fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <linux/slab.h>
  26 #include "compat.h"
  27 #include "hash.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "print-tree.h"
  31 #include "transaction.h"
  32 #include "volumes.h"
  33 #include "locking.h"
  34 #include "free-space-cache.h"
  35
  36 static int update_block_group(struct btrfs_trans_handle *trans,
  37                               struct btrfs_root *root,
  38                               u64 bytenr, u64 num_bytes, int alloc);
  39 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
  40                                  u64 num_bytes, int reserve, int sinfo);
  41 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  42                                 struct btrfs_root *root,
  43                                 u64 bytenr, u64 num_bytes, u64 parent,
  44                                 u64 root_objectid, u64 owner_objectid,
  45                                 u64 owner_offset, int refs_to_drop,
  46                                 struct btrfs_delayed_extent_op *extra_op);
  47 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  48                                     struct extent_buffer *leaf,
  49                                     struct btrfs_extent_item *ei);
  50 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  51                                       struct btrfs_root *root,
  52                                       u64 parent, u64 root_objectid,
  53                                       u64 flags, u64 owner, u64 offset,
  54                                       struct btrfs_key *ins, int ref_mod);
  55 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  56                                      struct btrfs_root *root,
  57                                      u64 parent, u64 root_objectid,
  58                                      u64 flags, struct btrfs_disk_key *key,
  59                                      int level, struct btrfs_key *ins);
  60 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  61                           struct btrfs_root *extent_root, u64 alloc_bytes,
  62                           u64 flags, int force);
  63 static int find_next_key(struct btrfs_path *path, int level,
  64                          struct btrfs_key *key);
  65 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  66                             int dump_block_groups);
  67
  68 static noinline int
  69 block_group_cache_done(struct btrfs_block_group_cache *cache)
  70 {
  71         smp_mb();
  72         return cache->cached == BTRFS_CACHE_FINISHED;
  73 }
  74
  75 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  76 {
  77         return (cache->flags & bits) == bits;
  78 }
  79
  80 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  81 {
  82         atomic_inc(&cache->count);
  83 }
  84
  85 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
  86 {
  87         if (atomic_dec_and_test(&cache->count)) {
  88                 WARN_ON(cache->pinned > 0);
  89                 WARN_ON(cache->reserved > 0);
  90                 WARN_ON(cache->reserved_pinned > 0);
  91                 kfree(cache);
  92         }
  93 }
  94
  95 /*
  96  * this adds the block group to the fs_info rb tree for the block group
  97  * cache
  98  */
  99 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 100                                 struct btrfs_block_group_cache *block_group)
 101 {
 102         struct rb_node **p;
 103         struct rb_node *parent = NULL;
 104         struct btrfs_block_group_cache *cache;
 105
 106         spin_lock(&info->block_group_cache_lock);
 107         p = &info->block_group_cache_tree.rb_node;
 108
 109         while (*p) {
 110                 parent = *p;
 111                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 112                                  cache_node);
 113                 if (block_group->key.objectid < cache->key.objectid) {
 114                         p = &(*p)->rb_left;
 115                 } else if (block_group->key.objectid > cache->key.objectid) {
 116                         p = &(*p)->rb_right;
 117                 } else {
 118                         spin_unlock(&info->block_group_cache_lock);
 119                         return -EEXIST;
 120                 }
 121         }
 122
 123         rb_link_node(&block_group->cache_node, parent, p);
 124         rb_insert_color(&block_group->cache_node,
 125                         &info->block_group_cache_tree);
 126         spin_unlock(&info->block_group_cache_lock);
 127
 128         return 0;
 129 }
 130
 131 /*
 132  * This will return the block group at or after bytenr if contains is 0, else
 133  * it will return the block group that contains the bytenr
 134  */
 135 static struct btrfs_block_group_cache *
 136 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 137                               int contains)
 138 {
 139         struct btrfs_block_group_cache *cache, *ret = NULL;
 140         struct rb_node *n;
 141         u64 end, start;
 142
 143         spin_lock(&info->block_group_cache_lock);
 144         n = info->block_group_cache_tree.rb_node;
 145
 146         while (n) {
 147                 cache = rb_entry(n, struct btrfs_block_group_cache,
 148                                  cache_node);
 149                 end = cache->key.objectid + cache->key.offset - 1;
 150                 start = cache->key.objectid;
 151
 152                 if (bytenr < start) {
 153                         if (!contains && (!ret || start < ret->key.objectid))
 154                                 ret = cache;
 155                         n = n->rb_left;
 156                 } else if (bytenr > start) {
 157                         if (contains && bytenr <= end) {
 158                                 ret = cache;
 159                                 break;
 160                         }
 161                         n = n->rb_right;
 162                 } else {
 163                         ret = cache;
 164                         break;
 165                 }
 166         }
 167         if (ret)
 168                 btrfs_get_block_group(ret);
 169         spin_unlock(&info->block_group_cache_lock);
 170
 171         return ret;
 172 }
 173
 174 static int add_excluded_extent(struct btrfs_root *root,
 175                                u64 start, u64 num_bytes)
 176 {
 177         u64 end = start + num_bytes - 1;
 178         set_extent_bits(&root->fs_info->freed_extents[0],
 179                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 180         set_extent_bits(&root->fs_info->freed_extents[1],
 181                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 182         return 0;
 183 }
 184
 185 static void free_excluded_extents(struct btrfs_root *root,
 186                                   struct btrfs_block_group_cache *cache)
 187 {
 188         u64 start, end;
 189
 190         start = cache->key.objectid;
 191         end = start + cache->key.offset - 1;
 192
 193         clear_extent_bits(&root->fs_info->freed_extents[0],
 194                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 195         clear_extent_bits(&root->fs_info->freed_extents[1],
 196                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 197 }
 198
 199 static int exclude_super_stripes(struct btrfs_root *root,
 200                                  struct btrfs_block_group_cache *cache)
 201 {
 202         u64 bytenr;
 203         u64 *logical;
 204         int stripe_len;
 205         int i, nr, ret;
 206
 207         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 208                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 209                 cache->bytes_super += stripe_len;
 210                 ret = add_excluded_extent(root, cache->key.objectid,
 211                                           stripe_len);
 212                 BUG_ON(ret);
 213         }
 214
 215         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 216                 bytenr = btrfs_sb_offset(i);
 217                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 218                                        cache->key.objectid, bytenr,
 219                                        0, &logical, &nr, &stripe_len);
 220                 BUG_ON(ret);
 221
 222                 while (nr--) {
 223                         cache->bytes_super += stripe_len;
 224                         ret = add_excluded_extent(root, logical[nr],
 225                                                   stripe_len);
 226                         BUG_ON(ret);
 227                 }
 228
 229                 kfree(logical);
 230         }
 231         return 0;
 232 }
 233
 234 static struct btrfs_caching_control *
 235 get_caching_control(struct btrfs_block_group_cache *cache)
 236 {
 237         struct btrfs_caching_control *ctl;
 238
 239         spin_lock(&cache->lock);
 240         if (cache->cached != BTRFS_CACHE_STARTED) {
 241                 spin_unlock(&cache->lock);
 242                 return NULL;
 243         }
 244
 245         /* We're loading it the fast way, so we don't have a caching_ctl. */
 246         if (!cache->caching_ctl) {
 247                 spin_unlock(&cache->lock);
 248                 return NULL;
 249         }
 250
 251         ctl = cache->caching_ctl;
 252         atomic_inc(&ctl->count);
 253         spin_unlock(&cache->lock);
 254         return ctl;
 255 }
 256
 257 static void put_caching_control(struct btrfs_caching_control *ctl)
 258 {
 259         if (atomic_dec_and_test(&ctl->count))
 260                 kfree(ctl);
 261 }
 262
 263 /*
 264  * this is only called by cache_block_group, since we could have freed extents
 265  * we need to check the pinned_extents for any extents that can't be used yet
 266  * since their free space will be released as soon as the transaction commits.
 267  */
 268 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 269                               struct btrfs_fs_info *info, u64 start, u64 end)
 270 {
 271         u64 extent_start, extent_end, size, total_added = 0;
 272         int ret;
 273
 274         while (start < end) {
 275                 ret = find_first_extent_bit(info->pinned_extents, start,
 276                                             &extent_start, &extent_end,
 277                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 278                 if (ret)
 279                         break;
 280
 281                 if (extent_start <= start) {
 282                         start = extent_end + 1;
 283                 } else if (extent_start > start && extent_start < end) {
 284                         size = extent_start - start;
 285                         total_added += size;
 286                         ret = btrfs_add_free_space(block_group, start,
 287                                                    size);
 288                         BUG_ON(ret);
 289                         start = extent_end + 1;
 290                 } else {
 291                         break;
 292                 }
 293         }
 294
 295         if (start < end) {
 296                 size = end - start;
 297                 total_added += size;
 298                 ret = btrfs_add_free_space(block_group, start, size);
 299                 BUG_ON(ret);
 300         }
 301
 302         return total_added;
 303 }
 304
 305 static int caching_kthread(void *data)
 306 {
 307         struct btrfs_block_group_cache *block_group = data;
 308         struct btrfs_fs_info *fs_info = block_group->fs_info;
 309         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 310         struct btrfs_root *extent_root = fs_info->extent_root;
 311         struct btrfs_path *path;
 312         struct extent_buffer *leaf;
 313         struct btrfs_key key;
 314         u64 total_found = 0;
 315         u64 last = 0;
 316         u32 nritems;
 317         int ret = 0;
 318
 319         path = btrfs_alloc_path();
 320         if (!path)
 321                 return -ENOMEM;
 322
 323         exclude_super_stripes(extent_root, block_group);
 324         spin_lock(&block_group->space_info->lock);
 325         block_group->space_info->bytes_readonly += block_group->bytes_super;
 326         spin_unlock(&block_group->space_info->lock);
 327
 328         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 329
 330         /*
 331          * We don't want to deadlock with somebody trying to allocate a new
 332          * extent for the extent root while also trying to search the extent
 333          * root to add free space.  So we skip locking and search the commit
 334          * root, since its read-only
 335          */
 336         path->skip_locking = 1;
 337         path->search_commit_root = 1;
 338         path->reada = 2;
 339
 340         key.objectid = last;
 341         key.offset = 0;
 342         key.type = BTRFS_EXTENT_ITEM_KEY;
 343 again:
 344         mutex_lock(&caching_ctl->mutex);
 345         /* need to make sure the commit_root doesn't disappear */
 346         down_read(&fs_info->extent_commit_sem);
 347
 348         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 349         if (ret < 0)
 350                 goto err;
 351
 352         leaf = path->nodes[0];
 353         nritems = btrfs_header_nritems(leaf);
 354
 355         while (1) {
 356                 smp_mb();
 357                 if (fs_info->closing > 1) {
 358                         last = (u64)-1;
 359                         break;
 360                 }
 361
 362                 if (path->slots[0] < nritems) {
 363                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 364                 } else {
 365                         ret = find_next_key(path, 0, &key);
 366                         if (ret)
 367                                 break;
 368
 369                         caching_ctl->progress = last;
 370                         btrfs_release_path(extent_root, path);
 371                         up_read(&fs_info->extent_commit_sem);
 372                         mutex_unlock(&caching_ctl->mutex);
 373                         if (btrfs_transaction_in_commit(fs_info))
 374                                 schedule_timeout(1);
 375                         else
 376                                 cond_resched();
 377                         goto again;
 378                 }
 379
 380                 if (key.objectid < block_group->key.objectid) {
 381                         path->slots[0]++;
 382                         continue;
 383                 }
 384
 385                 if (key.objectid >= block_group->key.objectid +
 386                     block_group->key.offset)
 387                         break;
 388
 389                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 390                         total_found += add_new_free_space(block_group,
 391                                                           fs_info, last,
 392                                                           key.objectid);
 393                         last = key.objectid + key.offset;
 394
 395                         if (total_found > (1024 * 1024 * 2)) {
 396                                 total_found = 0;
 397                                 wake_up(&caching_ctl->wait);
 398                         }
 399                 }
 400                 path->slots[0]++;
 401         }
 402         ret = 0;
 403
 404         total_found += add_new_free_space(block_group, fs_info, last,
 405                                           block_group->key.objectid +
 406                                           block_group->key.offset);
 407         caching_ctl->progress = (u64)-1;
 408
 409         spin_lock(&block_group->lock);
 410         block_group->caching_ctl = NULL;
 411         block_group->cached = BTRFS_CACHE_FINISHED;
 412         spin_unlock(&block_group->lock);
 413
 414 err:
 415         btrfs_free_path(path);
 416         up_read(&fs_info->extent_commit_sem);
 417
 418         free_excluded_extents(extent_root, block_group);
 419
 420         mutex_unlock(&caching_ctl->mutex);
 421         wake_up(&caching_ctl->wait);
 422
 423         put_caching_control(caching_ctl);
 424         atomic_dec(&block_group->space_info->caching_threads);
 425         btrfs_put_block_group(block_group);
 426
 427         return 0;
 428 }
 429
 430 static int cache_block_group(struct btrfs_block_group_cache *cache,
 431                              struct btrfs_trans_handle *trans,
 432                              struct btrfs_root *root,
 433                              int load_cache_only)
 434 {
 435         struct btrfs_fs_info *fs_info = cache->fs_info;
 436         struct btrfs_caching_control *caching_ctl;
 437         struct task_struct *tsk;
 438         int ret = 0;
 439
 440         smp_mb();
 441         if (cache->cached != BTRFS_CACHE_NO)
 442                 return 0;
 443
 444         /*
 445          * We can't do the read from on-disk cache during a commit since we need
 446          * to have the normal tree locking.  Also if we are currently trying to
 447          * allocate blocks for the tree root we can't do the fast caching since
 448          * we likely hold important locks.
 449          */
 450         if (!trans->transaction->in_commit &&
 451             (root && root != root->fs_info->tree_root)) {
 452                 spin_lock(&cache->lock);
 453                 if (cache->cached != BTRFS_CACHE_NO) {
 454                         spin_unlock(&cache->lock);
 455                         return 0;
 456                 }
 457                 cache->cached = BTRFS_CACHE_STARTED;
 458                 spin_unlock(&cache->lock);
 459
 460                 ret = load_free_space_cache(fs_info, cache);
 461
 462                 spin_lock(&cache->lock);
 463                 if (ret == 1) {
 464                         cache->cached = BTRFS_CACHE_FINISHED;
 465                         cache->last_byte_to_unpin = (u64)-1;
 466                 } else {
 467                         cache->cached = BTRFS_CACHE_NO;
 468                 }
 469                 spin_unlock(&cache->lock);
 470                 if (ret == 1)
 471                         return 0;
 472         }
 473
 474         if (load_cache_only)
 475                 return 0;
 476
 477         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 478         BUG_ON(!caching_ctl);
 479
 480         INIT_LIST_HEAD(&caching_ctl->list);
 481         mutex_init(&caching_ctl->mutex);
 482         init_waitqueue_head(&caching_ctl->wait);
 483         caching_ctl->block_group = cache;
 484         caching_ctl->progress = cache->key.objectid;
 485         /* one for caching kthread, one for caching block group list */
 486         atomic_set(&caching_ctl->count, 2);
 487
 488         spin_lock(&cache->lock);
 489         if (cache->cached != BTRFS_CACHE_NO) {
 490                 spin_unlock(&cache->lock);
 491                 kfree(caching_ctl);
 492                 return 0;
 493         }
 494         cache->caching_ctl = caching_ctl;
 495         cache->cached = BTRFS_CACHE_STARTED;
 496         spin_unlock(&cache->lock);
 497
 498         down_write(&fs_info->extent_commit_sem);
 499         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 500         up_write(&fs_info->extent_commit_sem);
 501
 502         atomic_inc(&cache->space_info->caching_threads);
 503         btrfs_get_block_group(cache);
 504
 505         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 506                           cache->key.objectid);
 507         if (IS_ERR(tsk)) {
 508                 ret = PTR_ERR(tsk);
 509                 printk(KERN_ERR "error running thread %d\n", ret);
 510                 BUG();
 511         }
 512
 513         return ret;
 514 }
 515
 516 /*
 517  * return the block group that starts at or after bytenr
 518  */
 519 static struct btrfs_block_group_cache *
 520 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 521 {
 522         struct btrfs_block_group_cache *cache;
 523
 524         cache = block_group_cache_tree_search(info, bytenr, 0);
 525
 526         return cache;
 527 }
 528
 529 /*
 530  * return the block group that contains the given bytenr
 531  */
 532 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 533                                                  struct btrfs_fs_info *info,
 534                                                  u64 bytenr)
 535 {
 536         struct btrfs_block_group_cache *cache;
 537
 538         cache = block_group_cache_tree_search(info, bytenr, 1);
 539
 540         return cache;
 541 }
 542
 543 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 544                                                   u64 flags)
 545 {
 546         struct list_head *head = &info->space_info;
 547         struct btrfs_space_info *found;
 548
 549         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 550                  BTRFS_BLOCK_GROUP_METADATA;
 551
 552         rcu_read_lock();
 553         list_for_each_entry_rcu(found, head, list) {
 554                 if (found->flags & flags) {
 555                         rcu_read_unlock();
 556                         return found;
 557                 }
 558         }
 559         rcu_read_unlock();
 560         return NULL;
 561 }
 562
 563 /*
 564  * after adding space to the filesystem, we need to clear the full flags
 565  * on all the space infos.
 566  */
 567 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 568 {
 569         struct list_head *head = &info->space_info;
 570         struct btrfs_space_info *found;
 571
 572         rcu_read_lock();
 573         list_for_each_entry_rcu(found, head, list)
 574                 found->full = 0;
 575         rcu_read_unlock();
 576 }
 577
 578 static u64 div_factor(u64 num, int factor)
 579 {
 580         if (factor == 10)
 581                 return num;
 582         num *= factor;
 583         do_div(num, 10);
 584         return num;
 585 }
 586
 587 static u64 div_factor_fine(u64 num, int factor)
 588 {
 589         if (factor == 100)
 590                 return num;
 591         num *= factor;
 592         do_div(num, 100);
 593         return num;
 594 }
 595
 596 u64 btrfs_find_block_group(struct btrfs_root *root,
 597                            u64 search_start, u64 search_hint, int owner)
 598 {
 599         struct btrfs_block_group_cache *cache;
 600         u64 used;
 601         u64 last = max(search_hint, search_start);
 602         u64 group_start = 0;
 603         int full_search = 0;
 604         int factor = 9;
 605         int wrapped = 0;
 606 again:
 607         while (1) {
 608                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 609                 if (!cache)
 610                         break;
 611
 612                 spin_lock(&cache->lock);
 613                 last = cache->key.objectid + cache->key.offset;
 614                 used = btrfs_block_group_used(&cache->item);
 615
 616                 if ((full_search || !cache->ro) &&
 617                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 618                         if (used + cache->pinned + cache->reserved <
 619                             div_factor(cache->key.offset, factor)) {
 620                                 group_start = cache->key.objectid;
 621                                 spin_unlock(&cache->lock);
 622                                 btrfs_put_block_group(cache);
 623                                 goto found;
 624                         }
 625                 }
 626                 spin_unlock(&cache->lock);
 627                 btrfs_put_block_group(cache);
 628                 cond_resched();
 629         }
 630         if (!wrapped) {
 631                 last = search_start;
 632                 wrapped = 1;
 633                 goto again;
 634         }
 635         if (!full_search && factor < 10) {
 636                 last = search_start;
 637                 full_search = 1;
 638                 factor = 10;
 639                 goto again;
 640         }
 641 found:
 642         return group_start;
 643 }
 644
 645 /* simple helper to search for an existing extent at a given offset */
 646 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 647 {
 648         int ret;
 649         struct btrfs_key key;
 650         struct btrfs_path *path;
 651
 652         path = btrfs_alloc_path();
 653         BUG_ON(!path);
 654         key.objectid = start;
 655         key.offset = len;
 656         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 657         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 658                                 0, 0);
 659         btrfs_free_path(path);
 660         return ret;
 661 }
 662
 663 /*
 664  * helper function to lookup reference count and flags of extent.
 665  *
 666  * the head node for delayed ref is used to store the sum of all the
 667  * reference count modifications queued up in the rbtree. the head
 668  * node may also store the extent flags to set. This way you can check
 669  * to see what the reference count and extent flags would be if all of
 670  * the delayed refs are not processed.
 671  */
 672 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 673                              struct btrfs_root *root, u64 bytenr,
 674                              u64 num_bytes, u64 *refs, u64 *flags)
 675 {
 676         struct btrfs_delayed_ref_head *head;
 677         struct btrfs_delayed_ref_root *delayed_refs;
 678         struct btrfs_path *path;
 679         struct btrfs_extent_item *ei;
 680         struct extent_buffer *leaf;
 681         struct btrfs_key key;
 682         u32 item_size;
 683         u64 num_refs;
 684         u64 extent_flags;
 685         int ret;
 686
 687         path = btrfs_alloc_path();
 688         if (!path)
 689                 return -ENOMEM;
 690
 691         key.objectid = bytenr;
 692         key.type = BTRFS_EXTENT_ITEM_KEY;
 693         key.offset = num_bytes;
 694         if (!trans) {
 695                 path->skip_locking = 1;
 696                 path->search_commit_root = 1;
 697         }
 698 again:
 699         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 700                                 &key, path, 0, 0);
 701         if (ret < 0)
 702                 goto out_free;
 703
 704         if (ret == 0) {
 705                 leaf = path->nodes[0];
 706                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 707                 if (item_size >= sizeof(*ei)) {
 708                         ei = btrfs_item_ptr(leaf, path->slots[0],
 709                                             struct btrfs_extent_item);
 710                         num_refs = btrfs_extent_refs(leaf, ei);
 711                         extent_flags = btrfs_extent_flags(leaf, ei);
 712                 } else {
 713 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 714                         struct btrfs_extent_item_v0 *ei0;
 715                         BUG_ON(item_size != sizeof(*ei0));
 716                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 717                                              struct btrfs_extent_item_v0);
 718                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 719                         /* FIXME: this isn't correct for data */
 720                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 721 #else
 722                         BUG();
 723 #endif
 724                 }
 725                 BUG_ON(num_refs == 0);
 726         } else {
 727                 num_refs = 0;
 728                 extent_flags = 0;
 729                 ret = 0;
 730         }
 731
 732         if (!trans)
 733                 goto out;
 734
 735         delayed_refs = &trans->transaction->delayed_refs;
 736         spin_lock(&delayed_refs->lock);
 737         head = btrfs_find_delayed_ref_head(trans, bytenr);
 738         if (head) {
 739                 if (!mutex_trylock(&head->mutex)) {
 740                         atomic_inc(&head->node.refs);
 741                         spin_unlock(&delayed_refs->lock);
 742
 743                         btrfs_release_path(root->fs_info->extent_root, path);
 744
 745                         mutex_lock(&head->mutex);
 746                         mutex_unlock(&head->mutex);
 747                         btrfs_put_delayed_ref(&head->node);
 748                         goto again;
 749                 }
 750                 if (head->extent_op && head->extent_op->update_flags)
 751                         extent_flags |= head->extent_op->flags_to_set;
 752                 else
 753                         BUG_ON(num_refs == 0);
 754
 755                 num_refs += head->node.ref_mod;
 756                 mutex_unlock(&head->mutex);
 757         }
 758         spin_unlock(&delayed_refs->lock);
 759 out:
 760         WARN_ON(num_refs == 0);
 761         if (refs)
 762                 *refs = num_refs;
 763         if (flags)
 764                 *flags = extent_flags;
 765 out_free:
 766         btrfs_free_path(path);
 767         return ret;
 768 }
 769
 770 /*
 771  * Back reference rules.  Back refs have three main goals:
 772  *
 773  * 1) differentiate between all holders of references to an extent so that
 774  *    when a reference is dropped we can make sure it was a valid reference
 775  *    before freeing the extent.
