797                 if (path->slots[0] >= nritems) {
6798                         ret = btrfs_next_leaf(extent_root, path);
6799                         if (ret < 0)
6800                                 goto out;
6801                         if (ret > 0)
6802                                 goto next;
6803                         leaf = path->nodes[0];
6804                 }
6805
6806                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6807                 if (found_key.objectid == bytenr &&
6808                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6809                         if (level < ref_path->shared_level)
6810                                 ref_path->shared_level = level;
6811                         goto found;
6812                 }
6813 next:
6814                 level--;
6815                 btrfs_release_path(extent_root, path);
6816                 cond_resched();
6817         }
6818         /* reached lowest level */
6819         ret = 1;
6820         goto out;
6821 walk_up:
6822         level = ref_path->current_level;
6823         while (level < BTRFS_MAX_LEVEL - 1) {
6824                 u64 ref_objectid;
6825
6826                 if (level >= 0)
6827                         bytenr = ref_path->nodes[level];
6828                 else
6829                         bytenr = ref_path->extent_start;
6830
6831                 BUG_ON(bytenr == 0);
6832
6833                 key.objectid = bytenr;
6834                 key.offset = 0;
6835                 key.type = BTRFS_EXTENT_REF_KEY;
6836
6837                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6838                 if (ret < 0)
6839                         goto out;
6840
6841                 leaf = path->nodes[0];
6842                 nritems = btrfs_header_nritems(leaf);
6843                 if (path->slots[0] >= nritems) {
6844                         ret = btrfs_next_leaf(extent_root, path);
6845                         if (ret < 0)
6846                                 goto out;
6847                         if (ret > 0) {
6848                                 /* the extent was freed by someone */
6849                                 if (ref_path->lowest_level == level)
6850                                         goto out;
6851                                 btrfs_release_path(extent_root, path);
6852                                 goto walk_down;
6853                         }
6854                         leaf = path->nodes[0];
6855                 }
6856
6857                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6858                 if (found_key.objectid != bytenr ||
6859                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6860                         /* the extent was freed by someone */
6861                         if (ref_path->lowest_level == level) {
6862                                 ret = 1;
6863                                 goto out;
6864                         }
6865                         btrfs_release_path(extent_root, path);
6866                         goto walk_down;
6867                 }
6868 found:
6869                 ref = btrfs_item_ptr(leaf, path->slots[0],
6870                                 struct btrfs_extent_ref);
6871                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6872                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6873                         if (first_time) {
6874                                 level = (int)ref_objectid;
6875                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6876                                 ref_path->lowest_level = level;
6877                                 ref_path->current_level = level;
6878                                 ref_path->nodes[level] = bytenr;
6879                         } else {
6880                                 WARN_ON(ref_objectid != level);
6881                         }
6882                 } else {
6883                         WARN_ON(level != -1);
6884                 }
6885                 first_time = 0;
6886
6887                 if (ref_path->lowest_level == level) {
6888                         ref_path->owner_objectid = ref_objectid;
6889                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6890                 }
6891
6892                 /*
6893                  * the block is tree root or the block isn't in reference
6894                  * counted tree.
6895                  */
6896                 if (found_key.objectid == found_key.offset ||
6897                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6898                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6899                         ref_path->root_generation =
6900                                 btrfs_ref_generation(leaf, ref);
6901                         if (level < 0) {
6902                                 /* special reference from the tree log */
6903                                 ref_path->nodes[0] = found_key.offset;
6904                                 ref_path->current_level = 0;
6905                         }
6906                         ret = 0;
6907                         goto out;
6908                 }
6909
6910                 level++;
6911                 BUG_ON(ref_path->nodes[level] != 0);
6912                 ref_path->nodes[level] = found_key.offset;
6913                 ref_path->current_level = level;
6914
6915                 /*
6916                  * the reference was created in the running transaction,
6917                  * no need to continue walking up.
6918                  */
6919                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6920                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6921                         ref_path->root_generation =
6922                                 btrfs_ref_generation(leaf, ref);
6923                         ret = 0;
6924                         goto out;
6925                 }
6926
6927                 btrfs_release_path(extent_root, path);
6928                 cond_resched();
6929         }
6930         /* reached max tree level, but no tree root found. */
6931         BUG();
6932 out:
6933         btrfs_free_path(path);
6934         return ret;
6935 }
6936
6937 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6938                                 struct btrfs_root *extent_root,
6939                                 struct btrfs_ref_path *ref_path,
6940                                 u64 extent_start)
6941 {
6942         memset(ref_path, 0, sizeof(*ref_path));
6943         ref_path->extent_start = extent_start;
6944
6945         return __next_ref_path(trans, extent_root, ref_path, 1);
6946 }
6947
6948 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6949                                struct btrfs_root *extent_root,
6950                                struct btrfs_ref_path *ref_path)
6951 {
6952         return __next_ref_path(trans, extent_root, ref_path, 0);
6953 }
6954
6955 static noinline int get_new_locations(struct inode *reloc_inode,
6956                                       struct btrfs_key *extent_key,
6957                                       u64 offset, int no_fragment,
6958                                       struct disk_extent **extents,
6959                                       int *nr_extents)
6960 {
6961         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6962         struct btrfs_path *path;
6963         struct btrfs_file_extent_item *fi;
6964         struct extent_buffer *leaf;
6965         struct disk_extent *exts = *extents;
6966         struct btrfs_key found_key;
6967         u64 cur_pos;
6968         u64 last_byte;
6969         u32 nritems;
6970         int nr = 0;
6971         int max = *nr_extents;
6972         int ret;
6973
6974         WARN_ON(!no_fragment && *extents);
6975         if (!exts) {
6976                 max = 1;
6977                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6978                 if (!exts)
6979                         return -ENOMEM;
6980         }
6981
6982         path = btrfs_alloc_path();
6983         if (!path) {
6984                 if (exts != *extents)
6985                         kfree(exts);
6986                 return -ENOMEM;
6987         }
6988
6989         cur_pos = extent_key->objectid - offset;
6990         last_byte = extent_key->objectid + extent_key->offset;
6991         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6992                                        cur_pos, 0);
6993         if (ret < 0)
6994                 goto out;
6995         if (ret > 0) {
6996                 ret = -ENOENT;
6997                 goto out;
6998         }
6999
7000         while (1) {
7001                 leaf = path->nodes[0];
7002                 nritems = btrfs_header_nritems(leaf);
7003                 if (path->slots[0] >= nritems) {
7004                         ret = btrfs_next_leaf(root, path);
7005                         if (ret < 0)
7006                                 goto out;
7007                         if (ret > 0)
7008                                 break;
7009                         leaf = path->nodes[0];
7010                 }
7011
7012                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7013                 if (found_key.offset != cur_pos ||
7014                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
7015                     found_key.objectid != reloc_inode->i_ino)
7016                         break;
7017
7018                 fi = btrfs_item_ptr(leaf, path->slots[0],
7019                                     struct btrfs_file_extent_item);
7020                 if (btrfs_file_extent_type(leaf, fi) !=
7021                     BTRFS_FILE_EXTENT_REG ||
7022                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7023                         break;
7024
7025                 if (nr == max) {
7026                         struct disk_extent *old = exts;
7027                         max *= 2;
7028                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
7029                         if (!exts) {
7030                                 ret = -ENOMEM;
7031                                 goto out;
7032                         }
7033                         memcpy(exts, old, sizeof(*exts) * nr);
7034                         if (old != *extents)
7035                                 kfree(old);
7036                 }
7037
7038                 exts[nr].disk_bytenr =
7039                         btrfs_file_extent_disk_bytenr(leaf, fi);
7040                 exts[nr].disk_num_bytes =
7041                         btrfs_file_extent_disk_num_bytes(leaf, fi);
7042                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
7043                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7044                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7045                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
7046                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
7047                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
7048                                                                            fi);
7049                 BUG_ON(exts[nr].offset > 0);
7050                 BUG_ON(exts[nr].compression || exts[nr].encryption);
7051                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
7052
7053                 cur_pos += exts[nr].num_bytes;
7054                 nr++;
7055
7056                 if (cur_pos + offset >= last_byte)
7057                         break;
7058
7059                 if (no_fragment) {
7060                         ret = 1;
7061                         goto out;
7062                 }
7063                 path->slots[0]++;
7064         }
7065
7066         BUG_ON(cur_pos + offset > last_byte);
7067         if (cur_pos + offset < last_byte) {
7068                 ret = -ENOENT;
7069                 goto out;
7070         }
7071         ret = 0;
7072 out:
7073         btrfs_free_path(path);
7074         if (ret) {
7075                 if (exts != *extents)
7076                         kfree(exts);
7077         } else {
7078                 *extents = exts;
7079                 *nr_extents = nr;
7080         }
7081         return ret;
7082 }
7083
7084 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
7085                                         struct btrfs_root *root,
7086                                         struct btrfs_path *path,
7087                                         struct btrfs_key *extent_key,
7088                                         struct btrfs_key *leaf_key,
7089                                         struct btrfs_ref_path *ref_path,
7090                                         struct disk_extent *new_extents,
7091                                         int nr_extents)
7092 {
7093         struct extent_buffer *leaf;
7094         struct btrfs_file_extent_item *fi;
7095         struct inode *inode = NULL;
7096         struct btrfs_key key;
7097         u64 lock_start = 0;
7098         u64 lock_end = 0;
7099         u64 num_bytes;
7100         u64 ext_offset;
7101         u64 search_end = (u64)-1;
7102         u32 nritems;
7103         int nr_scaned = 0;
7104         int extent_locked = 0;
7105         int extent_type;
7106         int ret;
7107
7108         memcpy(&key, leaf_key, sizeof(key));
7109         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7110                 if (key.objectid < ref_path->owner_objectid ||
7111                     (key.objectid == ref_path->owner_objectid &&
7112                      key.type < BTRFS_EXTENT_DATA_KEY)) {
7113                         key.objectid = ref_path->owner_objectid;
7114                         key.type = BTRFS_EXTENT_DATA_KEY;
7115                         key.offset = 0;
7116                 }
7117         }
7118
7119         while (1) {
7120                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7121                 if (ret < 0)
7122                         goto out;
7123
7124                 leaf = path->nodes[0];
7125                 nritems = btrfs_header_nritems(leaf);
7126 next:
7127                 if (extent_locked && ret > 0) {
7128                         /*
7129                          * the file extent item was modified by someone
7130                          * before the extent got locked.