 776  *
 777  * 2) Provide enough information to quickly find the holders of an extent
 778  *    if we notice a given block is corrupted or bad.
 779  *
 780  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 781  *    maintenance.  This is actually the same as #2, but with a slightly
 782  *    different use case.
 783  *
 784  * There are two kinds of back refs. The implicit back refs is optimized
 785  * for pointers in non-shared tree blocks. For a given pointer in a block,
 786  * back refs of this kind provide information about the block's owner tree
 787  * and the pointer's key. These information allow us to find the block by
 788  * b-tree searching. The full back refs is for pointers in tree blocks not
 789  * referenced by their owner trees. The location of tree block is recorded
 790  * in the back refs. Actually the full back refs is generic, and can be
 791  * used in all cases the implicit back refs is used. The major shortcoming
 792  * of the full back refs is its overhead. Every time a tree block gets
 793  * COWed, we have to update back refs entry for all pointers in it.
 794  *
 795  * For a newly allocated tree block, we use implicit back refs for
 796  * pointers in it. This means most tree related operations only involve
 797  * implicit back refs. For a tree block created in old transaction, the
 798  * only way to drop a reference to it is COW it. So we can detect the
 799  * event that tree block loses its owner tree's reference and do the
 800  * back refs conversion.
 801  *
 802  * When a tree block is COW'd through a tree, there are four cases:
 803  *
 804  * The reference count of the block is one and the tree is the block's
 805  * owner tree. Nothing to do in this case.
 806  *
 807  * The reference count of the block is one and the tree is not the
 808  * block's owner tree. In this case, full back refs is used for pointers
 809  * in the block. Remove these full back refs, add implicit back refs for
 810  * every pointers in the new block.
 811  *
 812  * The reference count of the block is greater than one and the tree is
 813  * the block's owner tree. In this case, implicit back refs is used for
 814  * pointers in the block. Add full back refs for every pointers in the
 815  * block, increase lower level extents' reference counts. The original
 816  * implicit back refs are entailed to the new block.
 817  *
 818  * The reference count of the block is greater than one and the tree is
 819  * not the block's owner tree. Add implicit back refs for every pointer in
 820  * the new block, increase lower level extents' reference count.
 821  *
 822  * Back Reference Key composing:
 823  *
 824  * The key objectid corresponds to the first byte in the extent,
 825  * The key type is used to differentiate between types of back refs.
 826  * There are different meanings of the key offset for different types
 827  * of back refs.
 828  *
 829  * File extents can be referenced by:
 830  *
 831  * - multiple snapshots, subvolumes, or different generations in one subvol
 832  * - different files inside a single subvolume
 833  * - different offsets inside a file (bookend extents in file.c)
 834  *
 835  * The extent ref structure for the implicit back refs has fields for:
 836  *
 837  * - Objectid of the subvolume root
 838  * - objectid of the file holding the reference
 839  * - original offset in the file
 840  * - how many bookend extents
 841  *
 842  * The key offset for the implicit back refs is hash of the first
 843  * three fields.
 844  *
 845  * The extent ref structure for the full back refs has field for:
 846  *
 847  * - number of pointers in the tree leaf
 848  *
 849  * The key offset for the implicit back refs is the first byte of
 850  * the tree leaf
 851  *
 852  * When a file extent is allocated, The implicit back refs is used.
 853  * the fields are filled in:
 854  *
 855  *     (root_key.objectid, inode objectid, offset in file, 1)
 856  *
 857  * When a file extent is removed file truncation, we find the
 858  * corresponding implicit back refs and check the following fields:
 859  *
 860  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 861  *
 862  * Btree extents can be referenced by:
 863  *
 864  * - Different subvolumes
 865  *
 866  * Both the implicit back refs and the full back refs for tree blocks
 867  * only consist of key. The key offset for the implicit back refs is
 868  * objectid of block's owner tree. The key offset for the full back refs
 869  * is the first byte of parent block.
 870  *
 871  * When implicit back refs is used, information about the lowest key and
 872  * level of the tree block are required. These information are stored in
 873  * tree block info structure.
 874  */
 875
 876 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 877 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 878                                   struct btrfs_root *root,
 879                                   struct btrfs_path *path,
 880                                   u64 owner, u32 extra_size)
 881 {
 882         struct btrfs_extent_item *item;
 883         struct btrfs_extent_item_v0 *ei0;
 884         struct btrfs_extent_ref_v0 *ref0;
 885         struct btrfs_tree_block_info *bi;
 886         struct extent_buffer *leaf;
 887         struct btrfs_key key;
 888         struct btrfs_key found_key;
 889         u32 new_size = sizeof(*item);
 890         u64 refs;
 891         int ret;
 892
 893         leaf = path->nodes[0];
 894         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 895
 896         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 897         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 898                              struct btrfs_extent_item_v0);
 899         refs = btrfs_extent_refs_v0(leaf, ei0);
 900
 901         if (owner == (u64)-1) {
 902                 while (1) {
 903                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 904                                 ret = btrfs_next_leaf(root, path);
 905                                 if (ret < 0)
 906                                         return ret;
 907                                 BUG_ON(ret > 0);
 908                                 leaf = path->nodes[0];
 909                         }
 910                         btrfs_item_key_to_cpu(leaf, &found_key,
 911                                               path->slots[0]);
 912                         BUG_ON(key.objectid != found_key.objectid);
 913                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 914                                 path->slots[0]++;
 915                                 continue;
 916                         }
 917                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 918                                               struct btrfs_extent_ref_v0);
 919                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 920                         break;
 921                 }
 922         }
 923         btrfs_release_path(root, path);
 924
 925         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 926                 new_size += sizeof(*bi);
 927
 928         new_size -= sizeof(*ei0);
 929         ret = btrfs_search_slot(trans, root, &key, path,
 930                                 new_size + extra_size, 1);
 931         if (ret < 0)
 932                 return ret;
 933         BUG_ON(ret);
 934
 935         ret = btrfs_extend_item(trans, root, path, new_size);
 936         BUG_ON(ret);
 937
 938         leaf = path->nodes[0];
 939         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 940         btrfs_set_extent_refs(leaf, item, refs);
 941         /* FIXME: get real generation */
 942         btrfs_set_extent_generation(leaf, item, 0);
 943         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 944                 btrfs_set_extent_flags(leaf, item,
 945                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 946                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 947                 bi = (struct btrfs_tree_block_info *)(item + 1);
 948                 /* FIXME: get first key of the block */
 949                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 950                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 951         } else {
 952                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 953         }
 954         btrfs_mark_buffer_dirty(leaf);
 955         return 0;
 956 }
 957 #endif
 958
 959 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 960 {
 961         u32 high_crc = ~(u32)0;
 962         u32 low_crc = ~(u32)0;
 963         __le64 lenum;
 964
 965         lenum = cpu_to_le64(root_objectid);
 966         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 967         lenum = cpu_to_le64(owner);
 968         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 969         lenum = cpu_to_le64(offset);
 970         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 971
 972         return ((u64)high_crc << 31) ^ (u64)low_crc;
 973 }
 974
 975 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 976                                      struct btrfs_extent_data_ref *ref)
 977 {
 978         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 979                                     btrfs_extent_data_ref_objectid(leaf, ref),
 980                                     btrfs_extent_data_ref_offset(leaf, ref));
 981 }
 982
 983 static int match_extent_data_ref(struct extent_buffer *leaf,
 984                                  struct btrfs_extent_data_ref *ref,
 985                                  u64 root_objectid, u64 owner, u64 offset)
 986 {
 987         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 988             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 989             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 990                 return 0;
 991         return 1;
 992 }
 993
 994 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 995                                            struct btrfs_root *root,
 996                                            struct btrfs_path *path,
 997                                            u64 bytenr, u64 parent,
 998                                            u64 root_objectid,
 999                                            u64 owner, u64 offset)
1000 {
1001         struct btrfs_key key;
1002         struct btrfs_extent_data_ref *ref;
1003         struct extent_buffer *leaf;
1004         u32 nritems;
1005         int ret;
1006         int recow;
1007         int err = -ENOENT;
1008
1009         key.objectid = bytenr;
1010         if (parent) {
1011                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1012                 key.offset = parent;
1013         } else {
1014                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1015                 key.offset = hash_extent_data_ref(root_objectid,
1016                                                   owner, offset);
1017         }
1018 again:
1019         recow = 0;
1020         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1021         if (ret < 0) {
1022                 err = ret;
1023                 goto fail;
1024         }
1025
1026         if (parent) {
1027                 if (!ret)
1028                         return 0;
1029 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1030                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1031                 btrfs_release_path(root, path);
1032                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1033                 if (ret < 0) {
1034                         err = ret;
1035                         goto fail;
1036                 }
1037                 if (!ret)
1038                         return 0;
1039 #endif
1040                 goto fail;
1041         }
1042
1043         leaf = path->nodes[0];
1044         nritems = btrfs_header_nritems(leaf);
1045         while (1) {
1046                 if (path->slots[0] >= nritems) {
1047                         ret = btrfs_next_leaf(root, path);
1048                         if (ret < 0)
1049                                 err = ret;
1050                         if (ret)
1051                                 goto fail;
1052
1053                         leaf = path->nodes[0];
1054                         nritems = btrfs_header_nritems(leaf);
1055                         recow = 1;
1056                 }
1057
1058                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1059                 if (key.objectid != bytenr ||
1060                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1061                         goto fail;
1062
1063                 ref = btrfs_item_ptr(leaf, path->slots[0],
1064                                      struct btrfs_extent_data_ref);
1065
1066                 if (match_extent_data_ref(leaf, ref, root_objectid,
1067                                           owner, offset)) {
1068                         if (recow) {
1069                                 btrfs_release_path(root, path);
1070                                 goto again;
1071                         }
1072                         err = 0;
1073                         break;
1074                 }
1075                 path->slots[0]++;
1076         }
1077 fail:
1078         return err;
1079 }
1080
1081 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1082                                            struct btrfs_root *root,
1083                                            struct btrfs_path *path,
1084                                            u64 bytenr, u64 parent,
1085                                            u64 root_objectid, u64 owner,
1086                                            u64 offset, int refs_to_add)
1087 {
1088         struct btrfs_key key;
1089         struct extent_buffer *leaf;
1090         u32 size;
1091         u32 num_refs;
1092         int ret;
1093
1094         key.objectid = bytenr;
1095         if (parent) {
1096                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1097                 key.offset = parent;
1098                 size = sizeof(struct btrfs_shared_data_ref);
1099         } else {
1100                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1101                 key.offset = hash_extent_data_ref(root_objectid,
1102                                                   owner, offset);
1103                 size = sizeof(struct btrfs_extent_data_ref);
1104         }
1105
1106         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1107         if (ret && ret != -EEXIST)
1108                 goto fail;
1109
1110         leaf = path->nodes[0];
1111         if (parent) {
1112                 struct btrfs_shared_data_ref *ref;
1113                 ref = btrfs_item_ptr(leaf, path->slots[0],
1114                                      struct btrfs_shared_data_ref);
1115                 if (ret == 0) {
1116                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1117                 } else {
1118                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1119                         num_refs += refs_to_add;
1120                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1121                 }
1122         } else {
1123                 struct btrfs_extent_data_ref *ref;
1124                 while (ret == -EEXIST) {
1125                         ref = btrfs_item_ptr(leaf, path->slots[0],
1126                                              struct btrfs_extent_data_ref);
1127                         if (match_extent_data_ref(leaf, ref, root_objectid,
1128                                                   owner, offset))
1129                                 break;
1130                         btrfs_release_path(root, path);
1131                         key.offset++;
1132                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1133                                                       size);
1134                         if (ret && ret != -EEXIST)
1135                                 goto fail;
1136
1137                         leaf = path->nodes[0];
1138                 }
1139                 ref = btrfs_item_ptr(leaf, path->slots[0],
1140                                      struct btrfs_extent_data_ref);
1141                 if (ret == 0) {
1142                         btrfs_set_extent_data_ref_root(leaf, ref,
1143                                                        root_objectid);
1144                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1145                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1146                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1147                 } else {
1148                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1149                         num_refs += refs_to_add;
1150                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1151                 }
1152         }
1153         btrfs_mark_buffer_dirty(leaf);
1154         ret = 0;
1155 fail:
1156         btrfs_release_path(root, path);
1157         return ret;
1158 }
1159
1160 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1161                                            struct btrfs_root *root,
1162                                            struct btrfs_path *path,
1163                                            int refs_to_drop)
1164 {
1165         struct btrfs_key key;
1166         struct btrfs_extent_data_ref *ref1 = NULL;
1167         struct btrfs_shared_data_ref *ref2 = NULL;
1168         struct extent_buffer *leaf;
1169         u32 num_refs = 0;
1170         int ret = 0;
1171
1172         leaf = path->nodes[0];
1173         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1174
1175         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1176                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1177                                       struct btrfs_extent_data_ref);
1178                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1179         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1180                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1181                                       struct btrfs_shared_data_ref);
1182                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1183 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1184         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1185                 struct btrfs_extent_ref_v0 *ref0;
1186                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1187                                       struct btrfs_extent_ref_v0);
1188                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1189 #endif
1190         } else {
1191                 BUG();
1192         }
1193
1194         BUG_ON(num_refs < refs_to_drop);
1195         num_refs -= refs_to_drop;
1196
1197         if (num_refs == 0) {
1198                 ret = btrfs_del_item(trans, root, path);
1199         } else {
1200                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1201                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1202                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1203                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1204 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1205                 else {
1206                         struct btrfs_extent_ref_v0 *ref0;
1207                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1208                                         struct btrfs_extent_ref_v0);
1209                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1210                 }
1211 #endif
1212                 btrfs_mark_buffer_dirty(leaf);
1213         }
1214         return ret;
1215 }
1216
1217 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1218                                           struct btrfs_path *path,
1219                                           struct btrfs_extent_inline_ref *iref)
1220 {
1221         struct btrfs_key key;
1222         struct extent_buffer *leaf;
1223         struct btrfs_extent_data_ref *ref1;
1224         struct btrfs_shared_data_ref *ref2;
1225         u32 num_refs = 0;
1226
1227         leaf = path->nodes[0];
1228         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1229         if (iref) {
1230                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1231                     BTRFS_EXTENT_DATA_REF_KEY) {
1232                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1233                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1234                 } else {
1235                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1236                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1237                 }
1238         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1239                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1240                                       struct btrfs_extent_data_ref);
1241                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1242         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1243                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1244                                       struct btrfs_shared_data_ref);
1245                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1246 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1247         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1248                 struct btrfs_extent_ref_v0 *ref0;
1249                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1250                                       struct btrfs_extent_ref_v0);
1251                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1252 #endif
1253         } else {
1254                 WARN_ON(1);
1255         }
1256         return num_refs;
1257 }
1258
1259 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1260                                           struct btrfs_root *root,
1261                                           struct btrfs_path *path,
1262                                           u64 bytenr, u64 parent,
1263                                           u64 root_objectid)
1264 {
1265         struct btrfs_key key;
1266         int ret;
1267
1268         key.objectid = bytenr;
1269         if (parent) {
1270                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1271                 key.offset = parent;
1272         } else {
1273                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1274                 key.offset = root_objectid;
1275         }
1276
1277         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1278         if (ret > 0)
1279                 ret = -ENOENT;
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1281         if (ret == -ENOENT && parent) {
1282                 btrfs_release_path(root, path);
1283                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1284                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1285                 if (ret > 0)
1286                         ret = -ENOENT;
1287         }
1288 #endif
1289         return ret;
1290 }
1291
1292 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1293                                           struct btrfs_root *root,
1294                                           struct btrfs_path *path,
1295                                           u64 bytenr, u64 parent,
1296                                           u64 root_objectid)
1297 {
1298         struct btrfs_key key;
1299         int ret;
1300
1301         key.objectid = bytenr;
1302         if (parent) {
1303                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1304                 key.offset = parent;
1305         } else {
1306                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1307                 key.offset = root_objectid;
1308         }
1309
1310         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1311         btrfs_release_path(root, path);
1312         return ret;
1313 }
1314
1315 static inline int extent_ref_type(u64 parent, u64 owner)
1316 {
1317         int type;
1318         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1319                 if (parent > 0)
1320                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1321                 else
1322                         type = BTRFS_TREE_BLOCK_REF_KEY;
1323         } else {
1324                 if (parent > 0)
1325                         type = BTRFS_SHARED_DATA_REF_KEY;
1326                 else
1327                         type = BTRFS_EXTENT_DATA_REF_KEY;
1328         }
1329         return type;
1330 }
1331
1332 static int find_next_key(struct btrfs_path *path, int level,
1333                          struct btrfs_key *key)
1334
1335 {
1336         for (; level < BTRFS_MAX_LEVEL; level++) {
1337                 if (!path->nodes[level])
1338                         break;
1339                 if (path->slots[level] + 1 >=
1340                     btrfs_header_nritems(path->nodes[level]))
1341                         continue;
1342                 if (level == 0)
1343                         btrfs_item_key_to_cpu(path->nodes[level], key,
1344                                               path->slots[level] + 1);
1345                 else
1346                         btrfs_node_key_to_cpu(path->nodes[level], key,
1347                                               path->slots[level] + 1);
1348                 return 0;
1349         }
1350         return 1;
1351 }
1352
1353 /*
1354  * look for inline back ref. if back ref is found, *ref_ret is set
1355  * to the address of inline back ref, and 0 is returned.
1356  *
1357  * if back ref isn't found, *ref_ret is set to the address where it
1358  * should be inserted, and -ENOENT is returned.
1359  *
1360  * if insert is true and there are too many inline back refs, the path
1361  * points to the extent item, and -EAGAIN is returned.
1362  *
1363  * NOTE: inline back refs are ordered in the same way that back ref
1364  *       items in the tree are ordered.
1365  */
1366 static noinline_for_stack
1367 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1368                                  struct btrfs_root *root,
1369                                  struct btrfs_path *path,
1370                                  struct btrfs_extent_inline_ref **ref_ret,
1371                                  u64 bytenr, u64 num_bytes,
1372                                  u64 parent, u64 root_objectid,
1373                                  u64 owner, u64 offset, int insert)
1374 {
1375         struct btrfs_key key;
1376         struct extent_buffer *leaf;
1377         struct btrfs_extent_item *ei;
1378         struct btrfs_extent_inline_ref *iref;
1379         u64 flags;
1380         u64 item_size;
1381         unsigned long ptr;
1382         unsigned long end;
1383         int extra_size;
1384         int type;
1385         int want;
1386         int ret;
1387         int err = 0;
1388
1389         key.objectid = bytenr;
1390         key.type = BTRFS_EXTENT_ITEM_KEY;
1391         key.offset = num_bytes;
1392
1393         want = extent_ref_type(parent, owner);
1394         if (insert) {
1395                 extra_size = btrfs_extent_inline_ref_size(want);
1396                 path->keep_locks = 1;
1397         } else
1398                 extra_size = -1;
1399         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1400         if (ret < 0) {
1401                 err = ret;
1402                 goto out;
1403         }
1404         BUG_ON(ret);
1405
1406         leaf = path->nodes[0];
1407         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1408 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1409         if (item_size < sizeof(*ei)) {
1410                 if (!insert) {
1411                         err = -ENOENT;
1412                         goto out;
1413                 }
1414                 ret = convert_extent_item_v0(trans, root, path, owner,
1415                                              extra_size);
1416                 if (ret < 0) {
1417                         err = ret;
1418                         goto out;
1419                 }
1420                 leaf = path->nodes[0];
1421                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1422         }
1423 #endif
1424         BUG_ON(item_size < sizeof(*ei));
1425
1426         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1427         flags = btrfs_extent_flags(leaf, ei);
1428
1429         ptr = (unsigned long)(ei + 1);
1430         end = (unsigned long)ei + item_size;
1431
1432         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1433                 ptr += sizeof(struct btrfs_tree_block_info);
1434                 BUG_ON(ptr > end);
1435         } else {
1436                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1437         }
1438
1439         err = -ENOENT;
1440         while (1) {
1441                 if (ptr >= end) {
1442                         WARN_ON(ptr > end);
1443                         break;
1444                 }
1445                 iref = (struct btrfs_extent_inline_ref *)ptr;
1446                 type = btrfs_extent_inline_ref_type(leaf, iref);
1447                 if (want < type)
1448                         break;
1449                 if (want > type) {
1450                         ptr += btrfs_extent_inline_ref_size(type);
1451                         continue;
1452                 }
1453
1454                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1455                         struct btrfs_extent_data_ref *dref;
1456                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1457                         if (match_extent_data_ref(leaf, dref, root_objectid,
1458                                                   owner, offset)) {
1459                                 err = 0;
1460                                 break;
1461                         }
1462                         if (hash_extent_data_ref_item(leaf, dref) <
1463                             hash_extent_data_ref(root_objectid, owner, offset))
1464                                 break;
1465                 } else {
1466                         u64 ref_offset;
1467                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1468                         if (parent > 0) {
1469                                 if (parent == ref_offset) {
1470                                         err = 0;
1471                                         break;
1472                                 }
1473                                 if (ref_offset < parent)
1474                                         break;
1475                         } else {
1476                                 if (root_objectid == ref_offset) {
1477                                         err = 0;
1478                                         break;
1479                                 }
1480                                 if (ref_offset < root_objectid)
1481                                         break;
1482                         }
1483                 }
1484                 ptr += btrfs_extent_inline_ref_size(type);
1485         }
1486         if (err == -ENOENT && insert) {
1487                 if (item_size + extra_size >=
1488                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1489                         err = -EAGAIN;
1490                         goto out;
1491                 }
1492                 /*
1493                  * To add new inline back ref, we have to make sure
1494                  * there is no corresponding back ref item.