7131                          */
7132                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7133                                       lock_end, GFP_NOFS);
7134                         extent_locked = 0;
7135                 }
7136
7137                 if (path->slots[0] >= nritems) {
7138                         if (++nr_scaned > 2)
7139                                 break;
7140
7141                         BUG_ON(extent_locked);
7142                         ret = btrfs_next_leaf(root, path);
7143                         if (ret < 0)
7144                                 goto out;
7145                         if (ret > 0)
7146                                 break;
7147                         leaf = path->nodes[0];
7148                         nritems = btrfs_header_nritems(leaf);
7149                 }
7150
7151                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7152
7153                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7154                         if ((key.objectid > ref_path->owner_objectid) ||
7155                             (key.objectid == ref_path->owner_objectid &&
7156                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7157                             key.offset >= search_end)
7158                                 break;
7159                 }
7160
7161                 if (inode && key.objectid != inode->i_ino) {
7162                         BUG_ON(extent_locked);
7163                         btrfs_release_path(root, path);
7164                         mutex_unlock(&inode->i_mutex);
7165                         iput(inode);
7166                         inode = NULL;
7167                         continue;
7168                 }
7169
7170                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7171                         path->slots[0]++;
7172                         ret = 1;
7173                         goto next;
7174                 }
7175                 fi = btrfs_item_ptr(leaf, path->slots[0],
7176                                     struct btrfs_file_extent_item);
7177                 extent_type = btrfs_file_extent_type(leaf, fi);
7178                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7179                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7180                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7181                      extent_key->objectid)) {
7182                         path->slots[0]++;
7183                         ret = 1;
7184                         goto next;
7185                 }
7186
7187                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7188                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7189
7190                 if (search_end == (u64)-1) {
7191                         search_end = key.offset - ext_offset +
7192                                 btrfs_file_extent_ram_bytes(leaf, fi);
7193                 }
7194
7195                 if (!extent_locked) {
7196                         lock_start = key.offset;
7197                         lock_end = lock_start + num_bytes - 1;
7198                 } else {
7199                         if (lock_start > key.offset ||
7200                             lock_end + 1 < key.offset + num_bytes) {
7201                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7202                                               lock_start, lock_end, GFP_NOFS);
7203                                 extent_locked = 0;
7204                         }
7205                 }
7206
7207                 if (!inode) {
7208                         btrfs_release_path(root, path);
7209
7210                         inode = btrfs_iget_locked(root->fs_info->sb,
7211                                                   key.objectid, root);
7212                         if (inode->i_state & I_NEW) {
7213                                 BTRFS_I(inode)->root = root;
7214                                 BTRFS_I(inode)->location.objectid =
7215                                         key.objectid;
7216                                 BTRFS_I(inode)->location.type =
7217                                         BTRFS_INODE_ITEM_KEY;
7218                                 BTRFS_I(inode)->location.offset = 0;
7219                                 btrfs_read_locked_inode(inode);
7220                                 unlock_new_inode(inode);
7221                         }
7222                         /*
7223                          * some code call btrfs_commit_transaction while
7224                          * holding the i_mutex, so we can't use mutex_lock
7225                          * here.
7226                          */
7227                         if (is_bad_inode(inode) ||
7228                             !mutex_trylock(&inode->i_mutex)) {
7229                                 iput(inode);
7230                                 inode = NULL;
7231                                 key.offset = (u64)-1;
7232                                 goto skip;
7233                         }
7234                 }
7235
7236                 if (!extent_locked) {
7237                         struct btrfs_ordered_extent *ordered;
7238
7239                         btrfs_release_path(root, path);
7240
7241                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7242                                     lock_end, GFP_NOFS);
7243                         ordered = btrfs_lookup_first_ordered_extent(inode,
7244                                                                     lock_end);
7245                         if (ordered &&
7246                             ordered->file_offset <= lock_end &&
7247                             ordered->file_offset + ordered->len > lock_start) {
7248                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7249                                               lock_start, lock_end, GFP_NOFS);
7250                                 btrfs_start_ordered_extent(inode, ordered, 1);
7251                                 btrfs_put_ordered_extent(ordered);
7252                                 key.offset += num_bytes;
7253                                 goto skip;
7254                         }
7255                         if (ordered)
7256                                 btrfs_put_ordered_extent(ordered);
7257
7258                         extent_locked = 1;
7259                         continue;
7260                 }
7261
7262                 if (nr_extents == 1) {
7263                         /* update extent pointer in place */
7264                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7265                                                 new_extents[0].disk_bytenr);
7266                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7267                                                 new_extents[0].disk_num_bytes);
7268                         btrfs_mark_buffer_dirty(leaf);
7269
7270                         btrfs_drop_extent_cache(inode, key.offset,
7271                                                 key.offset + num_bytes - 1, 0);
7272
7273                         ret = btrfs_inc_extent_ref(trans, root,
7274                                                 new_extents[0].disk_bytenr,
7275                                                 new_extents[0].disk_num_bytes,
7276                                                 leaf->start,
7277                                                 root->root_key.objectid,
7278                                                 trans->transid,
7279                                                 key.objectid);
7280                         BUG_ON(ret);
7281
7282                         ret = btrfs_free_extent(trans, root,
7283                                                 extent_key->objectid,
7284                                                 extent_key->offset,
7285                                                 leaf->start,
7286                                                 btrfs_header_owner(leaf),
7287                                                 btrfs_header_generation(leaf),
7288                                                 key.objectid, 0);
7289                         BUG_ON(ret);
7290
7291                         btrfs_release_path(root, path);
7292                         key.offset += num_bytes;
7293                 } else {
7294                         BUG_ON(1);
7295 #if 0
7296                         u64 alloc_hint;
7297                         u64 extent_len;
7298                         int i;
7299                         /*
7300                          * drop old extent pointer at first, then insert the
7301                          * new pointers one bye one
7302                          */
7303                         btrfs_release_path(root, path);
7304                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7305                                                  key.offset + num_bytes,
7306                                                  key.offset, &alloc_hint);
7307                         BUG_ON(ret);
7308
7309                         for (i = 0; i < nr_extents; i++) {
7310                                 if (ext_offset >= new_extents[i].num_bytes) {
7311                                         ext_offset -= new_extents[i].num_bytes;
7312                                         continue;
7313                                 }
7314                                 extent_len = min(new_extents[i].num_bytes -
7315                                                  ext_offset, num_bytes);
7316
7317                                 ret = btrfs_insert_empty_item(trans, root,
7318                                                               path, &key,
7319                                                               sizeof(*fi));
7320                                 BUG_ON(ret);
7321
7322                                 leaf = path->nodes[0];
7323                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7324                                                 struct btrfs_file_extent_item);
7325                                 btrfs_set_file_extent_generation(leaf, fi,
7326                                                         trans->transid);
7327                                 btrfs_set_file_extent_type(leaf, fi,
7328                                                         BTRFS_FILE_EXTENT_REG);
7329                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7330                                                 new_extents[i].disk_bytenr);
7331                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7332                                                 new_extents[i].disk_num_bytes);
7333                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7334                                                 new_extents[i].ram_bytes);
7335
7336                                 btrfs_set_file_extent_compression(leaf, fi,
7337                                                 new_extents[i].compression);
7338                                 btrfs_set_file_extent_encryption(leaf, fi,
7339                                                 new_extents[i].encryption);
7340                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7341                                                 new_extents[i].other_encoding);
7342
7343                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7344                                                         extent_len);
7345                                 ext_offset += new_extents[i].offset;
7346                                 btrfs_set_file_extent_offset(leaf, fi,
7347                                                         ext_offset);
7348                                 btrfs_mark_buffer_dirty(leaf);
7349
7350                                 btrfs_drop_extent_cache(inode, key.offset,
7351                                                 key.offset + extent_len - 1, 0);
7352
7353                                 ret = btrfs_inc_extent_ref(trans, root,
7354                                                 new_extents[i].disk_bytenr,
7355                                                 new_extents[i].disk_num_bytes,
7356                                                 leaf->start,
7357                                                 root->root_key.objectid,
7358                                                 trans->transid, key.objectid);
7359                                 BUG_ON(ret);
7360                                 btrfs_release_path(root, path);
7361
7362                                 inode_add_bytes(inode, extent_len);
7363
7364                                 ext_offset = 0;
7365                                 num_bytes -= extent_len;
7366                                 key.offset += extent_len;
7367
7368                                 if (num_bytes == 0)
7369                                         break;
7370                         }
7371                         BUG_ON(i >= nr_extents);
7372 #endif
7373                 }
7374
7375                 if (extent_locked) {
7376                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7377                                       lock_end, GFP_NOFS);
7378                         extent_locked = 0;
7379                 }
7380 skip:
7381                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7382                     key.offset >= search_end)
7383                         break;
7384
7385                 cond_resched();
7386         }
7387         ret = 0;
7388 out:
7389         btrfs_release_path(root, path);
7390         if (inode) {
7391                 mutex_unlock(&inode->i_mutex);
7392                 if (extent_locked) {
7393                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7394                                       lock_end, GFP_NOFS);
7395                 }
7396                 iput(inode);
7397         }
7398         return ret;
7399 }
7400
7401 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7402                                struct btrfs_root *root,
7403                                struct extent_buffer *buf, u64 orig_start)
7404 {
7405         int level;
7406         int ret;
7407
7408         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7409         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7410
7411         level = btrfs_header_level(buf);
7412         if (level == 0) {
7413                 struct btrfs_leaf_ref *ref;
7414                 struct btrfs_leaf_ref *orig_ref;
7415
7416                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7417                 if (!orig_ref)
7418                         return -ENOENT;
7419
7420                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7421                 if (!ref) {
7422                         btrfs_free_leaf_ref(root, orig_ref);
7423                         return -ENOMEM;
7424                 }
7425
7426                 ref->nritems = orig_ref->nritems;
7427                 memcpy(ref->extents, orig_ref->extents,
7428                         sizeof(ref->extents[0]) * ref->nritems);
7429
7430                 btrfs_free_leaf_ref(root, orig_ref);
7431
7432                 ref->root_gen = trans->transid;
7433                 ref->bytenr = buf->start;
7434                 ref->owner = btrfs_header_owner(buf);
7435                 ref->generation = btrfs_header_generation(buf);
7436
7437                 ret = btrfs_add_leaf_ref(root, ref, 0);
7438                 WARN_ON(ret);
7439                 btrfs_free_leaf_ref(root, ref);
7440         }
7441         return 0;
7442 }
7443
7444 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7445                                         struct extent_buffer *leaf,
7446                                         struct btrfs_block_group_cache *group,
7447                                         struct btrfs_root *target_root)
7448 {
7449         struct btrfs_key key;
7450         struct inode *inode = NULL;
7451         struct btrfs_file_extent_item *fi;
7452         struct extent_state *cached_state = NULL;
7453         u64 num_bytes;
7454         u64 skip_objectid = 0;
7455         u32 nritems;
7456         u32 i;
7457
7458         nritems = btrfs_header_nritems(leaf);
7459         for (i = 0; i < nritems; i++) {
7460                 btrfs_item_key_to_cpu(leaf, &key, i);
7461                 if (key.objectid == skip_objectid ||
7462                     key.type != BTRFS_EXTENT_DATA_KEY)
7463                         continue;
7464                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7465                 if (btrfs_file_extent_type(leaf, fi) ==
7466                     BTRFS_FILE_EXTENT_INLINE)
7467                         continue;
7468                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7469                         continue;
7470                 if (!inode || inode->i_ino != key.objectid) {
7471                         iput(inode);
7472                         inode = btrfs_ilookup(target_root->fs_info->sb,
7473                                               key.objectid, target_root, 1);
7474                 }
7475                 if (!inode) {
7476                         skip_objectid = key.objectid;
7477                         continue;
7478                 }
7479                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7480
7481                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7482                                  key.offset + num_bytes - 1, 0, &cached_state,
7483                                  GFP_NOFS);
7484                 btrfs_drop_extent_cache(inode, key.offset,
7485                                         key.offset + num_bytes - 1, 1);
7486                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7487                                      key.offset + num_bytes - 1, &cached_state,
7488                                      GFP_NOFS);
7489                 cond_resched();
7490         }
7491         iput(inode);
7492         return 0;
7493 }
7494
7495 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7496                                         struct btrfs_root *root,
7497                                         struct extent_buffer *leaf,
7498                                         struct btrfs_block_group_cache *group,
7499                                         struct inode *reloc_inode)
7500 {
7501         struct btrfs_key key;
7502         struct btrfs_key extent_key;
7503         struct btrfs_file_extent_item *fi;
7504         struct btrfs_leaf_ref *ref;
7505         struct disk_extent *new_extent;
7506         u64 bytenr;
7507         u64 num_bytes;
7508         u32 nritems;
7509         u32 i;
7510         int ext_index;
7511         int nr_extent;
7512         int ret;
7513
7514         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7515         if (!new_extent)
7516                 return -ENOMEM;
7517
7518         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7519         BUG_ON(!ref);
7520
7521         ext_index = -1;
7522         nritems = btrfs_header_nritems(leaf);
7523         for (i = 0; i < nritems; i++) {
7524                 btrfs_item_key_to_cpu(leaf, &key, i);
7525                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7526                         continue;
7527                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7528                 if (btrfs_file_extent_type(leaf, fi) ==
7529                     BTRFS_FILE_EXTENT_INLINE)
7530                         continue;
7531                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7532                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7533                 if (bytenr == 0)
7534                         continue;
7535
7536                 ext_index++;
7537                 if (bytenr >= group->key.objectid + group->key.offset ||
7538                     bytenr + num_bytes <= group->key.objectid)
7539                         continue;
7540
7541                 extent_key.objectid = bytenr;
7542                 extent_key.offset = num_bytes;
7543                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7544                 nr_extent = 1;
7545                 ret = get_new_locations(reloc_inode, &extent_key,
7546                                         group->key.objectid, 1,
7547                                         &new_extent, &nr_extent);
7548                 if (ret > 0)
7549                         continue;
7550                 BUG_ON(ret < 0);
7551
7552                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7553                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7554                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7555                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7556
7557                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7558                                                 new_extent->disk_bytenr);
7559                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7560                                                 new_extent->disk_num_bytes);
7561                 btrfs_mark_buffer_dirty(leaf);
7562
7563                 ret = btrfs_inc_extent_ref(trans, root,
7564                                         new_extent->disk_bytenr,
7565                                         new_extent->disk_num_bytes,
7566                                         leaf->start,
7567                                         root->root_key.objectid,
7568                                         trans->transid, key.objectid);
7569                 BUG_ON(ret);
7570
7571                 ret = btrfs_free_extent(trans, root,
7572                                         bytenr, num_bytes, leaf->start,
7573                                         btrfs_header_owner(leaf),
7574                                         btrfs_header_generation(leaf),
7575                                         key.objectid, 0);
7576                 BUG_ON(ret);
7577                 cond_resched();
7578         }
7579         kfree(new_extent);
7580         BUG_ON(ext_index + 1 != ref->nritems);
7581         btrfs_free_leaf_ref(root, ref);
7582         return 0;
7583 }
7584