1495                  * For simplicity, we just do not add new inline back
1496                  * ref if there is any kind of item for this block
1497                  */
1498                 if (find_next_key(path, 0, &key) == 0 &&
1499                     key.objectid == bytenr &&
1500                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1501                         err = -EAGAIN;
1502                         goto out;
1503                 }
1504         }
1505         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1506 out:
1507         if (insert) {
1508                 path->keep_locks = 0;
1509                 btrfs_unlock_up_safe(path, 1);
1510         }
1511         return err;
1512 }
1513
1514 /*
1515  * helper to add new inline back ref
1516  */
1517 static noinline_for_stack
1518 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1519                                 struct btrfs_root *root,
1520                                 struct btrfs_path *path,
1521                                 struct btrfs_extent_inline_ref *iref,
1522                                 u64 parent, u64 root_objectid,
1523                                 u64 owner, u64 offset, int refs_to_add,
1524                                 struct btrfs_delayed_extent_op *extent_op)
1525 {
1526         struct extent_buffer *leaf;
1527         struct btrfs_extent_item *ei;
1528         unsigned long ptr;
1529         unsigned long end;
1530         unsigned long item_offset;
1531         u64 refs;
1532         int size;
1533         int type;
1534         int ret;
1535
1536         leaf = path->nodes[0];
1537         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1538         item_offset = (unsigned long)iref - (unsigned long)ei;
1539
1540         type = extent_ref_type(parent, owner);
1541         size = btrfs_extent_inline_ref_size(type);
1542
1543         ret = btrfs_extend_item(trans, root, path, size);
1544         BUG_ON(ret);
1545
1546         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1547         refs = btrfs_extent_refs(leaf, ei);
1548         refs += refs_to_add;
1549         btrfs_set_extent_refs(leaf, ei, refs);
1550         if (extent_op)
1551                 __run_delayed_extent_op(extent_op, leaf, ei);
1552
1553         ptr = (unsigned long)ei + item_offset;
1554         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1555         if (ptr < end - size)
1556                 memmove_extent_buffer(leaf, ptr + size, ptr,
1557                                       end - size - ptr);
1558
1559         iref = (struct btrfs_extent_inline_ref *)ptr;
1560         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1561         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1562                 struct btrfs_extent_data_ref *dref;
1563                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1564                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1565                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1566                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1567                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1568         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1569                 struct btrfs_shared_data_ref *sref;
1570                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1571                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1572                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1573         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1574                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1575         } else {
1576                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1577         }
1578         btrfs_mark_buffer_dirty(leaf);
1579         return 0;
1580 }
1581
1582 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1583                                  struct btrfs_root *root,
1584                                  struct btrfs_path *path,
1585                                  struct btrfs_extent_inline_ref **ref_ret,
1586                                  u64 bytenr, u64 num_bytes, u64 parent,
1587                                  u64 root_objectid, u64 owner, u64 offset)
1588 {
1589         int ret;
1590
1591         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1592                                            bytenr, num_bytes, parent,
1593                                            root_objectid, owner, offset, 0);
1594         if (ret != -ENOENT)
1595                 return ret;
1596
1597         btrfs_release_path(root, path);
1598         *ref_ret = NULL;
1599
1600         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1601                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1602                                             root_objectid);
1603         } else {
1604                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1605                                              root_objectid, owner, offset);
1606         }
1607         return ret;
1608 }
1609
1610 /*
1611  * helper to update/remove inline back ref
1612  */
1613 static noinline_for_stack
1614 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1615                                  struct btrfs_root *root,
1616                                  struct btrfs_path *path,
1617                                  struct btrfs_extent_inline_ref *iref,
1618                                  int refs_to_mod,
1619                                  struct btrfs_delayed_extent_op *extent_op)
1620 {
1621         struct extent_buffer *leaf;
1622         struct btrfs_extent_item *ei;
1623         struct btrfs_extent_data_ref *dref = NULL;
1624         struct btrfs_shared_data_ref *sref = NULL;
1625         unsigned long ptr;
1626         unsigned long end;
1627         u32 item_size;
1628         int size;
1629         int type;
1630         int ret;
1631         u64 refs;
1632
1633         leaf = path->nodes[0];
1634         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1635         refs = btrfs_extent_refs(leaf, ei);
1636         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1637         refs += refs_to_mod;
1638         btrfs_set_extent_refs(leaf, ei, refs);
1639         if (extent_op)
1640                 __run_delayed_extent_op(extent_op, leaf, ei);
1641
1642         type = btrfs_extent_inline_ref_type(leaf, iref);
1643
1644         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1645                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1646                 refs = btrfs_extent_data_ref_count(leaf, dref);
1647         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1648                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1649                 refs = btrfs_shared_data_ref_count(leaf, sref);
1650         } else {
1651                 refs = 1;
1652                 BUG_ON(refs_to_mod != -1);
1653         }
1654
1655         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1656         refs += refs_to_mod;
1657
1658         if (refs > 0) {
1659                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1660                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1661                 else
1662                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1663         } else {
1664                 size =  btrfs_extent_inline_ref_size(type);
1665                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1666                 ptr = (unsigned long)iref;
1667                 end = (unsigned long)ei + item_size;
1668                 if (ptr + size < end)
1669                         memmove_extent_buffer(leaf, ptr, ptr + size,
1670                                               end - ptr - size);
1671                 item_size -= size;
1672                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1673                 BUG_ON(ret);
1674         }
1675         btrfs_mark_buffer_dirty(leaf);
1676         return 0;
1677 }
1678
1679 static noinline_for_stack
1680 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1681                                  struct btrfs_root *root,
1682                                  struct btrfs_path *path,
1683                                  u64 bytenr, u64 num_bytes, u64 parent,
1684                                  u64 root_objectid, u64 owner,
1685                                  u64 offset, int refs_to_add,
1686                                  struct btrfs_delayed_extent_op *extent_op)
1687 {
1688         struct btrfs_extent_inline_ref *iref;
1689         int ret;
1690
1691         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1692                                            bytenr, num_bytes, parent,
1693                                            root_objectid, owner, offset, 1);
1694         if (ret == 0) {
1695                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1696                 ret = update_inline_extent_backref(trans, root, path, iref,
1697                                                    refs_to_add, extent_op);
1698         } else if (ret == -ENOENT) {
1699                 ret = setup_inline_extent_backref(trans, root, path, iref,
1700                                                   parent, root_objectid,
1701                                                   owner, offset, refs_to_add,
1702                                                   extent_op);
1703         }
1704         return ret;
1705 }
1706
1707 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1708                                  struct btrfs_root *root,
1709                                  struct btrfs_path *path,
1710                                  u64 bytenr, u64 parent, u64 root_objectid,
1711                                  u64 owner, u64 offset, int refs_to_add)
1712 {
1713         int ret;
1714         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1715                 BUG_ON(refs_to_add != 1);
1716                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1717                                             parent, root_objectid);
1718         } else {
1719                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1720                                              parent, root_objectid,
1721                                              owner, offset, refs_to_add);
1722         }
1723         return ret;
1724 }
1725
1726 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1727                                  struct btrfs_root *root,
1728                                  struct btrfs_path *path,
1729                                  struct btrfs_extent_inline_ref *iref,
1730                                  int refs_to_drop, int is_data)
1731 {
1732         int ret;
1733
1734         BUG_ON(!is_data && refs_to_drop != 1);
1735         if (iref) {
1736                 ret = update_inline_extent_backref(trans, root, path, iref,
1737                                                    -refs_to_drop, NULL);
1738         } else if (is_data) {
1739                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1740         } else {
1741                 ret = btrfs_del_item(trans, root, path);
1742         }
1743         return ret;
1744 }
1745
1746 static void btrfs_issue_discard(struct block_device *bdev,
1747                                 u64 start, u64 len)
1748 {
1749         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL, 0);
1750 }
1751
1752 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1753                                 u64 num_bytes)
1754 {
1755         int ret;
1756         u64 map_length = num_bytes;
1757         struct btrfs_multi_bio *multi = NULL;
1758
1759         if (!btrfs_test_opt(root, DISCARD))
1760                 return 0;
1761
1762         /* Tell the block device(s) that the sectors can be discarded */
1763         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1764                               bytenr, &map_length, &multi, 0);
1765         if (!ret) {
1766                 struct btrfs_bio_stripe *stripe = multi->stripes;
1767                 int i;
1768
1769                 if (map_length > num_bytes)
1770                         map_length = num_bytes;
1771
1772                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1773                         btrfs_issue_discard(stripe->dev->bdev,
1774                                             stripe->physical,
1775                                             map_length);
1776                 }
1777                 kfree(multi);
1778         }
1779
1780         return ret;
1781 }
1782
1783 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1784                          struct btrfs_root *root,
1785                          u64 bytenr, u64 num_bytes, u64 parent,
1786                          u64 root_objectid, u64 owner, u64 offset)
1787 {
1788         int ret;
1789         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1790                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1791
1792         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1793                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1794                                         parent, root_objectid, (int)owner,
1795                                         BTRFS_ADD_DELAYED_REF, NULL);
1796         } else {
1797                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1798                                         parent, root_objectid, owner, offset,
1799                                         BTRFS_ADD_DELAYED_REF, NULL);
1800         }
1801         return ret;
1802 }
1803
1804 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1805                                   struct btrfs_root *root,
1806                                   u64 bytenr, u64 num_bytes,
1807                                   u64 parent, u64 root_objectid,
1808                                   u64 owner, u64 offset, int refs_to_add,
1809                                   struct btrfs_delayed_extent_op *extent_op)
1810 {
1811         struct btrfs_path *path;
1812         struct extent_buffer *leaf;
1813         struct btrfs_extent_item *item;
1814         u64 refs;
1815         int ret;
1816         int err = 0;
1817
1818         path = btrfs_alloc_path();
1819         if (!path)
1820                 return -ENOMEM;
1821
1822         path->reada = 1;
1823         path->leave_spinning = 1;
1824         /* this will setup the path even if it fails to insert the back ref */
1825         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1826                                            path, bytenr, num_bytes, parent,
1827                                            root_objectid, owner, offset,
1828                                            refs_to_add, extent_op);
1829         if (ret == 0)
1830                 goto out;
1831
1832         if (ret != -EAGAIN) {
1833                 err = ret;
1834                 goto out;
1835         }
1836
1837         leaf = path->nodes[0];
1838         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1839         refs = btrfs_extent_refs(leaf, item);
1840         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1841         if (extent_op)
1842                 __run_delayed_extent_op(extent_op, leaf, item);
1843
1844         btrfs_mark_buffer_dirty(leaf);
1845         btrfs_release_path(root->fs_info->extent_root, path);
1846
1847         path->reada = 1;
1848         path->leave_spinning = 1;
1849
1850         /* now insert the actual backref */
1851         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1852                                     path, bytenr, parent, root_objectid,
1853                                     owner, offset, refs_to_add);
1854         BUG_ON(ret);
1855 out:
1856         btrfs_free_path(path);
1857         return err;
1858 }
1859
1860 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1861                                 struct btrfs_root *root,
1862                                 struct btrfs_delayed_ref_node *node,
1863                                 struct btrfs_delayed_extent_op *extent_op,
1864                                 int insert_reserved)
1865 {
1866         int ret = 0;
1867         struct btrfs_delayed_data_ref *ref;
1868         struct btrfs_key ins;
1869         u64 parent = 0;
1870         u64 ref_root = 0;
1871         u64 flags = 0;
1872
1873         ins.objectid = node->bytenr;
1874         ins.offset = node->num_bytes;
1875         ins.type = BTRFS_EXTENT_ITEM_KEY;
1876
1877         ref = btrfs_delayed_node_to_data_ref(node);
1878         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1879                 parent = ref->parent;
1880         else
1881                 ref_root = ref->root;
1882
1883         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1884                 if (extent_op) {
1885                         BUG_ON(extent_op->update_key);
1886                         flags |= extent_op->flags_to_set;
1887                 }
1888                 ret = alloc_reserved_file_extent(trans, root,
1889                                                  parent, ref_root, flags,
1890                                                  ref->objectid, ref->offset,
1891                                                  &ins, node->ref_mod);
1892         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1893                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1894                                              node->num_bytes, parent,
1895                                              ref_root, ref->objectid,
1896                                              ref->offset, node->ref_mod,
1897                                              extent_op);
1898         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1899                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1900                                           node->num_bytes, parent,
1901                                           ref_root, ref->objectid,
1902                                           ref->offset, node->ref_mod,
1903                                           extent_op);
1904         } else {
1905                 BUG();
1906         }
1907         return ret;
1908 }
1909
1910 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1911                                     struct extent_buffer *leaf,
1912                                     struct btrfs_extent_item *ei)
1913 {
1914         u64 flags = btrfs_extent_flags(leaf, ei);
1915         if (extent_op->update_flags) {
1916                 flags |= extent_op->flags_to_set;
1917                 btrfs_set_extent_flags(leaf, ei, flags);
1918         }
1919
1920         if (extent_op->update_key) {
1921                 struct btrfs_tree_block_info *bi;
1922                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1923                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1924                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1925         }
1926 }
1927
1928 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1929                                  struct btrfs_root *root,
1930                                  struct btrfs_delayed_ref_node *node,
1931                                  struct btrfs_delayed_extent_op *extent_op)
1932 {
1933         struct btrfs_key key;
1934         struct btrfs_path *path;
1935         struct btrfs_extent_item *ei;
1936         struct extent_buffer *leaf;
1937         u32 item_size;
1938         int ret;
1939         int err = 0;
1940
1941         path = btrfs_alloc_path();
1942         if (!path)
1943                 return -ENOMEM;
1944
1945         key.objectid = node->bytenr;
1946         key.type = BTRFS_EXTENT_ITEM_KEY;
1947         key.offset = node->num_bytes;
1948
1949         path->reada = 1;
1950         path->leave_spinning = 1;
1951         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1952                                 path, 0, 1);
1953         if (ret < 0) {
1954                 err = ret;
1955                 goto out;
1956         }
1957         if (ret > 0) {
1958                 err = -EIO;
1959                 goto out;
1960         }
1961
1962         leaf = path->nodes[0];
1963         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1964 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1965         if (item_size < sizeof(*ei)) {
1966                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1967                                              path, (u64)-1, 0);
1968                 if (ret < 0) {
1969                         err = ret;
1970                         goto out;
1971                 }
1972                 leaf = path->nodes[0];
1973                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1974         }
1975 #endif
1976         BUG_ON(item_size < sizeof(*ei));
1977         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1978         __run_delayed_extent_op(extent_op, leaf, ei);
1979
1980         btrfs_mark_buffer_dirty(leaf);
1981 out:
1982         btrfs_free_path(path);
1983         return err;
1984 }
1985
1986 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1987                                 struct btrfs_root *root,
1988                                 struct btrfs_delayed_ref_node *node,
1989                                 struct btrfs_delayed_extent_op *extent_op,
1990                                 int insert_reserved)
1991 {
1992         int ret = 0;
1993         struct btrfs_delayed_tree_ref *ref;
1994         struct btrfs_key ins;
1995         u64 parent = 0;
1996         u64 ref_root = 0;
1997
1998         ins.objectid = node->bytenr;
1999         ins.offset = node->num_bytes;
2000         ins.type = BTRFS_EXTENT_ITEM_KEY;
2001
2002         ref = btrfs_delayed_node_to_tree_ref(node);
2003         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2004                 parent = ref->parent;
2005         else
2006                 ref_root = ref->root;
2007
2008         BUG_ON(node->ref_mod != 1);
2009         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2010                 BUG_ON(!extent_op || !extent_op->update_flags ||
2011                        !extent_op->update_key);
2012                 ret = alloc_reserved_tree_block(trans, root,
2013                                                 parent, ref_root,
2014                                                 extent_op->flags_to_set,
2015                                                 &extent_op->key,
2016                                                 ref->level, &ins);
2017         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2018                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2019                                              node->num_bytes, parent, ref_root,
2020                                              ref->level, 0, 1, extent_op);
2021         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2022                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2023                                           node->num_bytes, parent, ref_root,
2024                                           ref->level, 0, 1, extent_op);
2025         } else {
2026                 BUG();
2027         }
2028         return ret;
2029 }
2030
2031 /* helper function to actually process a single delayed ref entry */
2032 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2033                                struct btrfs_root *root,
2034                                struct btrfs_delayed_ref_node *node,
2035                                struct btrfs_delayed_extent_op *extent_op,
2036                                int insert_reserved)
2037 {
2038         int ret;
2039         if (btrfs_delayed_ref_is_head(node)) {
2040                 struct btrfs_delayed_ref_head *head;
2041                 /*
2042                  * we've hit the end of the chain and we were supposed
2043                  * to insert this extent into the tree.  But, it got
2044                  * deleted before we ever needed to insert it, so all
2045                  * we have to do is clean up the accounting
2046                  */
2047                 BUG_ON(extent_op);
2048                 head = btrfs_delayed_node_to_head(node);
2049                 if (insert_reserved) {
2050                         btrfs_pin_extent(root, node->bytenr,
2051                                          node->num_bytes, 1);
2052                         if (head->is_data) {
2053                                 ret = btrfs_del_csums(trans, root,
2054                                                       node->bytenr,
2055                                                       node->num_bytes);
2056                                 BUG_ON(ret);
2057                         }
2058                 }
2059                 mutex_unlock(&head->mutex);
2060                 return 0;
2061         }
2062
2063         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2064             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2065                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2066                                            insert_reserved);
2067         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2068                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2069                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2070                                            insert_reserved);
2071         else
2072                 BUG();
2073         return ret;
2074 }
2075
2076 static noinline struct btrfs_delayed_ref_node *
2077 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2078 {
2079         struct rb_node *node;
2080         struct btrfs_delayed_ref_node *ref;
2081         int action = BTRFS_ADD_DELAYED_REF;
2082 again:
2083         /*
2084          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2085          * this prevents ref count from going down to zero when
2086          * there still are pending delayed ref.
2087          */
2088         node = rb_prev(&head->node.rb_node);
2089         while (1) {
2090                 if (!node)
2091                         break;
2092                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2093                                 rb_node);
2094                 if (ref->bytenr != head->node.bytenr)
2095                         break;
2096                 if (ref->action == action)
2097                         return ref;
2098                 node = rb_prev(node);
2099         }
2100         if (action == BTRFS_ADD_DELAYED_REF) {
2101                 action = BTRFS_DROP_DELAYED_REF;
2102                 goto again;
2103         }
2104         return NULL;
2105 }
2106
2107 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2108                                        struct btrfs_root *root,
2109                                        struct list_head *cluster)
2110 {
2111         struct btrfs_delayed_ref_root *delayed_refs;
2112         struct btrfs_delayed_ref_node *ref;
2113         struct btrfs_delayed_ref_head *locked_ref = NULL;
2114         struct btrfs_delayed_extent_op *extent_op;
2115         int ret;
2116         int count = 0;
2117         int must_insert_reserved = 0;
2118
2119         delayed_refs = &trans->transaction->delayed_refs;
2120         while (1) {
2121                 if (!locked_ref) {
2122                         /* pick a new head ref from the cluster list */
2123                         if (list_empty(cluster))
2124                                 break;
2125
2126                         locked_ref = list_entry(cluster->next,
2127                                      struct btrfs_delayed_ref_head, cluster);
2128
2129                         /* grab the lock that says we are going to process
2130                          * all the refs for this head */
2131                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2132
2133                         /*
2134                          * we may have dropped the spin lock to get the head
2135                          * mutex lock, and that might have given someone else
2136                          * time to free the head.  If that's true, it has been
2137                          * removed from our list and we can move on.
2138                          */
2139                         if (ret == -EAGAIN) {
2140                                 locked_ref = NULL;
2141                                 count++;
2142                                 continue;
2143                         }
2144                 }
2145
2146                 /*
2147                  * record the must insert reserved flag before we
2148                  * drop the spin lock.
2149                  */
2150                 must_insert_reserved = locked_ref->must_insert_reserved;
2151                 locked_ref->must_insert_reserved = 0;
2152
2153                 extent_op = locked_ref->extent_op;
2154                 locked_ref->extent_op = NULL;
2155
2156                 /*
2157                  * locked_ref is the head node, so we have to go one
2158                  * node back for any delayed ref updates
2159                  */
2160                 ref = select_delayed_ref(locked_ref);
2161                 if (!ref) {
2162                         /* All delayed refs have been processed, Go ahead
2163                          * and send the head node to run_one_delayed_ref,
2164                          * so that any accounting fixes can happen
2165                          */
2166                         ref = &locked_ref->node;
2167
2168                         if (extent_op && must_insert_reserved) {
2169                                 kfree(extent_op);
2170                                 extent_op = NULL;
2171                         }
2172
2173                         if (extent_op) {
2174                                 spin_unlock(&delayed_refs->lock);
2175
2176                                 ret = run_delayed_extent_op(trans, root,
2177                                                             ref, extent_op);
2178                                 BUG_ON(ret);
2179                                 kfree(extent_op);
2180
2181                                 cond_resched();
2182                                 spin_lock(&delayed_refs->lock);
2183                                 continue;
2184                         }
2185
2186                         list_del_init(&locked_ref->cluster);
2187                         locked_ref = NULL;
2188                 }
2189
2190                 ref->in_tree = 0;
2191                 rb_erase(&ref->rb_node, &delayed_refs->root);
2192                 delayed_refs->num_entries--;
2193
2194                 spin_unlock(&delayed_refs->lock);
2195
2196                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2197                                           must_insert_reserved);
2198                 BUG_ON(ret);
2199
2200                 btrfs_put_delayed_ref(ref);
2201                 kfree(extent_op);
2202                 count++;
2203
2204                 cond_resched();
2205                 spin_lock(&delayed_refs->lock);
2206         }
2207         return count;
2208 }
2209
2210 /*
2211  * this starts processing the delayed reference count updates and
2212  * extent insertions we have queued up so far.  count can be
2213  * 0, which means to process everything in the tree at the start
2214  * of the run (but not newly added entries), or it can be some target
2215  * number you'd like to process.