7585 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7586                           struct btrfs_root *root)
7587 {
7588         struct btrfs_root *reloc_root;
7589         int ret;
7590
7591         if (root->reloc_root) {
7592                 reloc_root = root->reloc_root;
7593                 root->reloc_root = NULL;
7594                 list_add(&reloc_root->dead_list,
7595                          &root->fs_info->dead_reloc_roots);
7596
7597                 btrfs_set_root_bytenr(&reloc_root->root_item,
7598                                       reloc_root->node->start);
7599                 btrfs_set_root_level(&root->root_item,
7600                                      btrfs_header_level(reloc_root->node));
7601                 memset(&reloc_root->root_item.drop_progress, 0,
7602                         sizeof(struct btrfs_disk_key));
7603                 reloc_root->root_item.drop_level = 0;
7604
7605                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7606                                         &reloc_root->root_key,
7607                                         &reloc_root->root_item);
7608                 BUG_ON(ret);
7609         }
7610         return 0;
7611 }
7612
7613 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7614 {
7615         struct btrfs_trans_handle *trans;
7616         struct btrfs_root *reloc_root;
7617         struct btrfs_root *prev_root = NULL;
7618         struct list_head dead_roots;
7619         int ret;
7620         unsigned long nr;
7621
7622         INIT_LIST_HEAD(&dead_roots);
7623         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7624
7625         while (!list_empty(&dead_roots)) {
7626                 reloc_root = list_entry(dead_roots.prev,
7627                                         struct btrfs_root, dead_list);
7628                 list_del_init(&reloc_root->dead_list);
7629
7630                 BUG_ON(reloc_root->commit_root != NULL);
7631                 while (1) {
7632                         trans = btrfs_join_transaction(root, 1);
7633                         BUG_ON(IS_ERR(trans));
7634
7635                         mutex_lock(&root->fs_info->drop_mutex);
7636                         ret = btrfs_drop_snapshot(trans, reloc_root);
7637                         if (ret != -EAGAIN)
7638                                 break;
7639                         mutex_unlock(&root->fs_info->drop_mutex);
7640
7641                         nr = trans->blocks_used;
7642                         ret = btrfs_end_transaction(trans, root);
7643                         BUG_ON(ret);
7644                         btrfs_btree_balance_dirty(root, nr);
7645                 }
7646
7647                 free_extent_buffer(reloc_root->node);
7648
7649                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7650                                      &reloc_root->root_key);
7651                 BUG_ON(ret);
7652                 mutex_unlock(&root->fs_info->drop_mutex);
7653
7654                 nr = trans->blocks_used;
7655                 ret = btrfs_end_transaction(trans, root);
7656                 BUG_ON(ret);
7657                 btrfs_btree_balance_dirty(root, nr);
7658
7659                 kfree(prev_root);
7660                 prev_root = reloc_root;
7661         }
7662         if (prev_root) {
7663                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7664                 kfree(prev_root);
7665         }
7666         return 0;
7667 }
7668
7669 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7670 {
7671         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7672         return 0;
7673 }
7674
7675 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7676 {
7677         struct btrfs_root *reloc_root;
7678         struct btrfs_trans_handle *trans;
7679         struct btrfs_key location;
7680         int found;
7681         int ret;
7682
7683         mutex_lock(&root->fs_info->tree_reloc_mutex);
7684         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7685         BUG_ON(ret);
7686         found = !list_empty(&root->fs_info->dead_reloc_roots);
7687         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7688
7689         if (found) {
7690                 trans = btrfs_start_transaction(root, 1);
7691                 BUG_ON(IS_ERR(trans));
7692                 ret = btrfs_commit_transaction(trans, root);
7693                 BUG_ON(ret);
7694         }
7695
7696         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7697         location.offset = (u64)-1;
7698         location.type = BTRFS_ROOT_ITEM_KEY;
7699
7700         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7701         BUG_ON(!reloc_root);
7702         ret = btrfs_orphan_cleanup(reloc_root);
7703         BUG_ON(ret);
7704         return 0;
7705 }
7706
7707 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7708                                     struct btrfs_root *root)
7709 {
7710         struct btrfs_root *reloc_root;
7711         struct extent_buffer *eb;
7712         struct btrfs_root_item *root_item;
7713         struct btrfs_key root_key;
7714         int ret;
7715
7716         BUG_ON(!root->ref_cows);
7717         if (root->reloc_root)
7718                 return 0;
7719
7720         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7721         if (!root_item)
7722                 return -ENOMEM;
7723
7724         ret = btrfs_copy_root(trans, root, root->commit_root,
7725                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7726         BUG_ON(ret);
7727
7728         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7729         root_key.offset = root->root_key.objectid;
7730         root_key.type = BTRFS_ROOT_ITEM_KEY;
7731
7732         memcpy(root_item, &root->root_item, sizeof(root_item));
7733         btrfs_set_root_refs(root_item, 0);
7734         btrfs_set_root_bytenr(root_item, eb->start);
7735         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7736         btrfs_set_root_generation(root_item, trans->transid);
7737
7738         btrfs_tree_unlock(eb);
7739         free_extent_buffer(eb);
7740
7741         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7742                                 &root_key, root_item);
7743         BUG_ON(ret);
7744         kfree(root_item);
7745
7746         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7747                                                  &root_key);
7748         BUG_ON(IS_ERR(reloc_root));
7749         reloc_root->last_trans = trans->transid;
7750         reloc_root->commit_root = NULL;
7751         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7752
7753         root->reloc_root = reloc_root;
7754         return 0;
7755 }
7756
7757 /*
7758  * Core function of space balance.
7759  *
7760  * The idea is using reloc trees to relocate tree blocks in reference
7761  * counted roots. There is one reloc tree for each subvol, and all
7762  * reloc trees share same root key objectid. Reloc trees are snapshots
7763  * of the latest committed roots of subvols (root->commit_root).
7764  *
7765  * To relocate a tree block referenced by a subvol, there are two steps.
7766  * COW the block through subvol's reloc tree, then update block pointer
7767  * in the subvol to point to the new block. Since all reloc trees share
7768  * same root key objectid, doing special handing for tree blocks owned
7769  * by them is easy. Once a tree block has been COWed in one reloc tree,
7770  * we can use the resulting new block directly when the same block is
7771  * required to COW again through other reloc trees. By this way, relocated
7772  * tree blocks are shared between reloc trees, so they are also shared
7773  * between subvols.
7774  */
7775 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7776                                       struct btrfs_root *root,
7777                                       struct btrfs_path *path,
7778                                       struct btrfs_key *first_key,
7779                                       struct btrfs_ref_path *ref_path,
7780                                       struct btrfs_block_group_cache *group,
7781                                       struct inode *reloc_inode)
7782 {
7783         struct btrfs_root *reloc_root;
7784         struct extent_buffer *eb = NULL;
7785         struct btrfs_key *keys;
7786         u64 *nodes;
7787         int level;
7788         int shared_level;
7789         int lowest_level = 0;
7790         int ret;
7791
7792         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7793                 lowest_level = ref_path->owner_objectid;
7794
7795         if (!root->ref_cows) {
7796                 path->lowest_level = lowest_level;
7797                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7798                 BUG_ON(ret < 0);
7799                 path->lowest_level = 0;
7800                 btrfs_release_path(root, path);
7801                 return 0;
7802         }
7803
7804         mutex_lock(&root->fs_info->tree_reloc_mutex);
7805         ret = init_reloc_tree(trans, root);
7806         BUG_ON(ret);
7807         reloc_root = root->reloc_root;
7808
7809         shared_level = ref_path->shared_level;
7810         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7811
7812         keys = ref_path->node_keys;
7813         nodes = ref_path->new_nodes;
7814         memset(&keys[shared_level + 1], 0,
7815                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7816         memset(&nodes[shared_level + 1], 0,
7817                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7818
7819         if (nodes[lowest_level] == 0) {
7820                 path->lowest_level = lowest_level;
7821                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7822                                         0, 1);
7823                 BUG_ON(ret);
7824                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7825                         eb = path->nodes[level];
7826                         if (!eb || eb == reloc_root->node)
7827                                 break;
7828                         nodes[level] = eb->start;
7829                         if (level == 0)
7830                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7831                         else
7832                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7833                 }
7834                 if (nodes[0] &&
7835                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7836                         eb = path->nodes[0];
7837                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7838                                                       group, reloc_inode);
7839                         BUG_ON(ret);
7840                 }
7841                 btrfs_release_path(reloc_root, path);
7842         } else {
7843                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7844                                        lowest_level);
7845                 BUG_ON(ret);
7846         }
7847
7848         /*
7849          * replace tree blocks in the fs tree with tree blocks in
7850          * the reloc tree.