2216  */
2217 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2218                            struct btrfs_root *root, unsigned long count)
2219 {
2220         struct rb_node *node;
2221         struct btrfs_delayed_ref_root *delayed_refs;
2222         struct btrfs_delayed_ref_node *ref;
2223         struct list_head cluster;
2224         int ret;
2225         int run_all = count == (unsigned long)-1;
2226         int run_most = 0;
2227
2228         if (root == root->fs_info->extent_root)
2229                 root = root->fs_info->tree_root;
2230
2231         delayed_refs = &trans->transaction->delayed_refs;
2232         INIT_LIST_HEAD(&cluster);
2233 again:
2234         spin_lock(&delayed_refs->lock);
2235         if (count == 0) {
2236                 count = delayed_refs->num_entries * 2;
2237                 run_most = 1;
2238         }
2239         while (1) {
2240                 if (!(run_all || run_most) &&
2241                     delayed_refs->num_heads_ready < 64)
2242                         break;
2243
2244                 /*
2245                  * go find something we can process in the rbtree.  We start at
2246                  * the beginning of the tree, and then build a cluster
2247                  * of refs to process starting at the first one we are able to
2248                  * lock
2249                  */
2250                 ret = btrfs_find_ref_cluster(trans, &cluster,
2251                                              delayed_refs->run_delayed_start);
2252                 if (ret)
2253                         break;
2254
2255                 ret = run_clustered_refs(trans, root, &cluster);
2256                 BUG_ON(ret < 0);
2257
2258                 count -= min_t(unsigned long, ret, count);
2259
2260                 if (count == 0)
2261                         break;
2262         }
2263
2264         if (run_all) {
2265                 node = rb_first(&delayed_refs->root);
2266                 if (!node)
2267                         goto out;
2268                 count = (unsigned long)-1;
2269
2270                 while (node) {
2271                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2272                                        rb_node);
2273                         if (btrfs_delayed_ref_is_head(ref)) {
2274                                 struct btrfs_delayed_ref_head *head;
2275
2276                                 head = btrfs_delayed_node_to_head(ref);
2277                                 atomic_inc(&ref->refs);
2278
2279                                 spin_unlock(&delayed_refs->lock);
2280                                 mutex_lock(&head->mutex);
2281                                 mutex_unlock(&head->mutex);
2282
2283                                 btrfs_put_delayed_ref(ref);
2284                                 cond_resched();
2285                                 goto again;
2286                         }
2287                         node = rb_next(node);
2288                 }
2289                 spin_unlock(&delayed_refs->lock);
2290                 schedule_timeout(1);
2291                 goto again;
2292         }
2293 out:
2294         spin_unlock(&delayed_refs->lock);
2295         return 0;
2296 }
2297
2298 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2299                                 struct btrfs_root *root,
2300                                 u64 bytenr, u64 num_bytes, u64 flags,
2301                                 int is_data)
2302 {
2303         struct btrfs_delayed_extent_op *extent_op;
2304         int ret;
2305
2306         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2307         if (!extent_op)
2308                 return -ENOMEM;
2309
2310         extent_op->flags_to_set = flags;
2311         extent_op->update_flags = 1;
2312         extent_op->update_key = 0;
2313         extent_op->is_data = is_data ? 1 : 0;
2314
2315         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2316         if (ret)
2317                 kfree(extent_op);
2318         return ret;
2319 }
2320
2321 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2322                                       struct btrfs_root *root,
2323                                       struct btrfs_path *path,
2324                                       u64 objectid, u64 offset, u64 bytenr)
2325 {
2326         struct btrfs_delayed_ref_head *head;
2327         struct btrfs_delayed_ref_node *ref;
2328         struct btrfs_delayed_data_ref *data_ref;
2329         struct btrfs_delayed_ref_root *delayed_refs;
2330         struct rb_node *node;
2331         int ret = 0;
2332
2333         ret = -ENOENT;
2334         delayed_refs = &trans->transaction->delayed_refs;
2335         spin_lock(&delayed_refs->lock);
2336         head = btrfs_find_delayed_ref_head(trans, bytenr);
2337         if (!head)
2338                 goto out;
2339
2340         if (!mutex_trylock(&head->mutex)) {
2341                 atomic_inc(&head->node.refs);
2342                 spin_unlock(&delayed_refs->lock);
2343
2344                 btrfs_release_path(root->fs_info->extent_root, path);
2345
2346                 mutex_lock(&head->mutex);
2347                 mutex_unlock(&head->mutex);
2348                 btrfs_put_delayed_ref(&head->node);
2349                 return -EAGAIN;
2350         }
2351
2352         node = rb_prev(&head->node.rb_node);
2353         if (!node)
2354                 goto out_unlock;
2355
2356         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2357
2358         if (ref->bytenr != bytenr)
2359                 goto out_unlock;
2360
2361         ret = 1;
2362         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2363                 goto out_unlock;
2364
2365         data_ref = btrfs_delayed_node_to_data_ref(ref);
2366
2367         node = rb_prev(node);
2368         if (node) {
2369                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2370                 if (ref->bytenr == bytenr)
2371                         goto out_unlock;
2372         }
2373
2374         if (data_ref->root != root->root_key.objectid ||
2375             data_ref->objectid != objectid || data_ref->offset != offset)
2376                 goto out_unlock;
2377
2378         ret = 0;
2379 out_unlock:
2380         mutex_unlock(&head->mutex);
2381 out:
2382         spin_unlock(&delayed_refs->lock);
2383         return ret;
2384 }
2385
2386 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2387                                         struct btrfs_root *root,
2388                                         struct btrfs_path *path,
2389                                         u64 objectid, u64 offset, u64 bytenr)
2390 {
2391         struct btrfs_root *extent_root = root->fs_info->extent_root;
2392         struct extent_buffer *leaf;
2393         struct btrfs_extent_data_ref *ref;
2394         struct btrfs_extent_inline_ref *iref;
2395         struct btrfs_extent_item *ei;
2396         struct btrfs_key key;
2397         u32 item_size;
2398         int ret;
2399
2400         key.objectid = bytenr;
2401         key.offset = (u64)-1;
2402         key.type = BTRFS_EXTENT_ITEM_KEY;
2403
2404         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2405         if (ret < 0)
2406                 goto out;
2407         BUG_ON(ret == 0);
2408
2409         ret = -ENOENT;
2410         if (path->slots[0] == 0)
2411                 goto out;
2412
2413         path->slots[0]--;
2414         leaf = path->nodes[0];
2415         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2416
2417         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2418                 goto out;
2419
2420         ret = 1;
2421         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2422 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2423         if (item_size < sizeof(*ei)) {
2424                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2425                 goto out;
2426         }
2427 #endif
2428         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2429
2430         if (item_size != sizeof(*ei) +
2431             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2432                 goto out;
2433
2434         if (btrfs_extent_generation(leaf, ei) <=
2435             btrfs_root_last_snapshot(&root->root_item))
2436                 goto out;
2437
2438         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2439         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2440             BTRFS_EXTENT_DATA_REF_KEY)
2441                 goto out;
2442
2443         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2444         if (btrfs_extent_refs(leaf, ei) !=
2445             btrfs_extent_data_ref_count(leaf, ref) ||
2446             btrfs_extent_data_ref_root(leaf, ref) !=
2447             root->root_key.objectid ||
2448             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2449             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2450                 goto out;
2451
2452         ret = 0;
2453 out:
2454         return ret;
2455 }
2456
2457 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2458                           struct btrfs_root *root,
2459                           u64 objectid, u64 offset, u64 bytenr)
2460 {
2461         struct btrfs_path *path;
2462         int ret;
2463         int ret2;
2464
2465         path = btrfs_alloc_path();
2466         if (!path)
2467                 return -ENOENT;
2468
2469         do {
2470                 ret = check_committed_ref(trans, root, path, objectid,
2471                                           offset, bytenr);
2472                 if (ret && ret != -ENOENT)
2473                         goto out;
2474
2475                 ret2 = check_delayed_ref(trans, root, path, objectid,
2476                                          offset, bytenr);
2477         } while (ret2 == -EAGAIN);
2478
2479         if (ret2 && ret2 != -ENOENT) {
2480                 ret = ret2;
2481                 goto out;
2482         }
2483
2484         if (ret != -ENOENT || ret2 != -ENOENT)
2485                 ret = 0;
2486 out:
2487         btrfs_free_path(path);
2488         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2489                 WARN_ON(ret > 0);
2490         return ret;
2491 }
2492
2493 #if 0
2494 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2495                     struct extent_buffer *buf, u32 nr_extents)
2496 {
2497         struct btrfs_key key;
2498         struct btrfs_file_extent_item *fi;
2499         u64 root_gen;
2500         u32 nritems;
2501         int i;
2502         int level;
2503         int ret = 0;
2504         int shared = 0;
2505
2506         if (!root->ref_cows)
2507                 return 0;
2508
2509         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2510                 shared = 0;
2511                 root_gen = root->root_key.offset;
2512         } else {
2513                 shared = 1;
2514                 root_gen = trans->transid - 1;
2515         }
2516
2517         level = btrfs_header_level(buf);
2518         nritems = btrfs_header_nritems(buf);
2519
2520         if (level == 0) {
2521                 struct btrfs_leaf_ref *ref;
2522                 struct btrfs_extent_info *info;
2523
2524                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2525                 if (!ref) {
2526                         ret = -ENOMEM;
2527                         goto out;
2528                 }
2529
2530                 ref->root_gen = root_gen;
2531                 ref->bytenr = buf->start;
2532                 ref->owner = btrfs_header_owner(buf);
2533                 ref->generation = btrfs_header_generation(buf);
2534                 ref->nritems = nr_extents;
2535                 info = ref->extents;
2536
2537                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2538                         u64 disk_bytenr;
2539                         btrfs_item_key_to_cpu(buf, &key, i);
2540                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2541                                 continue;
2542                         fi = btrfs_item_ptr(buf, i,
2543                                             struct btrfs_file_extent_item);
2544                         if (btrfs_file_extent_type(buf, fi) ==
2545                             BTRFS_FILE_EXTENT_INLINE)
2546                                 continue;
2547                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2548                         if (disk_bytenr == 0)
2549                                 continue;
2550
2551                         info->bytenr = disk_bytenr;
2552                         info->num_bytes =
2553                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2554                         info->objectid = key.objectid;
2555                         info->offset = key.offset;
2556                         info++;
2557                 }
2558
2559                 ret = btrfs_add_leaf_ref(root, ref, shared);
2560                 if (ret == -EEXIST && shared) {
2561                         struct btrfs_leaf_ref *old;
2562                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2563                         BUG_ON(!old);
2564                         btrfs_remove_leaf_ref(root, old);
2565                         btrfs_free_leaf_ref(root, old);
2566                         ret = btrfs_add_leaf_ref(root, ref, shared);
2567                 }
2568                 WARN_ON(ret);
2569                 btrfs_free_leaf_ref(root, ref);
2570         }
2571 out:
2572         return ret;
2573 }
2574
2575 /* when a block goes through cow, we update the reference counts of
2576  * everything that block points to.  The internal pointers of the block
2577  * can be in just about any order, and it is likely to have clusters of
2578  * things that are close together and clusters of things that are not.
2579  *
2580  * To help reduce the seeks that come with updating all of these reference
2581  * counts, sort them by byte number before actual updates are done.
2582  *
2583  * struct refsort is used to match byte number to slot in the btree block.
2584  * we sort based on the byte number and then use the slot to actually
2585  * find the item.
2586  *
2587  * struct refsort is smaller than strcut btrfs_item and smaller than
2588  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2589  * for a btree block, there's no way for a kmalloc of refsorts for a
2590  * single node to be bigger than a page.
2591  */
2592 struct refsort {
2593         u64 bytenr;
2594         u32 slot;
2595 };
2596
2597 /*
2598  * for passing into sort()
2599  */
2600 static int refsort_cmp(const void *a_void, const void *b_void)
2601 {
2602         const struct refsort *a = a_void;
2603         const struct refsort *b = b_void;
2604
2605         if (a->bytenr < b->bytenr)
2606                 return -1;
2607         if (a->bytenr > b->bytenr)
2608                 return 1;
2609         return 0;
2610 }
2611 #endif
2612
2613 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2614                            struct btrfs_root *root,
2615                            struct extent_buffer *buf,
2616                            int full_backref, int inc)
2617 {
2618         u64 bytenr;
2619         u64 num_bytes;
2620         u64 parent;
2621         u64 ref_root;
2622         u32 nritems;
2623         struct btrfs_key key;
2624         struct btrfs_file_extent_item *fi;
2625         int i;
2626         int level;
2627         int ret = 0;
2628         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2629                             u64, u64, u64, u64, u64, u64);
2630
2631         ref_root = btrfs_header_owner(buf);
2632         nritems = btrfs_header_nritems(buf);
2633         level = btrfs_header_level(buf);
2634
2635         if (!root->ref_cows && level == 0)
2636                 return 0;
2637
2638         if (inc)
2639                 process_func = btrfs_inc_extent_ref;
2640         else
2641                 process_func = btrfs_free_extent;
2642
2643         if (full_backref)
2644                 parent = buf->start;
2645         else
2646                 parent = 0;
2647
2648         for (i = 0; i < nritems; i++) {
2649                 if (level == 0) {
2650                         btrfs_item_key_to_cpu(buf, &key, i);
2651                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2652                                 continue;
2653                         fi = btrfs_item_ptr(buf, i,
2654                                             struct btrfs_file_extent_item);
2655                         if (btrfs_file_extent_type(buf, fi) ==
2656                             BTRFS_FILE_EXTENT_INLINE)
2657                                 continue;
2658                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2659                         if (bytenr == 0)
2660                                 continue;
2661
2662                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2663                         key.offset -= btrfs_file_extent_offset(buf, fi);
2664                         ret = process_func(trans, root, bytenr, num_bytes,
2665                                            parent, ref_root, key.objectid,
2666                                            key.offset);
2667                         if (ret)
2668                                 goto fail;
2669                 } else {
2670                         bytenr = btrfs_node_blockptr(buf, i);
2671                         num_bytes = btrfs_level_size(root, level - 1);
2672                         ret = process_func(trans, root, bytenr, num_bytes,
2673                                            parent, ref_root, level - 1, 0);
2674                         if (ret)
2675                                 goto fail;
2676                 }
2677         }
2678         return 0;
2679 fail:
2680         BUG();
2681         return ret;
2682 }
2683
2684 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2685                   struct extent_buffer *buf, int full_backref)
2686 {
2687         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2688 }
2689
2690 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2691                   struct extent_buffer *buf, int full_backref)
2692 {
2693         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2694 }
2695
2696 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2697                                  struct btrfs_root *root,
2698                                  struct btrfs_path *path,
2699                                  struct btrfs_block_group_cache *cache)
2700 {
2701         int ret;
2702         struct btrfs_root *extent_root = root->fs_info->extent_root;
2703         unsigned long bi;
2704         struct extent_buffer *leaf;
2705
2706         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2707         if (ret < 0)
2708                 goto fail;
2709         BUG_ON(ret);
2710
2711         leaf = path->nodes[0];
2712         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2713         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2714         btrfs_mark_buffer_dirty(leaf);
2715         btrfs_release_path(extent_root, path);
2716 fail:
2717         if (ret)
2718                 return ret;
2719         return 0;
2720
2721 }
2722
2723 static struct btrfs_block_group_cache *
2724 next_block_group(struct btrfs_root *root,
2725                  struct btrfs_block_group_cache *cache)
2726 {
2727         struct rb_node *node;
2728         spin_lock(&root->fs_info->block_group_cache_lock);
2729         node = rb_next(&cache->cache_node);
2730         btrfs_put_block_group(cache);
2731         if (node) {
2732                 cache = rb_entry(node, struct btrfs_block_group_cache,
2733                                  cache_node);
2734                 btrfs_get_block_group(cache);
2735         } else
2736                 cache = NULL;
2737         spin_unlock(&root->fs_info->block_group_cache_lock);
2738         return cache;
2739 }
2740
2741 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2742                             struct btrfs_trans_handle *trans,
2743                             struct btrfs_path *path)
2744 {
2745         struct btrfs_root *root = block_group->fs_info->tree_root;
2746         struct inode *inode = NULL;
2747         u64 alloc_hint = 0;
2748         int dcs = BTRFS_DC_ERROR;
2749         int num_pages = 0;
2750         int retries = 0;
2751         int ret = 0;
2752
2753         /*
2754          * If this block group is smaller than 100 megs don't bother caching the
2755          * block group.
2756          */
2757         if (block_group->key.offset < (100 * 1024 * 1024)) {
2758                 spin_lock(&block_group->lock);
2759                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2760                 spin_unlock(&block_group->lock);
2761                 return 0;
2762         }
2763
2764 again:
2765         inode = lookup_free_space_inode(root, block_group, path);
2766         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2767                 ret = PTR_ERR(inode);
2768                 btrfs_release_path(root, path);
2769                 goto out;
2770         }
2771
2772         if (IS_ERR(inode)) {
2773                 BUG_ON(retries);
2774                 retries++;
2775
2776                 if (block_group->ro)
2777                         goto out_free;
2778
2779                 ret = create_free_space_inode(root, trans, block_group, path);
2780                 if (ret)
2781                         goto out_free;
2782                 goto again;
2783         }
2784
2785         /*
2786          * We want to set the generation to 0, that way if anything goes wrong
2787          * from here on out we know not to trust this cache when we load up next
2788          * time.
2789          */
2790         BTRFS_I(inode)->generation = 0;
2791         ret = btrfs_update_inode(trans, root, inode);
2792         WARN_ON(ret);
2793
2794         if (i_size_read(inode) > 0) {
2795                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2796                                                       inode);
2797                 if (ret)
2798                         goto out_put;
2799         }
2800
2801         spin_lock(&block_group->lock);
2802         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2803                 /* We're not cached, don't bother trying to write stuff out */
2804                 dcs = BTRFS_DC_WRITTEN;
2805                 spin_unlock(&block_group->lock);
2806                 goto out_put;
2807         }
2808         spin_unlock(&block_group->lock);
2809
2810         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2811         if (!num_pages)
2812                 num_pages = 1;
2813
2814         /*
2815          * Just to make absolutely sure we have enough space, we're going to
2816          * preallocate 12 pages worth of space for each block group.  In
2817          * practice we ought to use at most 8, but we need extra space so we can
2818          * add our header and have a terminator between the extents and the
2819          * bitmaps.
2820          */
2821         num_pages *= 16;
2822         num_pages *= PAGE_CACHE_SIZE;
2823
2824         ret = btrfs_check_data_free_space(inode, num_pages);
2825         if (ret)
2826                 goto out_put;
2827
2828         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2829                                               num_pages, num_pages,
2830                                               &alloc_hint);
2831         if (!ret)
2832                 dcs = BTRFS_DC_SETUP;
2833         btrfs_free_reserved_data_space(inode, num_pages);
2834 out_put:
2835         iput(inode);
2836 out_free:
2837         btrfs_release_path(root, path);
2838 out:
2839         spin_lock(&block_group->lock);
2840         block_group->disk_cache_state = dcs;
2841         spin_unlock(&block_group->lock);
2842
2843         return ret;
2844 }
2845
2846 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2847                                    struct btrfs_root *root)
2848 {
2849         struct btrfs_block_group_cache *cache;
2850         int err = 0;
2851         struct btrfs_path *path;
2852         u64 last = 0;
2853
2854         path = btrfs_alloc_path();
2855         if (!path)
2856                 return -ENOMEM;
2857
2858 again:
2859         while (1) {
2860                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2861                 while (cache) {
2862                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2863                                 break;
2864                         cache = next_block_group(root, cache);
2865                 }
2866                 if (!cache) {
2867                         if (last == 0)
2868                                 break;
2869                         last = 0;
2870                         continue;
2871                 }
2872                 err = cache_save_setup(cache, trans, path);
2873                 last = cache->key.objectid + cache->key.offset;
2874                 btrfs_put_block_group(cache);
2875         }
2876
2877         while (1) {
2878                 if (last == 0) {
2879                         err = btrfs_run_delayed_refs(trans, root,
2880                                                      (unsigned long)-1);
2881                         BUG_ON(err);
2882                 }
2883
2884                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2885                 while (cache) {
2886                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2887                                 btrfs_put_block_group(cache);
2888                                 goto again;
2889                         }
2890
2891                         if (cache->dirty)
2892                                 break;
2893                         cache = next_block_group(root, cache);
2894                 }
2895                 if (!cache) {
2896                         if (last == 0)
2897                                 break;
2898                         last = 0;
2899                         continue;
2900                 }
2901
2902                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2903                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2904                 cache->dirty = 0;
2905                 last = cache->key.objectid + cache->key.offset;
2906
2907                 err = write_one_cache_group(trans, root, path, cache);
2908                 BUG_ON(err);
2909                 btrfs_put_block_group(cache);
2910         }
2911
2912         while (1) {
2913                 /*
2914                  * I don't think this is needed since we're just marking our
2915                  * preallocated extent as written, but just in case it can't
2916                  * hurt.
2917                  */
2918                 if (last == 0) {
2919                         err = btrfs_run_delayed_refs(trans, root,
2920                                                      (unsigned long)-1);
2921                         BUG_ON(err);
2922                 }
2923
2924                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2925                 while (cache) {
2926                         /*
2927                          * Really this shouldn't happen, but it could if we
2928                          * couldn't write the entire preallocated extent and
2929                          * splitting the extent resulted in a new block.
2930                          */
2931                         if (cache->dirty) {
2932                                 btrfs_put_block_group(cache);
2933                                 goto again;
2934                         }
2935                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2936                                 break;
2937                         cache = next_block_group(root, cache);
2938                 }
2939                 if (!cache) {
2940                         if (last == 0)
2941                                 break;
2942                         last = 0;
2943                         continue;
2944                 }
2945
2946                 btrfs_write_out_cache(root, trans, cache, path);
2947
2948                 /*
2949                  * If we didn't have an error then the cache state is still
2950                  * NEED_WRITE, so we can set it to WRITTEN.
2951                  */
2952                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2953                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2954                 last = cache->key.objectid + cache->key.offset;
2955                 btrfs_put_block_group(cache);
2956         }
2957
2958         btrfs_free_path(path);
2959         return 0;
2960 }
2961
2962 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2963 {
2964         struct btrfs_block_group_cache *block_group;
2965         int readonly = 0;
2966
2967         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2968         if (!block_group || block_group->ro)
2969                 readonly = 1;
2970         if (block_group)
2971                 btrfs_put_block_group(block_group);
2972         return readonly;
2973 }
2974
2975 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2976                              u64 total_bytes, u64 bytes_used,
2977                              struct btrfs_space_info **space_info)
2978 {
2979         struct btrfs_space_info *found;
2980         int i;
2981         int factor;
2982
2983         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2984                      BTRFS_BLOCK_GROUP_RAID10))
2985                 factor = 2;
2986         else
2987                 factor = 1;
2988
2989         found = __find_space_info(info, flags);
2990         if (found) {
2991                 spin_lock(&found->lock);
2992                 found->total_bytes += total_bytes;
2993                 found->disk_total += total_bytes * factor;
2994                 found->bytes_used += bytes_used;
2995                 found->disk_used += bytes_used * factor;
2996                 found->full = 0;
2997                 spin_unlock(&found->lock);
2998                 *space_info = found;
2999                 return 0;
3000         }
3001         found = kzalloc(sizeof(*found), GFP_NOFS);
3002         if (!found)
3003                 return -ENOMEM;
3004
3005         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3006                 INIT_LIST_HEAD(&found->block_groups[i]);
3007         init_rwsem(&found->groups_sem);
3008         spin_lock_init(&found->lock);
3009         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3010                                 BTRFS_BLOCK_GROUP_SYSTEM |
3011                                 BTRFS_BLOCK_GROUP_METADATA);
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->bytes_pinned = 0;
3017         found->bytes_reserved = 0;
3018         found->bytes_readonly = 0;
3019         found->bytes_may_use = 0;
3020         found->full = 0;
3021         found->force_alloc = 0;
3022         *space_info = found;
3023         list_add_rcu(&found->list, &info->space_info);
3024         atomic_set(&found->caching_threads, 0);
3025         return 0;
3026 }
3027
3028 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3029 {
3030         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3031                                    BTRFS_BLOCK_GROUP_RAID1 |
3032                                    BTRFS_BLOCK_GROUP_RAID10 |
3033                                    BTRFS_BLOCK_GROUP_DUP);
3034         if (extra_flags) {
3035                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3036                         fs_info->avail_data_alloc_bits |= extra_flags;
3037                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3038                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3039                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3040                         fs_info->avail_system_alloc_bits |= extra_flags;
3041         }
3042 }
3043
3044 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3045 {
3046         /*
3047          * we add in the count of missing devices because we want
3048          * to make sure that any RAID levels on a degraded FS
3049          * continue to be honored.
3050          */
3051         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3052                 root->fs_info->fs_devices->missing_devices;
3053
3054         if (num_devices == 1)
3055                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3056         if (num_devices < 4)
3057                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3058
3059         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3060             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3061                       BTRFS_BLOCK_GROUP_RAID10))) {
3062                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3063         }
3064
3065         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3066             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3067                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3068         }
3069
3070         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3071             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3072              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3073              (flags & BTRFS_BLOCK_GROUP_DUP)))
3074                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3075         return flags;
3076 }
3077
3078 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3079 {
3080         if (flags & BTRFS_BLOCK_GROUP_DATA)
3081                 flags |= root->fs_info->avail_data_alloc_bits &
3082                          root->fs_info->data_alloc_profile;
3083         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3084                 flags |= root->fs_info->avail_system_alloc_bits &
3085                          root->fs_info->system_alloc_profile;
3086         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3087                 flags |= root->fs_info->avail_metadata_alloc_bits &
3088                          root->fs_info->metadata_alloc_profile;
3089         return btrfs_reduce_alloc_profile(root, flags);
3090 }
3091
3092 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3093 {
3094         u64 flags;
3095
3096         if (data)
3097                 flags = BTRFS_BLOCK_GROUP_DATA;
3098         else if (root == root->fs_info->chunk_root)
3099                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3100         else
3101                 flags = BTRFS_BLOCK_GROUP_METADATA;
3102
3103         return get_alloc_profile(root, flags);
3104 }
3105
3106 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3107 {
3108         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3109                                                        BTRFS_BLOCK_GROUP_DATA);
3110 }
3111
3112 /*
3113  * This will check the space that the inode allocates from to make sure we have
3114  * enough space for bytes.
3115  */
3116 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3117 {
3118         struct btrfs_space_info *data_sinfo;
3119         struct btrfs_root *root = BTRFS_I(inode)->root;
3120         u64 used;
3121         int ret = 0, committed = 0, alloc_chunk = 1;
3122
3123         /* make sure bytes are sectorsize aligned */
3124         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3125
3126         if (root == root->fs_info->tree_root) {
3127                 alloc_chunk = 0;
3128                 committed = 1;
3129         }
3130
3131         data_sinfo = BTRFS_I(inode)->space_info;
3132         if (!data_sinfo)
3133                 goto alloc;
3134
3135 again:
3136         /* make sure we have enough space to handle the data first */
3137         spin_lock(&data_sinfo->lock);
3138         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3139                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3140                 data_sinfo->bytes_may_use;
3141
3142         if (used + bytes > data_sinfo->total_bytes) {
3143                 struct btrfs_trans_handle *trans;
3144
3145                 /*
3146                  * if we don't have enough free bytes in this space then we need
3147                  * to alloc a new chunk.