7851          */
7852         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7853         BUG_ON(ret < 0);
7854
7855         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7856                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7857                                         0, 0);
7858                 BUG_ON(ret);
7859                 extent_buffer_get(path->nodes[0]);
7860                 eb = path->nodes[0];
7861                 btrfs_release_path(reloc_root, path);
7862                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7863                 BUG_ON(ret);
7864                 free_extent_buffer(eb);
7865         }
7866
7867         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7868         path->lowest_level = 0;
7869         return 0;
7870 }
7871
7872 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7873                                         struct btrfs_root *root,
7874                                         struct btrfs_path *path,
7875                                         struct btrfs_key *first_key,
7876                                         struct btrfs_ref_path *ref_path)
7877 {
7878         int ret;
7879
7880         ret = relocate_one_path(trans, root, path, first_key,
7881                                 ref_path, NULL, NULL);
7882         BUG_ON(ret);
7883
7884         return 0;
7885 }
7886
7887 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7888                                     struct btrfs_root *extent_root,
7889                                     struct btrfs_path *path,
7890                                     struct btrfs_key *extent_key)
7891 {
7892         int ret;
7893
7894         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7895         if (ret)
7896                 goto out;
7897         ret = btrfs_del_item(trans, extent_root, path);
7898 out:
7899         btrfs_release_path(extent_root, path);
7900         return ret;
7901 }
7902
7903 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7904                                                 struct btrfs_ref_path *ref_path)
7905 {
7906         struct btrfs_key root_key;
7907
7908         root_key.objectid = ref_path->root_objectid;
7909         root_key.type = BTRFS_ROOT_ITEM_KEY;
7910         if (is_cowonly_root(ref_path->root_objectid))
7911                 root_key.offset = 0;
7912         else
7913                 root_key.offset = (u64)-1;
7914
7915         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7916 }
7917
7918 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7919                                         struct btrfs_path *path,
7920                                         struct btrfs_key *extent_key,
7921                                         struct btrfs_block_group_cache *group,
7922                                         struct inode *reloc_inode, int pass)
7923 {
7924         struct btrfs_trans_handle *trans;
7925         struct btrfs_root *found_root;
7926         struct btrfs_ref_path *ref_path = NULL;
7927         struct disk_extent *new_extents = NULL;
7928         int nr_extents = 0;
7929         int loops;
7930         int ret;
7931         int level;
7932         struct btrfs_key first_key;
7933         u64 prev_block = 0;
7934
7935
7936         trans = btrfs_start_transaction(extent_root, 1);
7937         BUG_ON(IS_ERR(trans));
7938
7939         if (extent_key->objectid == 0) {
7940                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7941                 goto out;
7942         }
7943
7944         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7945         if (!ref_path) {
7946                 ret = -ENOMEM;
7947                 goto out;
7948         }
7949
7950         for (loops = 0; ; loops++) {
7951                 if (loops == 0) {
7952                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7953                                                    extent_key->objectid);
7954                 } else {
7955                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7956                 }
7957                 if (ret < 0)
7958                         goto out;
7959                 if (ret > 0)
7960                         break;
7961
7962                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7963                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7964                         continue;
7965
7966                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7967                 BUG_ON(!found_root);
7968                 /*
7969                  * for reference counted tree, only process reference paths
7970                  * rooted at the latest committed root.
7971                  */
7972                 if (found_root->ref_cows &&
7973                     ref_path->root_generation != found_root->root_key.offset)
7974                         continue;
7975
7976                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7977                         if (pass == 0) {
7978                                 /*
7979                                  * copy data extents to new locations
7980                                  */
7981                                 u64 group_start = group->key.objectid;
7982                                 ret = relocate_data_extent(reloc_inode,
7983                                                            extent_key,
7984                                                            group_start);
7985                                 if (ret < 0)
7986                                         goto out;
7987                                 break;
7988                         }
7989                         level = 0;
7990                 } else {
7991                         level = ref_path->owner_objectid;
7992                 }
7993
7994                 if (prev_block != ref_path->nodes[level]) {
7995                         struct extent_buffer *eb;
7996                         u64 block_start = ref_path->nodes[level];
7997                         u64 block_size = btrfs_level_size(found_root, level);
7998
7999                         eb = read_tree_block(found_root, block_start,
8000                                              block_size, 0);
8001                         if (!eb) {
8002                                 ret = -EIO;
8003                                 goto out;
8004                         }
8005                         btrfs_tree_lock(eb);
8006                         BUG_ON(level != btrfs_header_level(eb));
8007
8008                         if (level == 0)
8009                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
8010                         else
8011                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
8012
8013                         btrfs_tree_unlock(eb);
8014                         free_extent_buffer(eb);
8015                         prev_block = block_start;
8016                 }
8017
8018                 mutex_lock(&extent_root->fs_info->trans_mutex);
8019                 btrfs_record_root_in_trans(found_root);
8020                 mutex_unlock(&extent_root->fs_info->trans_mutex);
8021                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8022                         /*
8023                          * try to update data extent references while
8024                          * keeping metadata shared between snapshots.
8025                          */
8026                         if (pass == 1) {
8027                                 ret = relocate_one_path(trans, found_root,
8028                                                 path, &first_key, ref_path,
8029                                                 group, reloc_inode);
8030                                 if (ret < 0)
8031                                         goto out;
8032                                 continue;
8033                         }
8034                         /*
8035                          * use fallback method to process the remaining
8036                          * references.
8037                          */
8038                         if (!new_extents) {
8039                                 u64 group_start = group->key.objectid;
8040                                 new_extents = kmalloc(sizeof(*new_extents),
8041                                                       GFP_NOFS);
8042                                 nr_extents = 1;
8043                                 ret = get_new_locations(reloc_inode,
8044                                                         extent_key,
8045                                                         group_start, 1,
8046                                                         &new_extents,
8047                                                         &nr_extents);
8048                                 if (ret)
8049                                         goto out;
8050                         }
8051                         ret = replace_one_extent(trans, found_root,
8052                                                 path, extent_key,
8053                                                 &first_key, ref_path,
8054                                                 new_extents, nr_extents);
8055                 } else {
8056                         ret = relocate_tree_block(trans, found_root, path,
8057                                                   &first_key, ref_path);
8058                 }
8059                 if (ret < 0)
8060                         goto out;
8061         }
8062         ret = 0;
8063 out:
8064         btrfs_end_transaction(trans, extent_root);
8065         kfree(new_extents);
8066         kfree(ref_path);
8067         return ret;
8068 }
8069 #endif
8070
8071 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8072 {
8073         u64 num_devices;
8074         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8075                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8076
8077         /*
8078          * we add in the count of missing devices because we want
8079          * to make sure that any RAID levels on a degraded FS
8080          * continue to be honored.