3148                  */
3149                 if (!data_sinfo->full && alloc_chunk) {
3150                         u64 alloc_target;
3151
3152                         data_sinfo->force_alloc = 1;
3153                         spin_unlock(&data_sinfo->lock);
3154 alloc:
3155                         alloc_target = btrfs_get_alloc_profile(root, 1);
3156                         trans = btrfs_join_transaction(root, 1);
3157                         if (IS_ERR(trans))
3158                                 return PTR_ERR(trans);
3159
3160                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3161                                              bytes + 2 * 1024 * 1024,
3162                                              alloc_target, 0);
3163                         btrfs_end_transaction(trans, root);
3164                         if (ret < 0) {
3165                                 if (ret != -ENOSPC)
3166                                         return ret;
3167                                 else
3168                                         goto commit_trans;
3169                         }
3170
3171                         if (!data_sinfo) {
3172                                 btrfs_set_inode_space_info(root, inode);
3173                                 data_sinfo = BTRFS_I(inode)->space_info;
3174                         }
3175                         goto again;
3176                 }
3177                 spin_unlock(&data_sinfo->lock);
3178
3179                 /* commit the current transaction and try again */
3180 commit_trans:
3181                 if (!committed && !root->fs_info->open_ioctl_trans) {
3182                         committed = 1;
3183                         trans = btrfs_join_transaction(root, 1);
3184                         if (IS_ERR(trans))
3185                                 return PTR_ERR(trans);
3186                         ret = btrfs_commit_transaction(trans, root);
3187                         if (ret)
3188                                 return ret;
3189                         goto again;
3190                 }
3191
3192 #if 0 /* I hope we never need this code again, just in case */
3193                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3194                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3195                        "%llu bytes_readonly, %llu may use %llu total\n",
3196                        (unsigned long long)bytes,
3197                        (unsigned long long)data_sinfo->bytes_used,
3198                        (unsigned long long)data_sinfo->bytes_reserved,
3199                        (unsigned long long)data_sinfo->bytes_pinned,
3200                        (unsigned long long)data_sinfo->bytes_readonly,
3201                        (unsigned long long)data_sinfo->bytes_may_use,
3202                        (unsigned long long)data_sinfo->total_bytes);
3203 #endif
3204                 return -ENOSPC;
3205         }
3206         data_sinfo->bytes_may_use += bytes;
3207         BTRFS_I(inode)->reserved_bytes += bytes;
3208         spin_unlock(&data_sinfo->lock);
3209
3210         return 0;
3211 }
3212
3213 /*
3214  * called when we are clearing an delalloc extent from the
3215  * inode's io_tree or there was an error for whatever reason
3216  * after calling btrfs_check_data_free_space
3217  */
3218 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3219 {
3220         struct btrfs_root *root = BTRFS_I(inode)->root;
3221         struct btrfs_space_info *data_sinfo;
3222
3223         /* make sure bytes are sectorsize aligned */
3224         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3225
3226         data_sinfo = BTRFS_I(inode)->space_info;
3227         spin_lock(&data_sinfo->lock);
3228         data_sinfo->bytes_may_use -= bytes;
3229         BTRFS_I(inode)->reserved_bytes -= bytes;
3230         spin_unlock(&data_sinfo->lock);
3231 }
3232
3233 static void force_metadata_allocation(struct btrfs_fs_info *info)
3234 {
3235         struct list_head *head = &info->space_info;
3236         struct btrfs_space_info *found;
3237
3238         rcu_read_lock();
3239         list_for_each_entry_rcu(found, head, list) {
3240                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3241                         found->force_alloc = 1;
3242         }
3243         rcu_read_unlock();
3244 }
3245
3246 static int should_alloc_chunk(struct btrfs_root *root,
3247                               struct btrfs_space_info *sinfo, u64 alloc_bytes)
3248 {
3249         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3250         u64 thresh;
3251
3252         if (sinfo->bytes_used + sinfo->bytes_reserved +
3253             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3254                 return 0;
3255
3256         if (sinfo->bytes_used + sinfo->bytes_reserved +
3257             alloc_bytes < div_factor(num_bytes, 8))
3258                 return 0;
3259
3260         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3261         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3262
3263         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3264                 return 0;
3265
3266         return 1;
3267 }
3268
3269 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3270                           struct btrfs_root *extent_root, u64 alloc_bytes,
3271                           u64 flags, int force)
3272 {
3273         struct btrfs_space_info *space_info;
3274         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3275         int ret = 0;
3276
3277         mutex_lock(&fs_info->chunk_mutex);
3278
3279         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3280
3281         space_info = __find_space_info(extent_root->fs_info, flags);
3282         if (!space_info) {
3283                 ret = update_space_info(extent_root->fs_info, flags,
3284                                         0, 0, &space_info);
3285                 BUG_ON(ret);
3286         }
3287         BUG_ON(!space_info);
3288
3289         spin_lock(&space_info->lock);
3290         if (space_info->force_alloc)
3291                 force = 1;
3292         if (space_info->full) {
3293                 spin_unlock(&space_info->lock);
3294                 goto out;
3295         }
3296
3297         if (!force && !should_alloc_chunk(extent_root, space_info,
3298                                           alloc_bytes)) {
3299                 spin_unlock(&space_info->lock);
3300                 goto out;
3301         }
3302         spin_unlock(&space_info->lock);
3303
3304         /*
3305          * If we have mixed data/metadata chunks we want to make sure we keep
3306          * allocating mixed chunks instead of individual chunks.
3307          */
3308         if (btrfs_mixed_space_info(space_info))
3309                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3310
3311         /*
3312          * if we're doing a data chunk, go ahead and make sure that
3313          * we keep a reasonable number of metadata chunks allocated in the
3314          * FS as well.
3315          */
3316         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3317                 fs_info->data_chunk_allocations++;
3318                 if (!(fs_info->data_chunk_allocations %
3319                       fs_info->metadata_ratio))
3320                         force_metadata_allocation(fs_info);
3321         }
3322
3323         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3324         spin_lock(&space_info->lock);
3325         if (ret)
3326                 space_info->full = 1;
3327         else
3328                 ret = 1;
3329         space_info->force_alloc = 0;
3330         spin_unlock(&space_info->lock);
3331 out:
3332         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3333         return ret;
3334 }
3335
3336 /*
3337  * shrink metadata reservation for delalloc
3338  */
3339 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3340                            struct btrfs_root *root, u64 to_reclaim, int sync)
3341 {
3342         struct btrfs_block_rsv *block_rsv;
3343         struct btrfs_space_info *space_info;
3344         u64 reserved;
3345         u64 max_reclaim;
3346         u64 reclaimed = 0;
3347         int pause = 1;
3348         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3349
3350         block_rsv = &root->fs_info->delalloc_block_rsv;
3351         space_info = block_rsv->space_info;
3352
3353         smp_mb();
3354         reserved = space_info->bytes_reserved;
3355
3356         if (reserved == 0)
3357                 return 0;
3358
3359         max_reclaim = min(reserved, to_reclaim);
3360
3361         while (1) {
3362                 /* have the flusher threads jump in and do some IO */
3363                 smp_mb();
3364                 nr_pages = min_t(unsigned long, nr_pages,
3365                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3366                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3367
3368                 spin_lock(&space_info->lock);
3369                 if (reserved > space_info->bytes_reserved)
3370                         reclaimed += reserved - space_info->bytes_reserved;
3371                 reserved = space_info->bytes_reserved;
3372                 spin_unlock(&space_info->lock);
3373
3374                 if (reserved == 0 || reclaimed >= max_reclaim)
3375                         break;
3376
3377                 if (trans && trans->transaction->blocked)
3378                         return -EAGAIN;
3379
3380                 __set_current_state(TASK_INTERRUPTIBLE);
3381                 schedule_timeout(pause);
3382                 pause <<= 1;
3383                 if (pause > HZ / 10)
3384                         pause = HZ / 10;
3385
3386         }
3387         return reclaimed >= to_reclaim;
3388 }
3389
3390 /*
3391  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3392  * idea is if this is the first call (retries == 0) then we will add to our
3393  * reserved count if we can't make the allocation in order to hold our place
3394  * while we go and try and free up space.  That way for retries > 1 we don't try
3395  * and add space, we just check to see if the amount of unused space is >= the
3396  * total space, meaning that our reservation is valid.
3397  *
3398  * However if we don't intend to retry this reservation, pass -1 as retries so
3399  * that it short circuits this logic.
3400  */
3401 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3402                                   struct btrfs_root *root,
3403                                   struct btrfs_block_rsv *block_rsv,
3404                                   u64 orig_bytes, int flush)
3405 {
3406         struct btrfs_space_info *space_info = block_rsv->space_info;
3407         u64 unused;
3408         u64 num_bytes = orig_bytes;
3409         int retries = 0;
3410         int ret = 0;
3411         bool reserved = false;
3412         bool committed = false;
3413
3414 again:
3415         ret = -ENOSPC;
3416         if (reserved)
3417                 num_bytes = 0;
3418
3419         spin_lock(&space_info->lock);
3420         unused = space_info->bytes_used + space_info->bytes_reserved +
3421                  space_info->bytes_pinned + space_info->bytes_readonly +
3422                  space_info->bytes_may_use;
3423
3424         /*
3425          * The idea here is that we've not already over-reserved the block group
3426          * then we can go ahead and save our reservation first and then start
3427          * flushing if we need to.  Otherwise if we've already overcommitted
3428          * lets start flushing stuff first and then come back and try to make
3429          * our reservation.
3430          */
3431         if (unused <= space_info->total_bytes) {
3432                 unused = space_info->total_bytes - unused;
3433                 if (unused >= num_bytes) {
3434                         if (!reserved)
3435                                 space_info->bytes_reserved += orig_bytes;
3436                         ret = 0;
3437                 } else {
3438                         /*
3439                          * Ok set num_bytes to orig_bytes since we aren't
3440                          * overocmmitted, this way we only try and reclaim what
3441                          * we need.
3442                          */
3443                         num_bytes = orig_bytes;
3444                 }
3445         } else {
3446                 /*
3447                  * Ok we're over committed, set num_bytes to the overcommitted
3448                  * amount plus the amount of bytes that we need for this
3449                  * reservation.
3450                  */
3451                 num_bytes = unused - space_info->total_bytes +
3452                         (orig_bytes * (retries + 1));
3453         }
3454
3455         /*
3456          * Couldn't make our reservation, save our place so while we're trying
3457          * to reclaim space we can actually use it instead of somebody else
3458          * stealing it from us.
3459          */
3460         if (ret && !reserved) {
3461                 space_info->bytes_reserved += orig_bytes;
3462                 reserved = true;
3463         }
3464
3465         spin_unlock(&space_info->lock);
3466
3467         if (!ret)
3468                 return 0;
3469
3470         if (!flush)
3471                 goto out;
3472
3473         /*
3474          * We do synchronous shrinking since we don't actually unreserve
3475          * metadata until after the IO is completed.
3476          */
3477         ret = shrink_delalloc(trans, root, num_bytes, 1);
3478         if (ret > 0)
3479                 return 0;
3480         else if (ret < 0)
3481                 goto out;
3482
3483         /*
3484          * So if we were overcommitted it's possible that somebody else flushed
3485          * out enough space and we simply didn't have enough space to reclaim,
3486          * so go back around and try again.
3487          */
3488         if (retries < 2) {
3489                 retries++;
3490                 goto again;
3491         }
3492
3493         spin_lock(&space_info->lock);
3494         /*
3495          * Not enough space to be reclaimed, don't bother committing the
3496          * transaction.
3497          */
3498         if (space_info->bytes_pinned < orig_bytes)
3499                 ret = -ENOSPC;
3500         spin_unlock(&space_info->lock);
3501         if (ret)
3502                 goto out;
3503
3504         ret = -EAGAIN;
3505         if (trans || committed)
3506                 goto out;
3507
3508         ret = -ENOSPC;
3509         trans = btrfs_join_transaction(root, 1);
3510         if (IS_ERR(trans))
3511                 goto out;
3512         ret = btrfs_commit_transaction(trans, root);
3513         if (!ret) {
3514                 trans = NULL;
3515                 committed = true;
3516                 goto again;
3517         }
3518
3519 out:
3520         if (reserved) {
3521                 spin_lock(&space_info->lock);
3522                 space_info->bytes_reserved -= orig_bytes;
3523                 spin_unlock(&space_info->lock);
3524         }
3525
3526         return ret;
3527 }
3528
3529 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3530                                              struct btrfs_root *root)
3531 {
3532         struct btrfs_block_rsv *block_rsv;
3533         if (root->ref_cows)
3534                 block_rsv = trans->block_rsv;
3535         else
3536                 block_rsv = root->block_rsv;
3537
3538         if (!block_rsv)
3539                 block_rsv = &root->fs_info->empty_block_rsv;
3540
3541         return block_rsv;
3542 }
3543
3544 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3545                                u64 num_bytes)
3546 {
3547         int ret = -ENOSPC;
3548         spin_lock(&block_rsv->lock);
3549         if (block_rsv->reserved >= num_bytes) {
3550                 block_rsv->reserved -= num_bytes;
3551                 if (block_rsv->reserved < block_rsv->size)
3552                         block_rsv->full = 0;
3553                 ret = 0;
3554         }
3555         spin_unlock(&block_rsv->lock);
3556         return ret;
3557 }
3558
3559 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3560                                 u64 num_bytes, int update_size)
3561 {
3562         spin_lock(&block_rsv->lock);
3563         block_rsv->reserved += num_bytes;
3564         if (update_size)
3565                 block_rsv->size += num_bytes;
3566         else if (block_rsv->reserved >= block_rsv->size)
3567                 block_rsv->full = 1;
3568         spin_unlock(&block_rsv->lock);
3569 }
3570
3571 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3572                              struct btrfs_block_rsv *dest, u64 num_bytes)
3573 {
3574         struct btrfs_space_info *space_info = block_rsv->space_info;
3575
3576         spin_lock(&block_rsv->lock);
3577         if (num_bytes == (u64)-1)
3578                 num_bytes = block_rsv->size;
3579         block_rsv->size -= num_bytes;
3580         if (block_rsv->reserved >= block_rsv->size) {
3581                 num_bytes = block_rsv->reserved - block_rsv->size;
3582                 block_rsv->reserved = block_rsv->size;
3583                 block_rsv->full = 1;
3584         } else {
3585                 num_bytes = 0;
3586         }
3587         spin_unlock(&block_rsv->lock);
3588
3589         if (num_bytes > 0) {
3590                 if (dest) {
3591                         block_rsv_add_bytes(dest, num_bytes, 0);
3592                 } else {
3593                         spin_lock(&space_info->lock);
3594                         space_info->bytes_reserved -= num_bytes;
3595                         spin_unlock(&space_info->lock);
3596                 }
3597         }
3598 }
3599
3600 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3601                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3602 {
3603         int ret;
3604
3605         ret = block_rsv_use_bytes(src, num_bytes);
3606         if (ret)
3607                 return ret;
3608
3609         block_rsv_add_bytes(dst, num_bytes, 1);
3610         return 0;
3611 }
3612
3613 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3614 {
3615         memset(rsv, 0, sizeof(*rsv));
3616         spin_lock_init(&rsv->lock);
3617         atomic_set(&rsv->usage, 1);
3618         rsv->priority = 6;
3619         INIT_LIST_HEAD(&rsv->list);
3620 }
3621
3622 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3623 {
3624         struct btrfs_block_rsv *block_rsv;
3625         struct btrfs_fs_info *fs_info = root->fs_info;
3626
3627         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3628         if (!block_rsv)
3629                 return NULL;
3630
3631         btrfs_init_block_rsv(block_rsv);
3632         block_rsv->space_info = __find_space_info(fs_info,
3633                                                   BTRFS_BLOCK_GROUP_METADATA);
3634         return block_rsv;
3635 }
3636
3637 void btrfs_free_block_rsv(struct btrfs_root *root,
3638                           struct btrfs_block_rsv *rsv)
3639 {
3640         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3641                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3642                 if (!rsv->durable)
3643                         kfree(rsv);
3644         }
3645 }
3646
3647 /*
3648  * make the block_rsv struct be able to capture freed space.
3649  * the captured space will re-add to the the block_rsv struct
3650  * after transaction commit
3651  */
3652 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3653                                  struct btrfs_block_rsv *block_rsv)
3654 {
3655         block_rsv->durable = 1;
3656         mutex_lock(&fs_info->durable_block_rsv_mutex);
3657         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3658         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3659 }
3660
3661 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3662                         struct btrfs_root *root,
3663                         struct btrfs_block_rsv *block_rsv,
3664                         u64 num_bytes)
3665 {
3666         int ret;
3667
3668         if (num_bytes == 0)
3669                 return 0;
3670
3671         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3672         if (!ret) {
3673                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3674                 return 0;
3675         }
3676
3677         return ret;
3678 }
3679
3680 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3681                           struct btrfs_root *root,
3682                           struct btrfs_block_rsv *block_rsv,
3683                           u64 min_reserved, int min_factor)
3684 {
3685         u64 num_bytes = 0;
3686         int commit_trans = 0;
3687         int ret = -ENOSPC;
3688
3689         if (!block_rsv)
3690                 return 0;
3691
3692         spin_lock(&block_rsv->lock);
3693         if (min_factor > 0)
3694                 num_bytes = div_factor(block_rsv->size, min_factor);
3695         if (min_reserved > num_bytes)
3696                 num_bytes = min_reserved;
3697
3698         if (block_rsv->reserved >= num_bytes) {
3699                 ret = 0;
3700         } else {
3701                 num_bytes -= block_rsv->reserved;
3702                 if (block_rsv->durable &&
3703                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3704                         commit_trans = 1;
3705         }
3706         spin_unlock(&block_rsv->lock);
3707         if (!ret)
3708                 return 0;
3709
3710         if (block_rsv->refill_used) {
3711                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3712                                              num_bytes, 0);
3713                 if (!ret) {
3714                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3715                         return 0;
3716                 }
3717         }
3718
3719         if (commit_trans) {
3720                 if (trans)
3721                         return -EAGAIN;
3722
3723                 trans = btrfs_join_transaction(root, 1);
3724                 BUG_ON(IS_ERR(trans));
3725                 ret = btrfs_commit_transaction(trans, root);
3726                 return 0;
3727         }
3728
3729         return -ENOSPC;
3730 }
3731
3732 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3733                             struct btrfs_block_rsv *dst_rsv,
3734                             u64 num_bytes)
3735 {
3736         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3737 }
3738
3739 void btrfs_block_rsv_release(struct btrfs_root *root,
3740                              struct btrfs_block_rsv *block_rsv,
3741                              u64 num_bytes)
3742 {
3743         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3744         if (global_rsv->full || global_rsv == block_rsv ||
3745             block_rsv->space_info != global_rsv->space_info)
3746                 global_rsv = NULL;
3747         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3748 }
3749
3750 /*
3751  * helper to calculate size of global block reservation.
3752  * the desired value is sum of space used by extent tree,
3753  * checksum tree and root tree
3754  */
3755 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3756 {
3757         struct btrfs_space_info *sinfo;
3758         u64 num_bytes;
3759         u64 meta_used;
3760         u64 data_used;
3761         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3762 #if 0
3763         /*
3764          * per tree used space accounting can be inaccuracy, so we
3765          * can't rely on it.
3766          */
3767         spin_lock(&fs_info->extent_root->accounting_lock);
3768         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3769         spin_unlock(&fs_info->extent_root->accounting_lock);
3770
3771         spin_lock(&fs_info->csum_root->accounting_lock);
3772         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3773         spin_unlock(&fs_info->csum_root->accounting_lock);
3774
3775         spin_lock(&fs_info->tree_root->accounting_lock);
3776         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3777         spin_unlock(&fs_info->tree_root->accounting_lock);
3778 #endif
3779         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3780         spin_lock(&sinfo->lock);
3781         data_used = sinfo->bytes_used;
3782         spin_unlock(&sinfo->lock);
3783
3784         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3785         spin_lock(&sinfo->lock);
3786         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3787                 data_used = 0;
3788         meta_used = sinfo->bytes_used;
3789         spin_unlock(&sinfo->lock);
3790
3791         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3792                     csum_size * 2;
3793         num_bytes += div64_u64(data_used + meta_used, 50);
3794
3795         if (num_bytes * 3 > meta_used)
3796                 num_bytes = div64_u64(meta_used, 3);
3797
3798         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3799 }
3800
3801 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3802 {
3803         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3804         struct btrfs_space_info *sinfo = block_rsv->space_info;
3805         u64 num_bytes;
3806
3807         num_bytes = calc_global_metadata_size(fs_info);
3808
3809         spin_lock(&block_rsv->lock);
3810         spin_lock(&sinfo->lock);
3811
3812         block_rsv->size = num_bytes;
3813
3814         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3815                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3816                     sinfo->bytes_may_use;
3817
3818         if (sinfo->total_bytes > num_bytes) {
3819                 num_bytes = sinfo->total_bytes - num_bytes;
3820                 block_rsv->reserved += num_bytes;
3821                 sinfo->bytes_reserved += num_bytes;
3822         }
3823
3824         if (block_rsv->reserved >= block_rsv->size) {
3825                 num_bytes = block_rsv->reserved - block_rsv->size;
3826                 sinfo->bytes_reserved -= num_bytes;
3827                 block_rsv->reserved = block_rsv->size;
3828                 block_rsv->full = 1;
3829         }
3830 #if 0
3831         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3832                 block_rsv->size, block_rsv->reserved);
3833 #endif
3834         spin_unlock(&sinfo->lock);
3835         spin_unlock(&block_rsv->lock);
3836 }
3837
3838 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3839 {
3840         struct btrfs_space_info *space_info;
3841
3842         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3843         fs_info->chunk_block_rsv.space_info = space_info;
3844         fs_info->chunk_block_rsv.priority = 10;
3845
3846         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3847         fs_info->global_block_rsv.space_info = space_info;
3848         fs_info->global_block_rsv.priority = 10;
3849         fs_info->global_block_rsv.refill_used = 1;
3850         fs_info->delalloc_block_rsv.space_info = space_info;
3851         fs_info->trans_block_rsv.space_info = space_info;
3852         fs_info->empty_block_rsv.space_info = space_info;
3853         fs_info->empty_block_rsv.priority = 10;
3854
3855         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3856         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3857         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3858         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3859         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3860
3861         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3862
3863         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3864
3865         update_global_block_rsv(fs_info);
3866 }
3867
3868 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3869 {
3870         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3871         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3872         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3873         WARN_ON(fs_info->trans_block_rsv.size > 0);
3874         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3875         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3876         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3877 }
3878
3879 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3880 {
3881         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3882                 3 * num_items;
3883 }
3884
3885 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3886                                  struct btrfs_root *root,
3887                                  int num_items)
3888 {
3889         u64 num_bytes;
3890         int ret;
3891
3892         if (num_items == 0 || root->fs_info->chunk_root == root)
3893                 return 0;
3894
3895         num_bytes = calc_trans_metadata_size(root, num_items);
3896         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3897                                   num_bytes);
3898         if (!ret) {
3899                 trans->bytes_reserved += num_bytes;
3900                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3901         }
3902         return ret;
3903 }
3904
3905 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3906                                   struct btrfs_root *root)
3907 {
3908         if (!trans->bytes_reserved)
3909                 return;
3910
3911         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3912         btrfs_block_rsv_release(root, trans->block_rsv,
3913                                 trans->bytes_reserved);
3914         trans->bytes_reserved = 0;
3915 }
3916
3917 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3918                                   struct inode *inode)
3919 {
3920         struct btrfs_root *root = BTRFS_I(inode)->root;
3921         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3922         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3923
3924         /*
3925          * one for deleting orphan item, one for updating inode and
3926          * two for calling btrfs_truncate_inode_items.
3927          *
3928          * btrfs_truncate_inode_items is a delete operation, it frees
3929          * more space than it uses in most cases. So two units of
3930          * metadata space should be enough for calling it many times.
3931          * If all of the metadata space is used, we can commit
3932          * transaction and use space it freed.
3933          */
3934         u64 num_bytes = calc_trans_metadata_size(root, 4);
3935         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3936 }
3937
3938 void btrfs_orphan_release_metadata(struct inode *inode)
3939 {
3940         struct btrfs_root *root = BTRFS_I(inode)->root;
3941         u64 num_bytes = calc_trans_metadata_size(root, 4);
3942         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3943 }
3944
3945 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3946                                 struct btrfs_pending_snapshot *pending)
3947 {
3948         struct btrfs_root *root = pending->root;
3949         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3950         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3951         /*
3952          * two for root back/forward refs, two for directory entries
3953          * and one for root of the snapshot.
3954          */
3955         u64 num_bytes = calc_trans_metadata_size(root, 5);
3956         dst_rsv->space_info = src_rsv->space_info;
3957         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3958 }
3959
3960 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3961 {
3962         return num_bytes >>= 3;
3963 }
3964
3965 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3966 {
3967         struct btrfs_root *root = BTRFS_I(inode)->root;
3968         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3969         u64 to_reserve;
3970         int nr_extents;
3971         int ret;
3972
3973         if (btrfs_transaction_in_commit(root->fs_info))
3974                 schedule_timeout(1);
3975
3976         num_bytes = ALIGN(num_bytes, root->sectorsize);
3977
3978         spin_lock(&BTRFS_I(inode)->accounting_lock);
3979         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3980         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3981                 nr_extents -= BTRFS_I(inode)->reserved_extents;
3982                 to_reserve = calc_trans_metadata_size(root, nr_extents);
3983         } else {
3984                 nr_extents = 0;
3985                 to_reserve = 0;
3986         }
3987         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3988
3989         to_reserve += calc_csum_metadata_size(inode, num_bytes);
3990         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
3991         if (ret)
3992                 return ret;
3993
3994         spin_lock(&BTRFS_I(inode)->accounting_lock);
3995         BTRFS_I(inode)->reserved_extents += nr_extents;
3996         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3997         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3998
3999         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4000
4001         if (block_rsv->size > 512 * 1024 * 1024)
4002                 shrink_delalloc(NULL, root, to_reserve, 0);
4003
4004         return 0;
4005 }
4006
4007 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4008 {
4009         struct btrfs_root *root = BTRFS_I(inode)->root;
4010         u64 to_free;
4011         int nr_extents;
4012
4013         num_bytes = ALIGN(num_bytes, root->sectorsize);
4014         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4015
4016         spin_lock(&BTRFS_I(inode)->accounting_lock);
4017         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4018         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
4019                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
4020                 BTRFS_I(inode)->reserved_extents -= nr_extents;
4021         } else {
4022                 nr_extents = 0;
4023         }
4024         spin_unlock(&BTRFS_I(inode)->accounting_lock);
4025
4026         to_free = calc_csum_metadata_size(inode, num_bytes);
4027         if (nr_extents > 0)
4028                 to_free += calc_trans_metadata_size(root, nr_extents);
4029
4030         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4031                                 to_free);
4032 }
4033
4034 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4035 {
4036         int ret;
4037
4038         ret = btrfs_check_data_free_space(inode, num_bytes);
4039         if (ret)
4040                 return ret;
4041
4042         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4043         if (ret) {
4044                 btrfs_free_reserved_data_space(inode, num_bytes);
4045                 return ret;
4046         }
4047
4048         return 0;
4049 }
4050
4051 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4052 {
4053         btrfs_delalloc_release_metadata(inode, num_bytes);
4054         btrfs_free_reserved_data_space(inode, num_bytes);
4055 }
4056
4057 static int update_block_group(struct btrfs_trans_handle *trans,
4058                               struct btrfs_root *root,
4059                               u64 bytenr, u64 num_bytes, int alloc)
4060 {
4061         struct btrfs_block_group_cache *cache = NULL;
4062         struct btrfs_fs_info *info = root->fs_info;
4063         u64 total = num_bytes;
4064         u64 old_val;
4065         u64 byte_in_group;
4066         int factor;
4067
4068         /* block accounting for super block */
4069         spin_lock(&info->delalloc_lock);
4070         old_val = btrfs_super_bytes_used(&info->super_copy);
4071         if (alloc)
4072                 old_val += num_bytes;
4073         else
4074                 old_val -= num_bytes;
4075         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4076         spin_unlock(&info->delalloc_lock);
4077
4078         while (total) {
4079                 cache = btrfs_lookup_block_group(info, bytenr);
4080                 if (!cache)
4081                         return -1;
4082                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4083                                     BTRFS_BLOCK_GROUP_RAID1 |
4084                                     BTRFS_BLOCK_GROUP_RAID10))
4085                         factor = 2;
4086                 else
4087                         factor = 1;
4088                 /*
4089                  * If this block group has free space cache written out, we
4090                  * need to make sure to load it if we are removing space.  This
4091                  * is because we need the unpinning stage to actually add the
4092                  * space back to the block group, otherwise we will leak space.
4093                  */
4094                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4095                         cache_block_group(cache, trans, NULL, 1);
4096
4097                 byte_in_group = bytenr - cache->key.objectid;
4098                 WARN_ON(byte_in_group > cache->key.offset);
4099
4100                 spin_lock(&cache->space_info->lock);
4101                 spin_lock(&cache->lock);
4102
4103                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4104                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4105                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4106
4107                 cache->dirty = 1;
4108                 old_val = btrfs_block_group_used(&cache->item);
4109                 num_bytes = min(total, cache->key.offset - byte_in_group);
4110                 if (alloc) {
4111                         old_val += num_bytes;
4112                         btrfs_set_block_group_used(&cache->item, old_val);
4113                         cache->reserved -= num_bytes;
4114                         cache->space_info->bytes_reserved -= num_bytes;
4115                         cache->space_info->bytes_used += num_bytes;
4116                         cache->space_info->disk_used += num_bytes * factor;
4117                         spin_unlock(&cache->lock);
4118                         spin_unlock(&cache->space_info->lock);
4119                 } else {
4120                         old_val -= num_bytes;
4121                         btrfs_set_block_group_used(&cache->item, old_val);
4122                         cache->pinned += num_bytes;
4123                         cache->space_info->bytes_pinned += num_bytes;
4124                         cache->space_info->bytes_used -= num_bytes;
4125                         cache->space_info->disk_used -= num_bytes * factor;
4126                         spin_unlock(&cache->lock);
4127                         spin_unlock(&cache->space_info->lock);
4128
4129                         set_extent_dirty(info->pinned_extents,
4130                                          bytenr, bytenr + num_bytes - 1,
4131                                          GFP_NOFS | __GFP_NOFAIL);
4132                 }
4133                 btrfs_put_block_group(cache);
4134                 total -= num_bytes;
4135                 bytenr += num_bytes;
4136         }
4137         return 0;
4138 }
4139
4140 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4141 {
4142         struct btrfs_block_group_cache *cache;
4143         u64 bytenr;
4144
4145         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4146         if (!cache)
4147                 return 0;
4148
4149         bytenr = cache->key.objectid;
4150         btrfs_put_block_group(cache);
4151
4152         return bytenr;
4153 }
4154
4155 static int pin_down_extent(struct btrfs_root *root,
4156                            struct btrfs_block_group_cache *cache,
4157                            u64 bytenr, u64 num_bytes, int reserved)
4158 {
4159         spin_lock(&cache->space_info->lock);
4160         spin_lock(&cache->lock);
4161         cache->pinned += num_bytes;
4162         cache->space_info->bytes_pinned += num_bytes;
4163         if (reserved) {
4164                 cache->reserved -= num_bytes;
4165                 cache->space_info->bytes_reserved -= num_bytes;
4166         }
4167         spin_unlock(&cache->lock);
4168         spin_unlock(&cache->space_info->lock);
4169
4170         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4171                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4172         return 0;
4173 }
4174
4175 /*
4176  * this function must be called within transaction
4177  */
4178 int btrfs_pin_extent(struct btrfs_root *root,
4179                      u64 bytenr, u64 num_bytes, int reserved)
4180 {
4181         struct btrfs_block_group_cache *cache;
4182
4183         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4184         BUG_ON(!cache);
4185
4186         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4187
4188         btrfs_put_block_group(cache);
4189         return 0;
4190 }
4191
4192 /*
4193  * update size of reserved extents. this function may return -EAGAIN
4194  * if 'reserve' is true or 'sinfo' is false.
4195  */
4196 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4197                                  u64 num_bytes, int reserve, int sinfo)
4198 {
4199         int ret = 0;
4200         if (sinfo) {
4201                 struct btrfs_space_info *space_info = cache->space_info;
4202                 spin_lock(&space_info->lock);
4203                 spin_lock(&cache->lock);
4204                 if (reserve) {
4205                         if (cache->ro) {
4206                                 ret = -EAGAIN;
4207                         } else {
4208                                 cache->reserved += num_bytes;
4209                                 space_info->bytes_reserved += num_bytes;
4210                         }
4211                 } else {
4212                         if (cache->ro)
4213                                 space_info->bytes_readonly += num_bytes;
4214                         cache->reserved -= num_bytes;
4215                         space_info->bytes_reserved -= num_bytes;
4216                 }
4217                 spin_unlock(&cache->lock);
4218                 spin_unlock(&space_info->lock);
4219         } else {
4220                 spin_lock(&cache->lock);
4221                 if (cache->ro) {
4222                         ret = -EAGAIN;
4223                 } else {
4224                         if (reserve)
4225                                 cache->reserved += num_bytes;
4226                         else
4227                                 cache->reserved -= num_bytes;
4228                 }
4229                 spin_unlock(&cache->lock);
4230         }
4231         return ret;
4232 }
4233
4234 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4235                                 struct btrfs_root *root)
4236 {
4237         struct btrfs_fs_info *fs_info = root->fs_info;
4238         struct btrfs_caching_control *next;
4239         struct btrfs_caching_control *caching_ctl;
4240         struct btrfs_block_group_cache *cache;
4241
4242         down_write(&fs_info->extent_commit_sem);
4243
4244         list_for_each_entry_safe(caching_ctl, next,
4245                                  &fs_info->caching_block_groups, list) {
4246                 cache = caching_ctl->block_group;
4247                 if (block_group_cache_done(cache)) {
4248                         cache->last_byte_to_unpin = (u64)-1;
4249                         list_del_init(&caching_ctl->list);
4250                         put_caching_control(caching_ctl);
4251                 } else {
4252                         cache->last_byte_to_unpin = caching_ctl->progress;
4253                 }
4254         }
4255
4256         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4257                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4258         else
4259                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4260
4261         up_write(&fs_info->extent_commit_sem);
4262
4263         update_global_block_rsv(fs_info);
4264         return 0;
4265 }
4266
4267 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4268 {
4269         struct btrfs_fs_info *fs_info = root->fs_info;
4270         struct btrfs_block_group_cache *cache = NULL;
4271         u64 len;
4272
4273         while (start <= end) {
4274                 if (!cache ||
4275                     start >= cache->key.objectid + cache->key.offset) {
4276                         if (cache)
4277                                 btrfs_put_block_group(cache);
4278                         cache = btrfs_lookup_block_group(fs_info, start);
4279                         BUG_ON(!cache);
4280                 }
4281
4282                 len = cache->key.objectid + cache->key.offset - start;
4283                 len = min(len, end + 1 - start);
4284
4285                 if (start < cache->last_byte_to_unpin) {
4286                         len = min(len, cache->last_byte_to_unpin - start);
4287                         btrfs_add_free_space(cache, start, len);
4288                 }
4289
4290                 start += len;
4291
4292                 spin_lock(&cache->space_info->lock);
4293                 spin_lock(&cache->lock);
4294                 cache->pinned -= len;
4295                 cache->space_info->bytes_pinned -= len;
4296                 if (cache->ro) {
4297                         cache->space_info->bytes_readonly += len;
4298                 } else if (cache->reserved_pinned > 0) {
4299                         len = min(len, cache->reserved_pinned);
4300                         cache->reserved_pinned -= len;
4301                         cache->space_info->bytes_reserved += len;
4302                 }
4303                 spin_unlock(&cache->lock);
4304                 spin_unlock(&cache->space_info->lock);
4305         }
4306
4307         if (cache)
4308                 btrfs_put_block_group(cache);
4309         return 0;
4310 }
4311
4312 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4313                                struct btrfs_root *root)
4314 {
4315         struct btrfs_fs_info *fs_info = root->fs_info;
4316         struct extent_io_tree *unpin;
4317         struct btrfs_block_rsv *block_rsv;
4318         struct btrfs_block_rsv *next_rsv;
4319         u64 start;
4320         u64 end;
4321         int idx;
4322         int ret;
4323
4324         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4325                 unpin = &fs_info->freed_extents[1];
4326         else
4327                 unpin = &fs_info->freed_extents[0];
4328
4329         while (1) {
4330                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4331                                             EXTENT_DIRTY);
4332                 if (ret)
4333                         break;
4334
4335                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4336
4337                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4338                 unpin_extent_range(root, start, end);
4339                 cond_resched();
4340         }
4341
4342         mutex_lock(&fs_info->durable_block_rsv_mutex);
4343         list_for_each_entry_safe(block_rsv, next_rsv,
4344                                  &fs_info->durable_block_rsv_list, list) {
4345
4346                 idx = trans->transid & 0x1;
4347                 if (block_rsv->freed[idx] > 0) {
4348                         block_rsv_add_bytes(block_rsv,
4349                                             block_rsv->freed[idx], 0);
4350                         block_rsv->freed[idx] = 0;
4351                 }
4352                 if (atomic_read(&block_rsv->usage) == 0) {
4353                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4354
4355                         if (block_rsv->freed[0] == 0 &&
4356                             block_rsv->freed[1] == 0) {
4357                                 list_del_init(&block_rsv->list);
4358                                 kfree(block_rsv);
4359                         }
4360                 } else {
4361                         btrfs_block_rsv_release(root, block_rsv, 0);
4362                 }
4363         }
4364         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4365
4366         return 0;
4367 }
4368
4369 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4370                                 struct btrfs_root *root,
4371                                 u64 bytenr, u64 num_bytes, u64 parent,
4372                                 u64 root_objectid, u64 owner_objectid,
4373                                 u64 owner_offset, int refs_to_drop,
4374                                 struct btrfs_delayed_extent_op *extent_op)
4375 {
4376         struct btrfs_key key;
4377         struct btrfs_path *path;
4378         struct btrfs_fs_info *info = root->fs_info;
4379         struct btrfs_root *extent_root = info->extent_root;
4380         struct extent_buffer *leaf;
4381         struct btrfs_extent_item *ei;
4382         struct btrfs_extent_inline_ref *iref;
4383         int ret;
4384         int is_data;
4385         int extent_slot = 0;
4386         int found_extent = 0;
4387         int num_to_del = 1;
4388         u32 item_size;
4389         u64 refs;
4390
4391         path = btrfs_alloc_path();
4392         if (!path)
4393                 return -ENOMEM;
4394
4395         path->reada = 1;
4396         path->leave_spinning = 1;
4397
4398         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4399         BUG_ON(!is_data && refs_to_drop != 1);
4400
4401         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4402                                     bytenr, num_bytes, parent,
4403                                     root_objectid, owner_objectid,
4404                                     owner_offset);
4405         if (ret == 0) {
4406                 extent_slot = path->slots[0];
4407                 while (extent_slot >= 0) {
4408                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4409                                               extent_slot);
4410                         if (key.objectid != bytenr)
4411                                 break;
4412                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4413                             key.offset == num_bytes) {
4414                                 found_extent = 1;
4415                                 break;
4416                         }
4417                         if (path->slots[0] - extent_slot > 5)
4418                                 break;
4419                         extent_slot--;
4420                 }
4421 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4422                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4423                 if (found_extent && item_size < sizeof(*ei))
4424                         found_extent = 0;
4425 #endif
4426                 if (!found_extent) {
4427                         BUG_ON(iref);
4428                         ret = remove_extent_backref(trans, extent_root, path,
4429                                                     NULL, refs_to_drop,
4430                                                     is_data);
4431                         BUG_ON(ret);
4432                         btrfs_release_path(extent_root, path);
4433                         path->leave_spinning = 1;
4434
4435                         key.objectid = bytenr;
4436                         key.type = BTRFS_EXTENT_ITEM_KEY;
4437                         key.offset = num_bytes;
4438
4439                         ret = btrfs_search_slot(trans, extent_root,
4440                                                 &key, path, -1, 1);
4441                         if (ret) {
4442                                 printk(KERN_ERR "umm, got %d back from search"
4443                                        ", was looking for %llu\n", ret,
4444                                        (unsigned long long)bytenr);
4445                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4446                         }
4447                         BUG_ON(ret);
4448                         extent_slot = path->slots[0];
4449                 }
4450         } else {
4451                 btrfs_print_leaf(extent_root, path->nodes[0]);
4452                 WARN_ON(1);
4453                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4454                        "parent %llu root %llu  owner %llu offset %llu\n",
4455                        (unsigned long long)bytenr,
4456                        (unsigned long long)parent,
4457                        (unsigned long long)root_objectid,
4458                        (unsigned long long)owner_objectid,
4459                        (unsigned long long)owner_offset);
4460         }
4461
4462         leaf = path->nodes[0];
4463         item_size = btrfs_item_size_nr(leaf, extent_slot);
4464 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4465         if (item_size < sizeof(*ei)) {
4466                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4467                 ret = convert_extent_item_v0(trans, extent_root, path,
4468                                              owner_objectid, 0);
4469                 BUG_ON(ret < 0);
4470
4471                 btrfs_release_path(extent_root, path);
4472                 path->leave_spinning = 1;
4473
4474                 key.objectid = bytenr;
4475                 key.type = BTRFS_EXTENT_ITEM_KEY;
4476                 key.offset = num_bytes;
4477
4478                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4479                                         -1, 1);
4480                 if (ret) {
4481                         printk(KERN_ERR "umm, got %d back from search"
4482                                ", was looking for %llu\n", ret,
4483                                (unsigned long long)bytenr);
4484                         btrfs_print_leaf(extent_root, path->nodes[0]);
4485                 }
4486                 BUG_ON(ret);
4487                 extent_slot = path->slots[0];
4488                 leaf = path->nodes[0];
4489                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4490         }
4491 #endif
4492         BUG_ON(item_size < sizeof(*ei));
4493         ei = btrfs_item_ptr(leaf, extent_slot,
4494                             struct btrfs_extent_item);
4495         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4496                 struct btrfs_tree_block_info *bi;
4497                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4498                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4499                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4500         }
4501
4502         refs = btrfs_extent_refs(leaf, ei);
4503         BUG_ON(refs < refs_to_drop);
4504         refs -= refs_to_drop;
4505
4506         if (refs > 0) {
4507                 if (extent_op)
4508                         __run_delayed_extent_op(extent_op, leaf, ei);
4509                 /*
4510                  * In the case of inline back ref, reference count will
4511                  * be updated by remove_extent_backref
4512                  */
4513                 if (iref) {
4514                         BUG_ON(!found_extent);
4515                 } else {
4516                         btrfs_set_extent_refs(leaf, ei, refs);
4517                         btrfs_mark_buffer_dirty(leaf);
4518                 }
4519                 if (found_extent) {
4520                         ret = remove_extent_backref(trans, extent_root, path,
4521                                                     iref, refs_to_drop,
4522                                                     is_data);
4523                         BUG_ON(ret);
4524                 }
4525         } else {
4526                 if (found_extent) {
4527                         BUG_ON(is_data && refs_to_drop !=
4528                                extent_data_ref_count(root, path, iref));
4529                         if (iref) {
4530                                 BUG_ON(path->slots[0] != extent_slot);
4531                         } else {
4532                                 BUG_ON(path->slots[0] != extent_slot + 1);
4533                                 path->slots[0] = extent_slot;
4534                                 num_to_del = 2;
4535                         }
4536                 }
4537
4538                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4539                                       num_to_del);
4540                 BUG_ON(ret);
4541                 btrfs_release_path(extent_root, path);
4542
4543                 if (is_data) {
4544                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4545                         BUG_ON(ret);
4546                 } else {
4547                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4548                              bytenr >> PAGE_CACHE_SHIFT,
4549                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4550                 }
4551
4552                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4553                 BUG_ON(ret);
4554         }
4555         btrfs_free_path(path);
4556         return ret;
4557 }
4558
4559 /*
4560  * when we free an block, it is possible (and likely) that we free the last
4561  * delayed ref for that extent as well.  This searches the delayed ref tree for
4562  * a given extent, and if there are no other delayed refs to be processed, it
4563  * removes it from the tree.
4564  */
4565 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4566                                       struct btrfs_root *root, u64 bytenr)
4567 {
4568         struct btrfs_delayed_ref_head *head;
4569         struct btrfs_delayed_ref_root *delayed_refs;
4570         struct btrfs_delayed_ref_node *ref;
4571         struct rb_node *node;
4572         int ret = 0;
4573
4574         delayed_refs = &trans->transaction->delayed_refs;
4575         spin_lock(&delayed_refs->lock);
4576         head = btrfs_find_delayed_ref_head(trans, bytenr);
4577         if (!head)
4578                 goto out;
4579
4580         node = rb_prev(&head->node.rb_node);
4581         if (!node)
4582                 goto out;
4583
4584         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4585
4586         /* there are still entries for this ref, we can't drop it */
4587         if (ref->bytenr == bytenr)
4588                 goto out;
4589
4590         if (head->extent_op) {
4591                 if (!head->must_insert_reserved)
4592                         goto out;
4593                 kfree(head->extent_op);
4594                 head->extent_op = NULL;
4595         }
4596
4597         /*
4598          * waiting for the lock here would deadlock.  If someone else has it
4599          * locked they are already in the process of dropping it anyway
4600          */
4601         if (!mutex_trylock(&head->mutex))
4602                 goto out;
4603
4604         /*
4605          * at this point we have a head with no other entries.  Go
4606          * ahead and process it.
4607          */
4608         head->node.in_tree = 0;
4609         rb_erase(&head->node.rb_node, &delayed_refs->root);
4610
4611         delayed_refs->num_entries--;
4612
4613         /*
4614          * we don't take a ref on the node because we're removing it from the
4615          * tree, so we just steal the ref the tree was holding.
4616          */
4617         delayed_refs->num_heads--;
4618         if (list_empty(&head->cluster))
4619                 delayed_refs->num_heads_ready--;
4620
4621         list_del_init(&head->cluster);
4622         spin_unlock(&delayed_refs->lock);
4623
4624         BUG_ON(head->extent_op);
4625         if (head->must_insert_reserved)
4626                 ret = 1;
4627
4628         mutex_unlock(&head->mutex);
4629         btrfs_put_delayed_ref(&head->node);
4630         return ret;
4631 out:
4632         spin_unlock(&delayed_refs->lock);
4633         return 0;
4634 }
4635
4636 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4637                            struct btrfs_root *root,
4638                            struct extent_buffer *buf,
4639                            u64 parent, int last_ref)
4640 {
4641         struct btrfs_block_rsv *block_rsv;
4642         struct btrfs_block_group_cache *cache = NULL;
4643         int ret;
4644
4645         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4646                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4647                                                 parent, root->root_key.objectid,
4648                                                 btrfs_header_level(buf),
4649                                                 BTRFS_DROP_DELAYED_REF, NULL);
4650                 BUG_ON(ret);
4651         }
4652
4653         if (!last_ref)
4654                 return;
4655
4656         block_rsv = get_block_rsv(trans, root);
4657         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4658         if (block_rsv->space_info != cache->space_info)
4659                 goto out;
4660
4661         if (btrfs_header_generation(buf) == trans->transid) {
4662                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4663                         ret = check_ref_cleanup(trans, root, buf->start);
4664                         if (!ret)
4665                                 goto pin;
4666                 }
4667
4668                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4669                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4670                         goto pin;
4671                 }
4672
4673                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4674
4675                 btrfs_add_free_space(cache, buf->start, buf->len);
4676                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4677                 if (ret == -EAGAIN) {
4678                         /* block group became read-only */
4679                         update_reserved_bytes(cache, buf->len, 0, 1);
4680                         goto out;
4681                 }
4682
4683                 ret = 1;
4684                 spin_lock(&block_rsv->lock);
4685                 if (block_rsv->reserved < block_rsv->size) {
4686                         block_rsv->reserved += buf->len;
4687                         ret = 0;
4688                 }
4689                 spin_unlock(&block_rsv->lock);
4690
4691                 if (ret) {
4692                         spin_lock(&cache->space_info->lock);
4693                         cache->space_info->bytes_reserved -= buf->len;
4694                         spin_unlock(&cache->space_info->lock);
4695                 }
4696                 goto out;
4697         }
4698 pin:
4699         if (block_rsv->durable && !cache->ro) {
4700                 ret = 0;
4701                 spin_lock(&cache->lock);
4702                 if (!cache->ro) {
4703                         cache->reserved_pinned += buf->len;
4704                         ret = 1;
4705                 }
4706                 spin_unlock(&cache->lock);
4707
4708                 if (ret) {
4709                         spin_lock(&block_rsv->lock);
4710                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4711                         spin_unlock(&block_rsv->lock);
4712                 }
4713         }
4714 out:
4715         btrfs_put_block_group(cache);
4716 }
4717
4718 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4719                       struct btrfs_root *root,
4720                       u64 bytenr, u64 num_bytes, u64 parent,
4721                       u64 root_objectid, u64 owner, u64 offset)
4722 {
4723         int ret;
4724
4725         /*
4726          * tree log blocks never actually go into the extent allocation
4727          * tree, just update pinning info and exit early.