8081          */
8082         num_devices = root->fs_info->fs_devices->rw_devices +
8083                 root->fs_info->fs_devices->missing_devices;
8084
8085         if (num_devices == 1) {
8086                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8087                 stripped = flags & ~stripped;
8088
8089                 /* turn raid0 into single device chunks */
8090                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8091                         return stripped;
8092
8093                 /* turn mirroring into duplication */
8094                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8095                              BTRFS_BLOCK_GROUP_RAID10))
8096                         return stripped | BTRFS_BLOCK_GROUP_DUP;
8097                 return flags;
8098         } else {
8099                 /* they already had raid on here, just return */
8100                 if (flags & stripped)
8101                         return flags;
8102
8103                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8104                 stripped = flags & ~stripped;
8105
8106                 /* switch duplicated blocks with raid1 */
8107                 if (flags & BTRFS_BLOCK_GROUP_DUP)
8108                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
8109
8110                 /* turn single device chunks into raid0 */
8111                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8112         }
8113         return flags;
8114 }
8115
8116 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8117 {
8118         struct btrfs_space_info *sinfo = cache->space_info;
8119         u64 num_bytes;
8120         int ret = -ENOSPC;
8121
8122         if (cache->ro)
8123                 return 0;
8124
8125         spin_lock(&sinfo->lock);
8126         spin_lock(&cache->lock);
8127         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8128                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8129
8130         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8131             sinfo->bytes_may_use + sinfo->bytes_readonly +
8132             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8133                 sinfo->bytes_readonly += num_bytes;
8134                 sinfo->bytes_reserved += cache->reserved_pinned;
8135                 cache->reserved_pinned = 0;
8136                 cache->ro = 1;
8137                 ret = 0;
8138         }
8139
8140         spin_unlock(&cache->lock);
8141         spin_unlock(&sinfo->lock);
8142         return ret;
8143 }
8144
8145 int btrfs_set_block_group_ro(struct btrfs_root *root,
8146                              struct btrfs_block_group_cache *cache)
8147
8148 {
8149         struct btrfs_trans_handle *trans;
8150         u64 alloc_flags;
8151         int ret;
8152
8153         BUG_ON(cache->ro);
8154
8155         trans = btrfs_join_transaction(root, 1);
8156         BUG_ON(IS_ERR(trans));
8157
8158         alloc_flags = update_block_group_flags(root, cache->flags);
8159         if (alloc_flags != cache->flags)
8160                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8161                                CHUNK_ALLOC_FORCE);
8162
8163         ret = set_block_group_ro(cache);
8164         if (!ret)
8165                 goto out;
8166         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8167         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8168                              CHUNK_ALLOC_FORCE);
8169         if (ret < 0)
8170                 goto out;
8171         ret = set_block_group_ro(cache);
8172 out:
8173         btrfs_end_transaction(trans, root);
8174         return ret;
8175 }
8176
8177 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8178                             struct btrfs_root *root, u64 type)
8179 {
8180         u64 alloc_flags = get_alloc_profile(root, type);
8181         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8182                               CHUNK_ALLOC_FORCE);
8183 }
8184
8185 /*
8186  * helper to account the unused space of all the readonly block group in the
8187  * list. takes mirrors into account.
8188  */
8189 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8190 {
8191         struct btrfs_block_group_cache *block_group;
8192         u64 free_bytes = 0;
8193         int factor;
8194
8195         list_for_each_entry(block_group, groups_list, list) {
8196                 spin_lock(&block_group->lock);
8197
8198                 if (!block_group->ro) {
8199                         spin_unlock(&block_group->lock);
8200                         continue;
8201                 }
8202
8203                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8204                                           BTRFS_BLOCK_GROUP_RAID10 |
8205                                           BTRFS_BLOCK_GROUP_DUP))
8206                         factor = 2;
8207                 else
8208                         factor = 1;
8209
8210                 free_bytes += (block_group->key.offset -
8211                                btrfs_block_group_used(&block_group->item)) *
8212                                factor;
8213
8214                 spin_unlock(&block_group->lock);
8215         }
8216
8217         return free_bytes;
8218 }
8219
8220 /*
8221  * helper to account the unused space of all the readonly block group in the
8222  * space_info. takes mirrors into account.
8223  */
8224 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8225 {
8226         int i;
8227         u64 free_bytes = 0;
8228
8229         spin_lock(&sinfo->lock);
8230
8231         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8232                 if (!list_empty(&sinfo->block_groups[i]))
8233                         free_bytes += __btrfs_get_ro_block_group_free_space(
8234                                                 &sinfo->block_groups[i]);
8235
8236         spin_unlock(&sinfo->lock);
8237
8238         return free_bytes;
8239 }
8240
8241 int btrfs_set_block_group_rw(struct btrfs_root *root,
8242                               struct btrfs_block_group_cache *cache)
8243 {
8244         struct btrfs_space_info *sinfo = cache->space_info;
8245         u64 num_bytes;
8246
8247         BUG_ON(!cache->ro);
8248
8249         spin_lock(&sinfo->lock);
8250         spin_lock(&cache->lock);
8251         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8252                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8253         sinfo->bytes_readonly -= num_bytes;
8254         cache->ro = 0;
8255         spin_unlock(&cache->lock);
8256         spin_unlock(&sinfo->lock);
8257         return 0;
8258 }
8259
8260 /*
8261  * checks to see if its even possible to relocate this block group.
8262  *
8263  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8264  * ok to go ahead and try.
8265  */
8266 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8267 {
8268         struct btrfs_block_group_cache *block_group;
8269         struct btrfs_space_info *space_info;
8270         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8271         struct btrfs_device *device;
8272         int full = 0;
8273         int ret = 0;
8274
8275         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8276
8277         /* odd, couldn't find the block group, leave it alone */
8278         if (!block_group)
8279                 return -1;
8280
8281         /* no bytes used, we're good */
8282         if (!btrfs_block_group_used(&block_group->item))
8283                 goto out;
8284
8285         space_info = block_group->space_info;
8286         spin_lock(&space_info->lock);
8287
8288         full = space_info->full;
8289
8290         /*
8291          * if this is the last block group we have in this space, we can't
8292          * relocate it unless we're able to allocate a new chunk below.
8293          *
8294          * Otherwise, we need to make sure we have room in the space to handle
8295          * all of the extents from this block group.  If we can, we're good
8296          */
8297         if ((space_info->total_bytes != block_group->key.offset) &&
8298            (space_info->bytes_used + space_info->bytes_reserved +
8299             space_info->bytes_pinned + space_info->bytes_readonly +
8300             btrfs_block_group_used(&block_group->item) <
8301             space_info->total_bytes)) {
8302                 spin_unlock(&space_info->lock);
8303                 goto out;
8304         }
8305         spin_unlock(&space_info->lock);
8306
8307         /*
8308          * ok we don't have enough space, but maybe we have free space on our
8309          * devices to allocate new chunks for relocation, so loop through our
8310          * alloc devices and guess if we have enough space.  However, if we
8311          * were marked as full, then we know there aren't enough chunks, and we
8312          * can just return.
8313          */
8314         ret = -1;
8315         if (full)
8316                 goto out;
8317
8318         mutex_lock(&root->fs_info->chunk_mutex);
8319         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8320                 u64 min_free = btrfs_block_group_used(&block_group->item);
8321                 u64 dev_offset;
8322
8323                 /*
8324                  * check to make sure we can actually find a chunk with enough
8325                  * space to fit our block group in.
8326                  */
8327                 if (device->total_bytes > device->bytes_used + min_free) {
8328                         ret = find_free_dev_extent(NULL, device, min_free,
8329                                                    &dev_offset, NULL);
8330                         if (!ret)
8331                                 break;
8332                         ret = -1;
8333                 }
8334         }
8335         mutex_unlock(&root->fs_info->chunk_mutex);
8336 out:
8337         btrfs_put_block_group(block_group);
8338         return ret;
8339 }
8340
8341 static int find_first_block_group(struct btrfs_root *root,
8342                 struct btrfs_path *path, struct btrfs_key *key)
8343 {
8344         int ret = 0;
8345         struct btrfs_key found_key;
8346         struct extent_buffer *leaf;
8347         int slot;
8348
8349         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8350         if (ret < 0)
8351                 goto out;
8352
8353         while (1) {
8354                 slot = path->slots[0];
8355                 leaf = path->nodes[0];
8356                 if (slot >= btrfs_header_nritems(leaf)) {
8357                         ret = btrfs_next_leaf(root, path);
8358                         if (ret == 0)
8359                                 continue;
8360                         if (ret < 0)
8361                                 goto out;
8362                         break;
8363                 }
8364                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8365
8366                 if (found_key.objectid >= key->objectid &&
8367                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8368                         ret = 0;
8369                         goto out;
8370                 }
8371                 path->slots[0]++;
8372         }
8373 out:
8374         return ret;
8375 }
8376
8377 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8378 {
8379         struct btrfs_block_group_cache *block_group;
8380         u64 last = 0;
8381
8382         while (1) {
8383                 struct inode *inode;
8384
8385                 block_group = btrfs_lookup_first_block_group(info, last);
8386                 while (block_group) {
8387                         spin_lock(&block_group->lock);
8388                         if (block_group->iref)
8389                                 break;
8390                         spin_unlock(&block_group->lock);
8391                         block_group = next_block_group(info->tree_root,
8392                                                        block_group);
8393                 }
8394                 if (!block_group) {
8395                         if (last == 0)
8396                                 break;
8397                         last = 0;
8398                         continue;
8399                 }
8400
8401                 inode = block_group->inode;
8402                 block_group->iref = 0;
8403                 block_group->inode = NULL;
8404                 spin_unlock(&block_group->lock);
8405                 iput(inode);
8406                 last = block_group->key.objectid + block_group->key.offset;
8407                 btrfs_put_block_group(block_group);
8408         }
8409 }
8410
8411 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8412 {
8413         struct btrfs_block_group_cache *block_group;
8414         struct btrfs_space_info *space_info;
8415         struct btrfs_caching_control *caching_ctl;
8416         struct rb_node *n;
8417
8418         down_write(&info->extent_commit_sem);
8419         while (!list_empty(&info->caching_block_groups)) {
8420                 caching_ctl = list_entry(info->caching_block_groups.next,
8421                                          struct btrfs_caching_control, list);
8422                 list_del(&caching_ctl->list);
8423                 put_caching_control(caching_ctl);
8424         }
8425         up_write(&info->extent_commit_sem);
8426
8427         spin_lock(&info->block_group_cache_lock);
8428         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8429                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8430                                        cache_node);
8431                 rb_erase(&block_group->cache_node,
8432                          &info->block_group_cache_tree);
8433                 spin_unlock(&info->block_group_cache_lock);
8434
8435                 down_write(&block_group->space_info->groups_sem);
8436                 list_del(&block_group->list);
8437                 up_write(&block_group->space_info->groups_sem);
8438
8439                 if (block_group->cached == BTRFS_CACHE_STARTED)
8440                         wait_block_group_cache_done(block_group);
8441
8442                 /*
8443                  * We haven't cached this block group, which means we could
8444                  * possibly have excluded extents on this block group.