4728          */
4729         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4730                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4731                 /* unlocks the pinned mutex */
4732                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4733                 ret = 0;
4734         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4735                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4736                                         parent, root_objectid, (int)owner,
4737                                         BTRFS_DROP_DELAYED_REF, NULL);
4738                 BUG_ON(ret);
4739         } else {
4740                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4741                                         parent, root_objectid, owner,
4742                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4743                 BUG_ON(ret);
4744         }
4745         return ret;
4746 }
4747
4748 static u64 stripe_align(struct btrfs_root *root, u64 val)
4749 {
4750         u64 mask = ((u64)root->stripesize - 1);
4751         u64 ret = (val + mask) & ~mask;
4752         return ret;
4753 }
4754
4755 /*
4756  * when we wait for progress in the block group caching, its because
4757  * our allocation attempt failed at least once.  So, we must sleep
4758  * and let some progress happen before we try again.
4759  *
4760  * This function will sleep at least once waiting for new free space to
4761  * show up, and then it will check the block group free space numbers
4762  * for our min num_bytes.  Another option is to have it go ahead
4763  * and look in the rbtree for a free extent of a given size, but this
4764  * is a good start.
4765  */
4766 static noinline int
4767 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4768                                 u64 num_bytes)
4769 {
4770         struct btrfs_caching_control *caching_ctl;
4771         DEFINE_WAIT(wait);
4772
4773         caching_ctl = get_caching_control(cache);
4774         if (!caching_ctl)
4775                 return 0;
4776
4777         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4778                    (cache->free_space >= num_bytes));
4779
4780         put_caching_control(caching_ctl);
4781         return 0;
4782 }
4783
4784 static noinline int
4785 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4786 {
4787         struct btrfs_caching_control *caching_ctl;
4788         DEFINE_WAIT(wait);
4789
4790         caching_ctl = get_caching_control(cache);
4791         if (!caching_ctl)
4792                 return 0;
4793
4794         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4795
4796         put_caching_control(caching_ctl);
4797         return 0;
4798 }
4799
4800 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4801 {
4802         int index;
4803         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4804                 index = 0;
4805         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4806                 index = 1;
4807         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4808                 index = 2;
4809         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4810                 index = 3;
4811         else
4812                 index = 4;
4813         return index;
4814 }
4815
4816 enum btrfs_loop_type {
4817         LOOP_FIND_IDEAL = 0,
4818         LOOP_CACHING_NOWAIT = 1,
4819         LOOP_CACHING_WAIT = 2,
4820         LOOP_ALLOC_CHUNK = 3,
4821         LOOP_NO_EMPTY_SIZE = 4,
4822 };
4823
4824 /*
4825  * walks the btree of allocated extents and find a hole of a given size.
4826  * The key ins is changed to record the hole:
4827  * ins->objectid == block start
4828  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4829  * ins->offset == number of blocks
4830  * Any available blocks before search_start are skipped.
4831  */
4832 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4833                                      struct btrfs_root *orig_root,
4834                                      u64 num_bytes, u64 empty_size,
4835                                      u64 search_start, u64 search_end,
4836                                      u64 hint_byte, struct btrfs_key *ins,
4837                                      int data)
4838 {
4839         int ret = 0;
4840         struct btrfs_root *root = orig_root->fs_info->extent_root;
4841         struct btrfs_free_cluster *last_ptr = NULL;
4842         struct btrfs_block_group_cache *block_group = NULL;
4843         int empty_cluster = 2 * 1024 * 1024;
4844         int allowed_chunk_alloc = 0;
4845         int done_chunk_alloc = 0;
4846         struct btrfs_space_info *space_info;
4847         int last_ptr_loop = 0;
4848         int loop = 0;
4849         int index = 0;
4850         bool found_uncached_bg = false;
4851         bool failed_cluster_refill = false;
4852         bool failed_alloc = false;
4853         bool use_cluster = true;
4854         u64 ideal_cache_percent = 0;
4855         u64 ideal_cache_offset = 0;
4856
4857         WARN_ON(num_bytes < root->sectorsize);
4858         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4859         ins->objectid = 0;
4860         ins->offset = 0;
4861
4862         space_info = __find_space_info(root->fs_info, data);
4863         if (!space_info) {
4864                 printk(KERN_ERR "No space info for %d\n", data);
4865                 return -ENOSPC;
4866         }
4867
4868         /*
4869          * If the space info is for both data and metadata it means we have a
4870          * small filesystem and we can't use the clustering stuff.
4871          */
4872         if (btrfs_mixed_space_info(space_info))
4873                 use_cluster = false;
4874
4875         if (orig_root->ref_cows || empty_size)
4876                 allowed_chunk_alloc = 1;
4877
4878         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4879                 last_ptr = &root->fs_info->meta_alloc_cluster;
4880                 if (!btrfs_test_opt(root, SSD))
4881                         empty_cluster = 64 * 1024;
4882         }
4883
4884         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4885             btrfs_test_opt(root, SSD)) {
4886                 last_ptr = &root->fs_info->data_alloc_cluster;
4887         }
4888
4889         if (last_ptr) {
4890                 spin_lock(&last_ptr->lock);
4891                 if (last_ptr->block_group)
4892                         hint_byte = last_ptr->window_start;
4893                 spin_unlock(&last_ptr->lock);
4894         }
4895
4896         search_start = max(search_start, first_logical_byte(root, 0));
4897         search_start = max(search_start, hint_byte);
4898
4899         if (!last_ptr)
4900                 empty_cluster = 0;
4901
4902         if (search_start == hint_byte) {
4903 ideal_cache:
4904                 block_group = btrfs_lookup_block_group(root->fs_info,
4905                                                        search_start);
4906                 /*
4907                  * we don't want to use the block group if it doesn't match our
4908                  * allocation bits, or if its not cached.
4909                  *
4910                  * However if we are re-searching with an ideal block group
4911                  * picked out then we don't care that the block group is cached.
4912                  */
4913                 if (block_group && block_group_bits(block_group, data) &&
4914                     (block_group->cached != BTRFS_CACHE_NO ||
4915                      search_start == ideal_cache_offset)) {
4916                         down_read(&space_info->groups_sem);
4917                         if (list_empty(&block_group->list) ||
4918                             block_group->ro) {
4919                                 /*
4920                                  * someone is removing this block group,
4921                                  * we can't jump into the have_block_group
4922                                  * target because our list pointers are not
4923                                  * valid
4924                                  */
4925                                 btrfs_put_block_group(block_group);
4926                                 up_read(&space_info->groups_sem);
4927                         } else {
4928                                 index = get_block_group_index(block_group);
4929                                 goto have_block_group;
4930                         }
4931                 } else if (block_group) {
4932                         btrfs_put_block_group(block_group);
4933                 }
4934         }
4935 search:
4936         down_read(&space_info->groups_sem);
4937         list_for_each_entry(block_group, &space_info->block_groups[index],
4938                             list) {
4939                 u64 offset;
4940                 int cached;
4941
4942                 btrfs_get_block_group(block_group);
4943                 search_start = block_group->key.objectid;
4944
4945                 /*
4946                  * this can happen if we end up cycling through all the
4947                  * raid types, but we want to make sure we only allocate
4948                  * for the proper type.
4949                  */
4950                 if (!block_group_bits(block_group, data)) {
4951                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
4952                                 BTRFS_BLOCK_GROUP_RAID1 |
4953                                 BTRFS_BLOCK_GROUP_RAID10;
4954
4955                         /*
4956                          * if they asked for extra copies and this block group
4957                          * doesn't provide them, bail.  This does allow us to
4958                          * fill raid0 from raid1.
4959                          */
4960                         if ((data & extra) && !(block_group->flags & extra))
4961                                 goto loop;
4962                 }
4963
4964 have_block_group:
4965                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4966                         u64 free_percent;
4967
4968                         ret = cache_block_group(block_group, trans,
4969                                                 orig_root, 1);
4970                         if (block_group->cached == BTRFS_CACHE_FINISHED)
4971                                 goto have_block_group;
4972
4973                         free_percent = btrfs_block_group_used(&block_group->item);
4974                         free_percent *= 100;
4975                         free_percent = div64_u64(free_percent,
4976                                                  block_group->key.offset);
4977                         free_percent = 100 - free_percent;
4978                         if (free_percent > ideal_cache_percent &&
4979                             likely(!block_group->ro)) {
4980                                 ideal_cache_offset = block_group->key.objectid;
4981                                 ideal_cache_percent = free_percent;
4982                         }
4983
4984                         /*
4985                          * We only want to start kthread caching if we are at
4986                          * the point where we will wait for caching to make
4987                          * progress, or if our ideal search is over and we've
4988                          * found somebody to start caching.
4989                          */
4990                         if (loop > LOOP_CACHING_NOWAIT ||
4991                             (loop > LOOP_FIND_IDEAL &&
4992                              atomic_read(&space_info->caching_threads) < 2)) {
4993                                 ret = cache_block_group(block_group, trans,
4994                                                         orig_root, 0);
4995                                 BUG_ON(ret);
4996                         }
4997                         found_uncached_bg = true;
4998
4999                         /*
5000                          * If loop is set for cached only, try the next block
5001                          * group.
5002                          */
5003                         if (loop == LOOP_FIND_IDEAL)
5004                                 goto loop;
5005                 }
5006
5007                 cached = block_group_cache_done(block_group);
5008                 if (unlikely(!cached))
5009                         found_uncached_bg = true;
5010
5011                 if (unlikely(block_group->ro))
5012                         goto loop;
5013
5014                 /*
5015                  * Ok we want to try and use the cluster allocator, so lets look
5016                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5017                  * have tried the cluster allocator plenty of times at this
5018                  * point and not have found anything, so we are likely way too
5019                  * fragmented for the clustering stuff to find anything, so lets
5020                  * just skip it and let the allocator find whatever block it can
5021                  * find
5022                  */
5023                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5024                         /*
5025                          * the refill lock keeps out other
5026                          * people trying to start a new cluster
5027                          */
5028                         spin_lock(&last_ptr->refill_lock);
5029                         if (last_ptr->block_group &&
5030                             (last_ptr->block_group->ro ||
5031                             !block_group_bits(last_ptr->block_group, data))) {
5032                                 offset = 0;
5033                                 goto refill_cluster;
5034                         }
5035
5036                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5037                                                  num_bytes, search_start);
5038                         if (offset) {
5039                                 /* we have a block, we're done */
5040                                 spin_unlock(&last_ptr->refill_lock);
5041                                 goto checks;
5042                         }
5043
5044                         spin_lock(&last_ptr->lock);
5045                         /*
5046                          * whoops, this cluster doesn't actually point to
5047                          * this block group.  Get a ref on the block
5048                          * group is does point to and try again
5049                          */
5050                         if (!last_ptr_loop && last_ptr->block_group &&
5051                             last_ptr->block_group != block_group) {
5052
5053                                 btrfs_put_block_group(block_group);
5054                                 block_group = last_ptr->block_group;
5055                                 btrfs_get_block_group(block_group);
5056                                 spin_unlock(&last_ptr->lock);
5057                                 spin_unlock(&last_ptr->refill_lock);
5058
5059                                 last_ptr_loop = 1;
5060                                 search_start = block_group->key.objectid;
5061                                 /*
5062                                  * we know this block group is properly
5063                                  * in the list because
5064                                  * btrfs_remove_block_group, drops the
5065                                  * cluster before it removes the block
5066                                  * group from the list
5067                                  */
5068                                 goto have_block_group;
5069                         }
5070                         spin_unlock(&last_ptr->lock);
5071 refill_cluster:
5072                         /*
5073                          * this cluster didn't work out, free it and
5074                          * start over
5075                          */
5076                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5077
5078                         last_ptr_loop = 0;
5079
5080                         /* allocate a cluster in this block group */
5081                         ret = btrfs_find_space_cluster(trans, root,
5082                                                block_group, last_ptr,
5083                                                offset, num_bytes,
5084                                                empty_cluster + empty_size);
5085                         if (ret == 0) {
5086                                 /*
5087                                  * now pull our allocation out of this
5088                                  * cluster
5089                                  */
5090                                 offset = btrfs_alloc_from_cluster(block_group,
5091                                                   last_ptr, num_bytes,
5092                                                   search_start);
5093                                 if (offset) {
5094                                         /* we found one, proceed */
5095                                         spin_unlock(&last_ptr->refill_lock);
5096                                         goto checks;
5097                                 }
5098                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5099                                    && !failed_cluster_refill) {
5100                                 spin_unlock(&last_ptr->refill_lock);
5101
5102                                 failed_cluster_refill = true;
5103                                 wait_block_group_cache_progress(block_group,
5104                                        num_bytes + empty_cluster + empty_size);
5105                                 goto have_block_group;
5106                         }
5107
5108                         /*
5109                          * at this point we either didn't find a cluster
5110                          * or we weren't able to allocate a block from our
5111                          * cluster.  Free the cluster we've been trying
5112                          * to use, and go to the next block group
5113                          */
5114                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5115                         spin_unlock(&last_ptr->refill_lock);
5116                         goto loop;
5117                 }
5118
5119                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5120                                                     num_bytes, empty_size);
5121                 /*
5122                  * If we didn't find a chunk, and we haven't failed on this
5123                  * block group before, and this block group is in the middle of
5124                  * caching and we are ok with waiting, then go ahead and wait
5125                  * for progress to be made, and set failed_alloc to true.
5126                  *
5127                  * If failed_alloc is true then we've already waited on this
5128                  * block group once and should move on to the next block group.
5129                  */
5130                 if (!offset && !failed_alloc && !cached &&
5131                     loop > LOOP_CACHING_NOWAIT) {
5132                         wait_block_group_cache_progress(block_group,
5133                                                 num_bytes + empty_size);
5134                         failed_alloc = true;
5135                         goto have_block_group;
5136                 } else if (!offset) {
5137                         goto loop;
5138                 }
5139 checks:
5140                 search_start = stripe_align(root, offset);
5141                 /* move on to the next group */
5142                 if (search_start + num_bytes >= search_end) {
5143                         btrfs_add_free_space(block_group, offset, num_bytes);
5144                         goto loop;
5145                 }
5146
5147                 /* move on to the next group */
5148                 if (search_start + num_bytes >
5149                     block_group->key.objectid + block_group->key.offset) {
5150                         btrfs_add_free_space(block_group, offset, num_bytes);
5151                         goto loop;
5152                 }
5153
5154                 ins->objectid = search_start;
5155                 ins->offset = num_bytes;
5156
5157                 if (offset < search_start)
5158                         btrfs_add_free_space(block_group, offset,
5159                                              search_start - offset);
5160                 BUG_ON(offset > search_start);
5161
5162                 ret = update_reserved_bytes(block_group, num_bytes, 1,
5163                                             (data & BTRFS_BLOCK_GROUP_DATA));
5164                 if (ret == -EAGAIN) {
5165                         btrfs_add_free_space(block_group, offset, num_bytes);
5166                         goto loop;
5167                 }
5168
5169                 /* we are all good, lets return */
5170                 ins->objectid = search_start;
5171                 ins->offset = num_bytes;
5172
5173                 if (offset < search_start)
5174                         btrfs_add_free_space(block_group, offset,
5175                                              search_start - offset);
5176                 BUG_ON(offset > search_start);
5177                 break;
5178 loop:
5179                 failed_cluster_refill = false;
5180                 failed_alloc = false;
5181                 BUG_ON(index != get_block_group_index(block_group));
5182                 btrfs_put_block_group(block_group);
5183         }
5184         up_read(&space_info->groups_sem);
5185
5186         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5187                 goto search;
5188
5189         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5190          *                      for them to make caching progress.  Also
5191          *                      determine the best possible bg to cache
5192          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5193          *                      caching kthreads as we move along
5194          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5195          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5196          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5197          *                      again
5198          */
5199         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5200             (found_uncached_bg || empty_size || empty_cluster ||
5201              allowed_chunk_alloc)) {
5202                 index = 0;
5203                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5204                         found_uncached_bg = false;
5205                         loop++;
5206                         if (!ideal_cache_percent &&
5207                             atomic_read(&space_info->caching_threads))
5208                                 goto search;
5209
5210                         /*
5211                          * 1 of the following 2 things have happened so far
5212                          *
5213                          * 1) We found an ideal block group for caching that
5214                          * is mostly full and will cache quickly, so we might
5215                          * as well wait for it.
5216                          *
5217                          * 2) We searched for cached only and we didn't find
5218                          * anything, and we didn't start any caching kthreads
5219                          * either, so chances are we will loop through and
5220                          * start a couple caching kthreads, and then come back
5221                          * around and just wait for them.  This will be slower
5222                          * because we will have 2 caching kthreads reading at
5223                          * the same time when we could have just started one
5224                          * and waited for it to get far enough to give us an
5225                          * allocation, so go ahead and go to the wait caching
5226                          * loop.
5227                          */
5228                         loop = LOOP_CACHING_WAIT;
5229                         search_start = ideal_cache_offset;
5230                         ideal_cache_percent = 0;
5231                         goto ideal_cache;
5232                 } else if (loop == LOOP_FIND_IDEAL) {
5233                         /*
5234                          * Didn't find a uncached bg, wait on anything we find
5235                          * next.