8445                  */
8446                 if (block_group->cached == BTRFS_CACHE_NO)
8447                         free_excluded_extents(info->extent_root, block_group);
8448
8449                 btrfs_remove_free_space_cache(block_group);
8450                 btrfs_put_block_group(block_group);
8451
8452                 spin_lock(&info->block_group_cache_lock);
8453         }
8454         spin_unlock(&info->block_group_cache_lock);
8455
8456         /* now that all the block groups are freed, go through and
8457          * free all the space_info structs.  This is only called during
8458          * the final stages of unmount, and so we know nobody is
8459          * using them.  We call synchronize_rcu() once before we start,
8460          * just to be on the safe side.
8461          */
8462         synchronize_rcu();
8463
8464         release_global_block_rsv(info);
8465
8466         while(!list_empty(&info->space_info)) {
8467                 space_info = list_entry(info->space_info.next,
8468                                         struct btrfs_space_info,
8469                                         list);
8470                 if (space_info->bytes_pinned > 0 ||
8471                     space_info->bytes_reserved > 0) {
8472                         WARN_ON(1);
8473                         dump_space_info(space_info, 0, 0);
8474                 }
8475                 list_del(&space_info->list);
8476                 kfree(space_info);
8477         }
8478         return 0;
8479 }
8480
8481 static void __link_block_group(struct btrfs_space_info *space_info,
8482                                struct btrfs_block_group_cache *cache)
8483 {
8484         int index = get_block_group_index(cache);
8485
8486         down_write(&space_info->groups_sem);
8487         list_add_tail(&cache->list, &space_info->block_groups[index]);
8488         up_write(&space_info->groups_sem);
8489 }
8490
8491 int btrfs_read_block_groups(struct btrfs_root *root)
8492 {
8493         struct btrfs_path *path;
8494         int ret;
8495         struct btrfs_block_group_cache *cache;
8496         struct btrfs_fs_info *info = root->fs_info;
8497         struct btrfs_space_info *space_info;
8498         struct btrfs_key key;
8499         struct btrfs_key found_key;
8500         struct extent_buffer *leaf;
8501         int need_clear = 0;
8502         u64 cache_gen;
8503
8504         root = info->extent_root;
8505         key.objectid = 0;
8506         key.offset = 0;
8507         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8508         path = btrfs_alloc_path();
8509         if (!path)
8510                 return -ENOMEM;
8511
8512         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8513         if (cache_gen != 0 &&
8514             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8515                 need_clear = 1;
8516         if (btrfs_test_opt(root, CLEAR_CACHE))
8517                 need_clear = 1;
8518         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8519                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8520
8521         while (1) {
8522                 ret = find_first_block_group(root, path, &key);
8523                 if (ret > 0)
8524                         break;
8525                 if (ret != 0)
8526                         goto error;
8527                 leaf = path->nodes[0];
8528                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8529                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8530                 if (!cache) {
8531                         ret = -ENOMEM;
8532                         goto error;
8533                 }
8534
8535                 atomic_set(&cache->count, 1);
8536                 spin_lock_init(&cache->lock);
8537                 spin_lock_init(&cache->tree_lock);
8538                 cache->fs_info = info;
8539                 INIT_LIST_HEAD(&cache->list);
8540                 INIT_LIST_HEAD(&cache->cluster_list);
8541
8542                 if (need_clear)
8543                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8544
8545                 /*
8546                  * we only want to have 32k of ram per block group for keeping
8547                  * track of free space, and if we pass 1/2 of that we want to
8548                  * start converting things over to using bitmaps
8549                  */
8550                 cache->extents_thresh = ((1024 * 32) / 2) /
8551                         sizeof(struct btrfs_free_space);
8552
8553                 read_extent_buffer(leaf, &cache->item,
8554                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8555                                    sizeof(cache->item));
8556                 memcpy(&cache->key, &found_key, sizeof(found_key));
8557
8558                 key.objectid = found_key.objectid + found_key.offset;
8559                 btrfs_release_path(root, path);
8560                 cache->flags = btrfs_block_group_flags(&cache->item);
8561                 cache->sectorsize = root->sectorsize;
8562
8563                 /*
8564                  * We need to exclude the super stripes now so that the space
8565                  * info has super bytes accounted for, otherwise we'll think
8566                  * we have more space than we actually do.
8567                  */
8568                 exclude_super_stripes(root, cache);
8569
8570                 /*
8571                  * check for two cases, either we are full, and therefore
8572                  * don't need to bother with the caching work since we won't
8573                  * find any space, or we are empty, and we can just add all
8574                  * the space in and be done with it.  This saves us _alot_ of
8575                  * time, particularly in the full case.
8576                  */
8577                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8578                         cache->last_byte_to_unpin = (u64)-1;
8579                         cache->cached = BTRFS_CACHE_FINISHED;
8580                         free_excluded_extents(root, cache);
8581                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8582                         cache->last_byte_to_unpin = (u64)-1;
8583                         cache->cached = BTRFS_CACHE_FINISHED;
8584                         add_new_free_space(cache, root->fs_info,
8585                                            found_key.objectid,
8586                                            found_key.objectid +
8587                                            found_key.offset);
8588                         free_excluded_extents(root, cache);
8589                 }
8590
8591                 ret = update_space_info(info, cache->flags, found_key.offset,
8592                                         btrfs_block_group_used(&cache->item),
8593                                         &space_info);
8594                 BUG_ON(ret);
8595                 cache->space_info = space_info;
8596                 spin_lock(&cache->space_info->lock);
8597                 cache->space_info->bytes_readonly += cache->bytes_super;
8598                 spin_unlock(&cache->space_info->lock);
8599
8600                 __link_block_group(space_info, cache);
8601
8602                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8603                 BUG_ON(ret);
8604
8605                 set_avail_alloc_bits(root->fs_info, cache->flags);
8606                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8607                         set_block_group_ro(cache);
8608         }
8609
8610         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8611                 if (!(get_alloc_profile(root, space_info->flags) &
8612                       (BTRFS_BLOCK_GROUP_RAID10 |
8613                        BTRFS_BLOCK_GROUP_RAID1 |
8614                        BTRFS_BLOCK_GROUP_DUP)))
8615                         continue;
8616                 /*
8617                  * avoid allocating from un-mirrored block group if there are
8618                  * mirrored block groups.
8619                  */
8620                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8621                         set_block_group_ro(cache);
8622                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8623                         set_block_group_ro(cache);
8624         }
8625
8626         init_global_block_rsv(info);
8627         ret = 0;
8628 error:
8629         btrfs_free_path(path);
8630         return ret;
8631 }
8632
8633 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8634                            struct btrfs_root *root, u64 bytes_used,
8635                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8636                            u64 size)
8637 {
8638         int ret;
8639         struct btrfs_root *extent_root;
8640         struct btrfs_block_group_cache *cache;
8641
8642         extent_root = root->fs_info->extent_root;
8643
8644         root->fs_info->last_trans_log_full_commit = trans->transid;
8645
8646         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8647         if (!cache)
8648                 return -ENOMEM;
8649
8650         cache->key.objectid = chunk_offset;
8651         cache->key.offset = size;
8652         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8653         cache->sectorsize = root->sectorsize;
8654         cache->fs_info = root->fs_info;
8655
8656         /*
8657          * we only want to have 32k of ram per block group for keeping track
8658          * of free space, and if we pass 1/2 of that we want to start
8659          * converting things over to using bitmaps
8660          */
8661         cache->extents_thresh = ((1024 * 32) / 2) /
8662                 sizeof(struct btrfs_free_space);
8663         atomic_set(&cache->count, 1);
8664         spin_lock_init(&cache->lock);
8665         spin_lock_init(&cache->tree_lock);
8666         INIT_LIST_HEAD(&cache->list);
8667         INIT_LIST_HEAD(&cache->cluster_list);
8668
8669         btrfs_set_block_group_used(&cache->item, bytes_used);
8670         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8671         cache->flags = type;
8672         btrfs_set_block_group_flags(&cache->item, type);
8673
8674         cache->last_byte_to_unpin = (u64)-1;
8675         cache->cached = BTRFS_CACHE_FINISHED;
8676         exclude_super_stripes(root, cache);
8677
8678         add_new_free_space(cache, root->fs_info, chunk_offset,
8679                            chunk_offset + size);
8680
8681         free_excluded_extents(root, cache);
8682
8683         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8684                                 &cache->space_info);
8685         BUG_ON(ret);
8686
8687         spin_lock(&cache->space_info->lock);
8688         cache->space_info->bytes_readonly += cache->bytes_super;
8689         spin_unlock(&cache->space_info->lock);
8690
8691         __link_block_group(cache->space_info, cache);
8692
8693         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8694         BUG_ON(ret);
8695
8696         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8697                                 sizeof(cache->item));
8698         BUG_ON(ret);
8699
8700         set_avail_alloc_bits(extent_root->fs_info, type);
8701
8702         return 0;
8703 }
8704
8705 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8706                              struct btrfs_root *root, u64 group_start)
8707 {
8708         struct btrfs_path *path;
8709         struct btrfs_block_group_cache *block_group;
8710         struct btrfs_free_cluster *cluster;
8711         struct btrfs_root *tree_root = root->fs_info->tree_root;
8712         struct btrfs_key key;
8713         struct inode *inode;
8714         int ret;
8715         int factor;
8716
8717         root = root->fs_info->extent_root;
8718
8719         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8720         BUG_ON(!block_group);
8721         BUG_ON(!block_group->ro);
8722
8723         /*
8724          * Free the reserved super bytes from this block group before
8725          * remove it.
8726          */
8727         free_excluded_extents(root, block_group);
8728
8729         memcpy(&key, &block_group->key, sizeof(key));
8730         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8731                                   BTRFS_BLOCK_GROUP_RAID1 |
8732                                   BTRFS_BLOCK_GROUP_RAID10))
8733                 factor = 2;
8734         else
8735                 factor = 1;
8736
8737         /* make sure this block group isn't part of an allocation cluster */
8738         cluster = &root->fs_info->data_alloc_cluster;
8739         spin_lock(&cluster->refill_lock);
8740         btrfs_return_cluster_to_free_space(block_group, cluster);
8741         spin_unlock(&cluster->refill_lock);
8742
8743         /*
8744          * make sure this block group isn't part of a metadata
8745          * allocation cluster
8746          */
8747         cluster = &root->fs_info->meta_alloc_cluster;
8748         spin_lock(&cluster->refill_lock);
8749         btrfs_return_cluster_to_free_space(block_group, cluster);
8750         spin_unlock(&cluster->refill_lock);
8751
8752         path = btrfs_alloc_path();
8753         BUG_ON(!path);
8754
8755         inode = lookup_free_space_inode(root, block_group, path);
8756         if (!IS_ERR(inode)) {
8757                 btrfs_orphan_add(trans, inode);
8758                 clear_nlink(inode);
8759                 /* One for the block groups ref */
8760                 spin_lock(&block_group->lock);
8761                 if (block_group->iref) {
8762                         block_group->iref = 0;
8763                         block_group->inode = NULL;
8764                         spin_unlock(&block_group->lock);
8765                         iput(inode);
8766                 } else {
8767                         spin_unlock(&block_group->lock);
8768                 }
8769                 /* One for our lookup ref */
8770                 iput(inode);
8771         }
8772
8773         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8774         key.offset = block_group->key.objectid;
8775         key.type = 0;
8776
8777         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8778         if (ret < 0)
8779                 goto out;
8780         if (ret > 0)
8781                 btrfs_release_path(tree_root, path);
8782         if (ret == 0) {
8783                 ret = btrfs_del_item(trans, tree_root, path);
8784                 if (ret)
8785                         goto out;
8786                 btrfs_release_path(tree_root, path);
8787         }
8788
8789         spin_lock(&root->fs_info->block_group_cache_lock);
8790         rb_erase(&block_group->cache_node,
8791                  &root->fs_info->block_group_cache_tree);
8792         spin_unlock(&root->fs_info->block_group_cache_lock);
8793
8794         down_write(&block_group->space_info->groups_sem);
8795         /*
8796          * we must use list_del_init so people can check to see if they
8797          * are still on the list after taking the semaphore
8798          */
8799         list_del_init(&block_group->list);
8800         up_write(&block_group->space_info->groups_sem);
8801
8802         if (block_group->cached == BTRFS_CACHE_STARTED)
8803                 wait_block_group_cache_done(block_group);
8804
8805         btrfs_remove_free_space_cache(block_group);
8806
8807         spin_lock(&block_group->space_info->lock);
8808         block_group->space_info->total_bytes -= block_group->key.offset;
8809         block_group->space_info->bytes_readonly -= block_group->key.offset;
8810         block_group->space_info->disk_total -= block_group->key.offset * factor;
8811         spin_unlock(&block_group->space_info->lock);
8812
8813         memcpy(&key, &block_group->key, sizeof(key));
8814
8815         btrfs_clear_space_info_full(root->fs_info);
8816
8817         btrfs_put_block_group(block_group);
8818         btrfs_put_block_group(block_group);
8819
8820         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8821         if (ret > 0)
8822                 ret = -EIO;
8823         if (ret < 0)
8824                 goto out;
8825
8826         ret = btrfs_del_item(trans, root, path);
8827 out:
8828         btrfs_free_path(path);
8829         return ret;
8830 }
8831
8832 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8833 {
8834         struct btrfs_space_info *space_info;
8835         int ret;
8836
8837         ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM, 0, 0,
8838                                                                  &space_info);
8839         if (ret)
8840                 return ret;
8841
8842         ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA, 0, 0,
8843                                                                  &space_info);
8844         if (ret)
8845                 return ret;
8846
8847         ret = update_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA, 0, 0,
8848                                                                  &space_info);
8849         if (ret)
8850                 return ret;
8851
8852         return ret;
8853 }
8854
8855 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8856 {
8857         return unpin_extent_range(root, start, end);
8858 }
8859
8860 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8861                                u64 num_bytes, u64 *actual_bytes)
8862 {
8863         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8864 }
8865
8866 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8867 {
8868         struct btrfs_fs_info *fs_info = root->fs_info;
8869         struct btrfs_block_group_cache *cache = NULL;
8870         u64 group_trimmed;
8871         u64 start;
8872         u64 end;
8873         u64 trimmed = 0;
8874         int ret = 0;
8875
8876         cache = btrfs_lookup_block_group(fs_info, range->start);
8877
8878         while (cache) {
8879                 if (cache->key.objectid >= (range->start + range->len)) {
8880                         btrfs_put_block_group(cache);
8881                         break;
8882                 }
8883
8884                 start = max(range->start, cache->key.objectid);
8885                 end = min(range->start + range->len,
8886                                 cache->key.objectid + cache->key.offset);
8887
8888                 if (end - start >= range->minlen) {
8889                         if (!block_group_cache_done(cache)) {
8890                                 ret = cache_block_group(cache, NULL, root, 0);
8891                                 if (!ret)
8892                                         wait_block_group_cache_done(cache);
8893                         }
8894                         ret = btrfs_trim_block_group(cache,
8895                                                      &group_trimmed,
8896                                                      start,
8897                                                      end,
8898                                                      range->minlen);
8899
8900                         trimmed += group_trimmed;
8901                         if (ret) {
8902                                 btrfs_put_block_group(cache);
8903                                 break;
8904                         }
8905                 }
8906
8907                 cache = next_block_group(fs_info->tree_root, cache);
8908         }
8909
8910         range->len = trimmed;
8911         return ret;
8912 }