5236                          */
5237                         loop = LOOP_CACHING_WAIT;
5238                         goto search;
5239                 }
5240
5241                 if (loop < LOOP_CACHING_WAIT) {
5242                         loop++;
5243                         goto search;
5244                 }
5245
5246                 if (loop == LOOP_ALLOC_CHUNK) {
5247                         empty_size = 0;
5248                         empty_cluster = 0;
5249                 }
5250
5251                 if (allowed_chunk_alloc) {
5252                         ret = do_chunk_alloc(trans, root, num_bytes +
5253                                              2 * 1024 * 1024, data, 1);
5254                         allowed_chunk_alloc = 0;
5255                         done_chunk_alloc = 1;
5256                 } else if (!done_chunk_alloc) {
5257                         space_info->force_alloc = 1;
5258                 }
5259
5260                 if (loop < LOOP_NO_EMPTY_SIZE) {
5261                         loop++;
5262                         goto search;
5263                 }
5264                 ret = -ENOSPC;
5265         } else if (!ins->objectid) {
5266                 ret = -ENOSPC;
5267         }
5268
5269         /* we found what we needed */
5270         if (ins->objectid) {
5271                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5272                         trans->block_group = block_group->key.objectid;
5273
5274                 btrfs_put_block_group(block_group);
5275                 ret = 0;
5276         }
5277
5278         return ret;
5279 }
5280
5281 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5282                             int dump_block_groups)
5283 {
5284         struct btrfs_block_group_cache *cache;
5285         int index = 0;
5286
5287         spin_lock(&info->lock);
5288         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5289                (unsigned long long)(info->total_bytes - info->bytes_used -
5290                                     info->bytes_pinned - info->bytes_reserved -
5291                                     info->bytes_readonly),
5292                (info->full) ? "" : "not ");
5293         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5294                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5295                (unsigned long long)info->total_bytes,
5296                (unsigned long long)info->bytes_used,
5297                (unsigned long long)info->bytes_pinned,
5298                (unsigned long long)info->bytes_reserved,
5299                (unsigned long long)info->bytes_may_use,
5300                (unsigned long long)info->bytes_readonly);
5301         spin_unlock(&info->lock);
5302
5303         if (!dump_block_groups)
5304                 return;
5305
5306         down_read(&info->groups_sem);
5307 again:
5308         list_for_each_entry(cache, &info->block_groups[index], list) {
5309                 spin_lock(&cache->lock);
5310                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5311                        "%llu pinned %llu reserved\n",
5312                        (unsigned long long)cache->key.objectid,
5313                        (unsigned long long)cache->key.offset,
5314                        (unsigned long long)btrfs_block_group_used(&cache->item),
5315                        (unsigned long long)cache->pinned,
5316                        (unsigned long long)cache->reserved);
5317                 btrfs_dump_free_space(cache, bytes);
5318                 spin_unlock(&cache->lock);
5319         }
5320         if (++index < BTRFS_NR_RAID_TYPES)
5321                 goto again;
5322         up_read(&info->groups_sem);
5323 }
5324
5325 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5326                          struct btrfs_root *root,
5327                          u64 num_bytes, u64 min_alloc_size,
5328                          u64 empty_size, u64 hint_byte,
5329                          u64 search_end, struct btrfs_key *ins,
5330                          u64 data)
5331 {
5332         int ret;
5333         u64 search_start = 0;
5334
5335         data = btrfs_get_alloc_profile(root, data);
5336 again:
5337         /*
5338          * the only place that sets empty_size is btrfs_realloc_node, which
5339          * is not called recursively on allocations
5340          */
5341         if (empty_size || root->ref_cows)
5342                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5343                                      num_bytes + 2 * 1024 * 1024, data, 0);
5344
5345         WARN_ON(num_bytes < root->sectorsize);
5346         ret = find_free_extent(trans, root, num_bytes, empty_size,
5347                                search_start, search_end, hint_byte,
5348                                ins, data);
5349
5350         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5351                 num_bytes = num_bytes >> 1;
5352                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5353                 num_bytes = max(num_bytes, min_alloc_size);
5354                 do_chunk_alloc(trans, root->fs_info->extent_root,
5355                                num_bytes, data, 1);
5356                 goto again;
5357         }
5358         if (ret == -ENOSPC) {
5359                 struct btrfs_space_info *sinfo;
5360
5361                 sinfo = __find_space_info(root->fs_info, data);
5362                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5363                        "wanted %llu\n", (unsigned long long)data,
5364                        (unsigned long long)num_bytes);
5365                 dump_space_info(sinfo, num_bytes, 1);
5366         }
5367
5368         return ret;
5369 }
5370
5371 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5372 {
5373         struct btrfs_block_group_cache *cache;
5374         int ret = 0;
5375
5376         cache = btrfs_lookup_block_group(root->fs_info, start);
5377         if (!cache) {
5378                 printk(KERN_ERR "Unable to find block group for %llu\n",
5379                        (unsigned long long)start);
5380                 return -ENOSPC;
5381         }
5382
5383         ret = btrfs_discard_extent(root, start, len);
5384
5385         btrfs_add_free_space(cache, start, len);
5386         update_reserved_bytes(cache, len, 0, 1);
5387         btrfs_put_block_group(cache);
5388
5389         return ret;
5390 }
5391
5392 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5393                                       struct btrfs_root *root,
5394                                       u64 parent, u64 root_objectid,
5395                                       u64 flags, u64 owner, u64 offset,
5396                                       struct btrfs_key *ins, int ref_mod)
5397 {
5398         int ret;
5399         struct btrfs_fs_info *fs_info = root->fs_info;
5400         struct btrfs_extent_item *extent_item;
5401         struct btrfs_extent_inline_ref *iref;
5402         struct btrfs_path *path;
5403         struct extent_buffer *leaf;
5404         int type;
5405         u32 size;
5406
5407         if (parent > 0)
5408                 type = BTRFS_SHARED_DATA_REF_KEY;
5409         else
5410                 type = BTRFS_EXTENT_DATA_REF_KEY;
5411
5412         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5413
5414         path = btrfs_alloc_path();
5415         BUG_ON(!path);
5416
5417         path->leave_spinning = 1;
5418         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5419                                       ins, size);
5420         BUG_ON(ret);
5421
5422         leaf = path->nodes[0];
5423         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5424                                      struct btrfs_extent_item);
5425         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5426         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5427         btrfs_set_extent_flags(leaf, extent_item,
5428                                flags | BTRFS_EXTENT_FLAG_DATA);
5429
5430         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5431         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5432         if (parent > 0) {
5433                 struct btrfs_shared_data_ref *ref;
5434                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5435                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5436                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5437         } else {
5438                 struct btrfs_extent_data_ref *ref;
5439                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5440                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5441                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5442                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5443                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5444         }
5445
5446         btrfs_mark_buffer_dirty(path->nodes[0]);
5447         btrfs_free_path(path);
5448
5449         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5450         if (ret) {
5451                 printk(KERN_ERR "btrfs update block group failed for %llu "
5452                        "%llu\n", (unsigned long long)ins->objectid,
5453                        (unsigned long long)ins->offset);
5454                 BUG();
5455         }
5456         return ret;
5457 }
5458
5459 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5460                                      struct btrfs_root *root,
5461                                      u64 parent, u64 root_objectid,
5462                                      u64 flags, struct btrfs_disk_key *key,
5463                                      int level, struct btrfs_key *ins)
5464 {
5465         int ret;
5466         struct btrfs_fs_info *fs_info = root->fs_info;
5467         struct btrfs_extent_item *extent_item;
5468         struct btrfs_tree_block_info *block_info;
5469         struct btrfs_extent_inline_ref *iref;
5470         struct btrfs_path *path;
5471         struct extent_buffer *leaf;
5472         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5473
5474         path = btrfs_alloc_path();
5475         BUG_ON(!path);
5476
5477         path->leave_spinning = 1;
5478         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5479                                       ins, size);
5480         BUG_ON(ret);
5481
5482         leaf = path->nodes[0];
5483         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5484                                      struct btrfs_extent_item);
5485         btrfs_set_extent_refs(leaf, extent_item, 1);
5486         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5487         btrfs_set_extent_flags(leaf, extent_item,
5488                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5489         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5490
5491         btrfs_set_tree_block_key(leaf, block_info, key);
5492         btrfs_set_tree_block_level(leaf, block_info, level);
5493
5494         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5495         if (parent > 0) {
5496                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5497                 btrfs_set_extent_inline_ref_type(leaf, iref,
5498                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5499                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5500         } else {
5501                 btrfs_set_extent_inline_ref_type(leaf, iref,
5502                                                  BTRFS_TREE_BLOCK_REF_KEY);
5503                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5504         }
5505
5506         btrfs_mark_buffer_dirty(leaf);
5507         btrfs_free_path(path);
5508
5509         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5510         if (ret) {
5511                 printk(KERN_ERR "btrfs update block group failed for %llu "
5512                        "%llu\n", (unsigned long long)ins->objectid,
5513                        (unsigned long long)ins->offset);
5514                 BUG();
5515         }
5516         return ret;
5517 }
5518
5519 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5520                                      struct btrfs_root *root,
5521                                      u64 root_objectid, u64 owner,
5522                                      u64 offset, struct btrfs_key *ins)
5523 {
5524         int ret;
5525
5526         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5527
5528         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5529                                          0, root_objectid, owner, offset,
5530                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5531         return ret;
5532 }
5533
5534 /*
5535  * this is used by the tree logging recovery code.  It records that
5536  * an extent has been allocated and makes sure to clear the free
5537  * space cache bits as well
5538  */
5539 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5540                                    struct btrfs_root *root,
5541                                    u64 root_objectid, u64 owner, u64 offset,
5542                                    struct btrfs_key *ins)
5543 {
5544         int ret;
5545         struct btrfs_block_group_cache *block_group;
5546         struct btrfs_caching_control *caching_ctl;
5547         u64 start = ins->objectid;
5548         u64 num_bytes = ins->offset;
5549
5550         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5551         cache_block_group(block_group, trans, NULL, 0);
5552         caching_ctl = get_caching_control(block_group);
5553
5554         if (!caching_ctl) {
5555                 BUG_ON(!block_group_cache_done(block_group));
5556                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5557                 BUG_ON(ret);
5558         } else {
5559                 mutex_lock(&caching_ctl->mutex);
5560
5561                 if (start >= caching_ctl->progress) {
5562                         ret = add_excluded_extent(root, start, num_bytes);
5563                         BUG_ON(ret);
5564                 } else if (start + num_bytes <= caching_ctl->progress) {
5565                         ret = btrfs_remove_free_space(block_group,
5566                                                       start, num_bytes);
5567                         BUG_ON(ret);
5568                 } else {
5569                         num_bytes = caching_ctl->progress - start;
5570                         ret = btrfs_remove_free_space(block_group,
5571                                                       start, num_bytes);
5572                         BUG_ON(ret);
5573
5574                         start = caching_ctl->progress;
5575                         num_bytes = ins->objectid + ins->offset -
5576                                     caching_ctl->progress;
5577                         ret = add_excluded_extent(root, start, num_bytes);
5578                         BUG_ON(ret);
5579                 }
5580
5581                 mutex_unlock(&caching_ctl->mutex);
5582                 put_caching_control(caching_ctl);
5583         }
5584
5585         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5586         BUG_ON(ret);
5587         btrfs_put_block_group(block_group);
5588         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5589                                          0, owner, offset, ins, 1);
5590         return ret;
5591 }
5592
5593 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5594                                             struct btrfs_root *root,
5595                                             u64 bytenr, u32 blocksize,
5596                                             int level)
5597 {
5598         struct extent_buffer *buf;
5599
5600         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5601         if (!buf)
5602                 return ERR_PTR(-ENOMEM);
5603         btrfs_set_header_generation(buf, trans->transid);
5604         btrfs_set_buffer_lockdep_class(buf, level);
5605         btrfs_tree_lock(buf);
5606         clean_tree_block(trans, root, buf);
5607
5608         btrfs_set_lock_blocking(buf);
5609         btrfs_set_buffer_uptodate(buf);
5610
5611         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5612                 /*
5613                  * we allow two log transactions at a time, use different
5614                  * EXENT bit to differentiate dirty pages.
5615                  */
5616                 if (root->log_transid % 2 == 0)
5617                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5618                                         buf->start + buf->len - 1, GFP_NOFS);
5619                 else
5620                         set_extent_new(&root->dirty_log_pages, buf->start,
5621                                         buf->start + buf->len - 1, GFP_NOFS);
5622         } else {
5623                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5624                          buf->start + buf->len - 1, GFP_NOFS);
5625         }
5626         trans->blocks_used++;
5627         /* this returns a buffer locked for blocking */
5628         return buf;
5629 }
5630
5631 static struct btrfs_block_rsv *
5632 use_block_rsv(struct btrfs_trans_handle *trans,
5633               struct btrfs_root *root, u32 blocksize)
5634 {
5635         struct btrfs_block_rsv *block_rsv;
5636         int ret;
5637
5638         block_rsv = get_block_rsv(trans, root);
5639
5640         if (block_rsv->size == 0) {
5641                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5642                                              blocksize, 0);
5643                 if (ret)
5644                         return ERR_PTR(ret);
5645                 return block_rsv;
5646         }
5647
5648         ret = block_rsv_use_bytes(block_rsv, blocksize);
5649         if (!ret)
5650                 return block_rsv;
5651
5652         return ERR_PTR(-ENOSPC);
5653 }
5654
5655 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5656 {
5657         block_rsv_add_bytes(block_rsv, blocksize, 0);
5658         block_rsv_release_bytes(block_rsv, NULL, 0);
5659 }
5660
5661 /*
5662  * finds a free extent and does all the dirty work required for allocation
5663  * returns the key for the extent through ins, and a tree buffer for
5664  * the first block of the extent through buf.
5665  *
5666  * returns the tree buffer or NULL.
5667  */
5668 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5669                                         struct btrfs_root *root, u32 blocksize,
5670                                         u64 parent, u64 root_objectid,
5671                                         struct btrfs_disk_key *key, int level,
5672                                         u64 hint, u64 empty_size)
5673 {
5674         struct btrfs_key ins;
5675         struct btrfs_block_rsv *block_rsv;
5676         struct extent_buffer *buf;
5677         u64 flags = 0;
5678         int ret;
5679
5680
5681         block_rsv = use_block_rsv(trans, root, blocksize);
5682         if (IS_ERR(block_rsv))
5683                 return ERR_CAST(block_rsv);
5684
5685         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5686                                    empty_size, hint, (u64)-1, &ins, 0);
5687         if (ret) {
5688                 unuse_block_rsv(block_rsv, blocksize);
5689                 return ERR_PTR(ret);
5690         }
5691
5692         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5693                                     blocksize, level);
5694         BUG_ON(IS_ERR(buf));
5695
5696         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5697                 if (parent == 0)
5698                         parent = ins.objectid;
5699                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5700         } else
5701                 BUG_ON(parent > 0);
5702
5703         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5704                 struct btrfs_delayed_extent_op *extent_op;
5705                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5706                 BUG_ON(!extent_op);
5707                 if (key)
5708                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5709                 else
5710                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5711                 extent_op->flags_to_set = flags;
5712                 extent_op->update_key = 1;
5713                 extent_op->update_flags = 1;
5714                 extent_op->is_data = 0;
5715
5716                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5717                                         ins.offset, parent, root_objectid,
5718                                         level, BTRFS_ADD_DELAYED_EXTENT,
5719                                         extent_op);
5720                 BUG_ON(ret);
5721         }
5722         return buf;
5723 }
5724
5725 struct walk_control {
5726         u64 refs[BTRFS_MAX_LEVEL];
5727         u64 flags[BTRFS_MAX_LEVEL];
5728         struct btrfs_key update_progress;
5729         int stage;
5730         int level;
5731         int shared_level;
5732         int update_ref;
5733         int keep_locks;
5734         int reada_slot;
5735         int reada_count;
5736 };
5737
5738 #define DROP_REFERENCE  1
5739 #define UPDATE_BACKREF  2
5740
5741 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5742                                      struct btrfs_root *root,
5743                                      struct walk_control *wc,
5744                                      struct btrfs_path *path)
5745 {
5746         u64 bytenr;
5747         u64 generation;
5748         u64 refs;
5749         u64 flags;
5750         u32 nritems;
5751         u32 blocksize;
5752         struct btrfs_key key;
5753         struct extent_buffer *eb;
5754         int ret;
5755         int slot;
5756         int nread = 0;
5757
5758         if (path->slots[wc->level] < wc->reada_slot) {
5759                 wc->reada_count = wc->reada_count * 2 / 3;
5760                 wc->reada_count = max(wc->reada_count, 2);
5761         } else {
5762                 wc->reada_count = wc->reada_count * 3 / 2;
5763                 wc->reada_count = min_t(int, wc->reada_count,
5764                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5765         }
5766
5767         eb = path->nodes[wc->level];
5768         nritems = btrfs_header_nritems(eb);
5769         blocksize = btrfs_level_size(root, wc->level - 1);
5770
5771         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5772                 if (nread >= wc->reada_count)
5773                         break;
5774
5775                 cond_resched();
5776                 bytenr = btrfs_node_blockptr(eb, slot);
5777                 generation = btrfs_node_ptr_generation(eb, slot);
5778
5779                 if (slot == path->slots[wc->level])
5780                         goto reada;
5781
5782                 if (wc->stage == UPDATE_BACKREF &&
5783                     generation <= root->root_key.offset)
5784                         continue;
5785
5786                 /* We don't lock the tree block, it's OK to be racy here */
5787                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5788                                                &refs, &flags);
5789                 BUG_ON(ret);
5790                 BUG_ON(refs == 0);
5791
5792                 if (wc->stage == DROP_REFERENCE) {
5793                         if (refs == 1)
5794                                 goto reada;
5795
5796                         if (wc->level == 1 &&
5797                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5798                                 continue;
5799                         if (!wc->update_ref ||
5800                             generation <= root->root_key.offset)
5801                                 continue;
5802                         btrfs_node_key_to_cpu(eb, &key, slot);
5803                         ret = btrfs_comp_cpu_keys(&key,
5804                                                   &wc->update_progress);
5805                         if (ret < 0)
5806                                 continue;
5807                 } else {
5808                         if (wc->level == 1 &&
5809                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5810                                 continue;
5811                 }
5812 reada:
5813                 ret = readahead_tree_block(root, bytenr, blocksize,
5814                                            generation);
5815                 if (ret)
5816                         break;
5817                 nread++;
5818         }
5819         wc->reada_slot = slot;
5820 }
5821
5822 /*
5823  * hepler to process tree block while walking down the tree.
5824  *
5825  * when wc->stage == UPDATE_BACKREF, this function updates
5826  * back refs for pointers in the block.
5827  *
5828  * NOTE: return value 1 means we should stop walking down.
5829  */
5830 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5831                                    struct btrfs_root *root,
5832                                    struct btrfs_path *path,
5833                                    struct walk_control *wc, int lookup_info)
5834 {
5835         int level = wc->level;
5836         struct extent_buffer *eb = path->nodes[level];
5837         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5838         int ret;
5839
5840         if (wc->stage == UPDATE_BACKREF &&
5841             btrfs_header_owner(eb) != root->root_key.objectid)
5842                 return 1;
5843
5844         /*
5845          * when reference count of tree block is 1, it won't increase
5846          * again. once full backref flag is set, we never clear it.
5847          */
5848         if (lookup_info &&
5849             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5850              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5851                 BUG_ON(!path->locks[level]);
5852                 ret = btrfs_lookup_extent_info(trans, root,
5853                                                eb->start, eb->len,
5854                                                &wc->refs[level],
5855                                                &wc->flags[level]);
5856                 BUG_ON(ret);
5857                 BUG_ON(wc->refs[level] == 0);
5858         }
5859
5860         if (wc->stage == DROP_REFERENCE) {
5861                 if (wc->refs[level] > 1)
5862                         return 1;
5863
5864                 if (path->locks[level] && !wc->keep_locks) {
5865                         btrfs_tree_unlock(eb);
5866                         path->locks[level] = 0;
5867                 }
5868                 return 0;
5869         }
5870
5871         /* wc->stage == UPDATE_BACKREF */
5872         if (!(wc->flags[level] & flag)) {
5873                 BUG_ON(!path->locks[level]);
5874                 ret = btrfs_inc_ref(trans, root, eb, 1);
5875                 BUG_ON(ret);
5876                 ret = btrfs_dec_ref(trans, root, eb, 0);
5877                 BUG_ON(ret);
5878                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5879                                                   eb->len, flag, 0);
5880                 BUG_ON(ret);
5881                 wc->flags[level] |= flag;
5882         }
5883
5884         /*
5885          * the block is shared by multiple trees, so it's not good to
5886          * keep the tree lock
5887          */
5888         if (path->locks[level] && level > 0) {
5889                 btrfs_tree_unlock(eb);
5890                 path->locks[level] = 0;
5891         }
5892         return 0;
5893 }
5894
5895 /*
5896  * hepler to process tree block pointer.
5897  *
5898  * when wc->stage == DROP_REFERENCE, this function checks
5899  * reference count of the block pointed to. if the block
5900  * is shared and we need update back refs for the subtree
5901  * rooted at the block, this function changes wc->stage to
5902  * UPDATE_BACKREF. if the block is shared and there is no
5903  * need to update back, this function drops the reference
5904  * to the block.
5905  *
5906  * NOTE: return value 1 means we should stop walking down.
5907  */
5908 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5909                                  struct btrfs_root *root,
5910                                  struct btrfs_path *path,
5911                                  struct walk_control *wc, int *lookup_info)
5912 {
5913         u64 bytenr;
5914         u64 generation;
5915         u64 parent;
5916         u32 blocksize;
5917         struct btrfs_key key;
5918         struct extent_buffer *next;
5919         int level = wc->level;
5920         int reada = 0;
5921         int ret = 0;
5922
5923         generation = btrfs_node_ptr_generation(path->nodes[level],
5924                                                path->slots[level]);
5925         /*
5926          * if the lower level block was created before the snapshot
5927          * was created, we know there is no need to update back refs
5928          * for the subtree
5929          */
5930         if (wc->stage == UPDATE_BACKREF &&
5931             generation <= root->root_key.offset) {
5932                 *lookup_info = 1;
5933                 return 1;
5934         }
5935
5936         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5937         blocksize = btrfs_level_size(root, level - 1);
5938
5939         next = btrfs_find_tree_block(root, bytenr, blocksize);
5940         if (!next) {
5941                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5942                 if (!next)
5943                         return -ENOMEM;
5944                 reada = 1;
5945         }
5946         btrfs_tree_lock(next);
5947         btrfs_set_lock_blocking(next);
5948
5949         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5950                                        &wc->refs[level - 1],
5951                                        &wc->flags[level - 1]);
5952         BUG_ON(ret);
5953         BUG_ON(wc->refs[level - 1] == 0);
5954         *lookup_info = 0;
5955
5956         if (wc->stage == DROP_REFERENCE) {
5957                 if (wc->refs[level - 1] > 1) {
5958                         if (level == 1 &&
5959                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5960                                 goto skip;
5961
5962                         if (!wc->update_ref ||
5963                             generation <= root->root_key.offset)
5964                                 goto skip;
5965
5966                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5967                                               path->slots[level]);
5968                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5969                         if (ret < 0)
5970                                 goto skip;
5971
5972                         wc->stage = UPDATE_BACKREF;
5973                         wc->shared_level = level - 1;
5974                 }
5975         } else {
5976                 if (level == 1 &&
5977                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5978                         goto skip;
5979         }
5980
5981         if (!btrfs_buffer_uptodate(next, generation)) {
5982                 btrfs_tree_unlock(next);
5983                 free_extent_buffer(next);
5984                 next = NULL;
5985                 *lookup_info = 1;
5986         }
5987
5988         if (!next) {
5989                 if (reada && level == 1)
5990                         reada_walk_down(trans, root, wc, path);
5991                 next = read_tree_block(root, bytenr, blocksize, generation);
5992                 btrfs_tree_lock(next);
5993                 btrfs_set_lock_blocking(next);
5994         }
5995
5996         level--;
5997         BUG_ON(level != btrfs_header_level(next));
5998         path->nodes[level] = next;
5999         path->slots[level] = 0;
6000         path->locks[level] = 1;
6001         wc->level = level;
6002         if (wc->level == 1)
6003                 wc->reada_slot = 0;
6004         return 0;
6005 skip:
6006         wc->refs[level - 1] = 0;
6007         wc->flags[level - 1] = 0;
6008         if (wc->stage == DROP_REFERENCE) {
6009                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6010                         parent = path->nodes[level]->start;
6011                 } else {
6012                         BUG_ON(root->root_key.objectid !=
6013                                btrfs_header_owner(path->nodes[level]));
6014                         parent = 0;
6015                 }
6016
6017                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6018                                         root->root_key.objectid, level - 1, 0);
6019                 BUG_ON(ret);
6020         }
6021         btrfs_tree_unlock(next);
6022         free_extent_buffer(next);
6023         *lookup_info = 1;
6024         return 1;
6025 }
6026
6027 /*
6028  * hepler to process tree block while walking up the tree.
6029  *
6030  * when wc->stage == DROP_REFERENCE, this function drops
6031  * reference count on the block.
6032  *
6033  * when wc->stage == UPDATE_BACKREF, this function changes
6034  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6035  * to UPDATE_BACKREF previously while processing the block.
6036  *
6037  * NOTE: return value 1 means we should stop walking up.
6038  */
6039 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6040                                  struct btrfs_root *root,
6041                                  struct btrfs_path *path,
6042                                  struct walk_control *wc)
6043 {
6044         int ret;
6045         int level = wc->level;
6046         struct extent_buffer *eb = path->nodes[level];
6047         u64 parent = 0;
6048
6049         if (wc->stage == UPDATE_BACKREF) {
6050                 BUG_ON(wc->shared_level < level);
6051                 if (level < wc->shared_level)
6052                         goto out;
6053
6054                 ret = find_next_key(path, level + 1, &wc->update_progress);
6055                 if (ret > 0)
6056                         wc->update_ref = 0;
6057
6058                 wc->stage = DROP_REFERENCE;
6059                 wc->shared_level = -1;
6060                 path->slots[level] = 0;
6061
6062                 /*
6063                  * check reference count again if the block isn't locked.
6064                  * we should start walking down the tree again if reference
6065                  * count is one.
6066                  */
6067                 if (!path->locks[level]) {
6068                         BUG_ON(level == 0);
6069                         btrfs_tree_lock(eb);
6070                         btrfs_set_lock_blocking(eb);
6071                         path->locks[level] = 1;
6072
6073                         ret = btrfs_lookup_extent_info(trans, root,
6074                                                        eb->start, eb->len,
6075                                                        &wc->refs[level],
6076                                                        &wc->flags[level]);
6077                         BUG_ON(ret);
6078                         BUG_ON(wc->refs[level] == 0);
6079                         if (wc->refs[level] == 1) {
6080                                 btrfs_tree_unlock(eb);
6081                                 path->locks[level] = 0;
6082                                 return 1;
6083                         }
6084                 }
6085         }
6086
6087         /* wc->stage == DROP_REFERENCE */
6088         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6089
6090         if (wc->refs[level] == 1) {
6091                 if (level == 0) {
6092                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6093                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6094                         else
6095                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6096                         BUG_ON(ret);
6097                 }
6098                 /* make block locked assertion in clean_tree_block happy */
6099                 if (!path->locks[level] &&
6100                     btrfs_header_generation(eb) == trans->transid) {
6101                         btrfs_tree_lock(eb);
6102                         btrfs_set_lock_blocking(eb);
6103                         path->locks[level] = 1;
6104                 }
6105                 clean_tree_block(trans, root, eb);
6106         }
6107
6108         if (eb == root->node) {
6109                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6110                         parent = eb->start;
6111                 else
6112                         BUG_ON(root->root_key.objectid !=
6113                                btrfs_header_owner(eb));
6114         } else {
6115                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6116                         parent = path->nodes[level + 1]->start;
6117                 else
6118                         BUG_ON(root->root_key.objectid !=
6119                                btrfs_header_owner(path->nodes[level + 1]));
6120         }
6121
6122         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6123 out:
6124         wc->refs[level] = 0;
6125         wc->flags[level] = 0;
6126         return 0;
6127 }
6128
6129 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6130                                    struct btrfs_root *root,
6131                                    struct btrfs_path *path,
6132                                    struct walk_control *wc)
6133 {
6134         int level = wc->level;
6135         int lookup_info = 1;
6136         int ret;
6137
6138         while (level >= 0) {
6139                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6140                 if (ret > 0)
6141                         break;
6142
6143                 if (level == 0)
6144                         break;
6145
6146                 if (path->slots[level] >=
6147                     btrfs_header_nritems(path->nodes[level]))
6148                         break;
6149
6150                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6151                 if (ret > 0) {
6152                         path->slots[level]++;
6153                         continue;
6154                 } else if (ret < 0)
6155                         return ret;
6156                 level = wc->level;
6157         }
6158         return 0;
6159 }
6160
6161 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6162                                  struct btrfs_root *root,
6163                                  struct btrfs_path *path,
6164                                  struct walk_control *wc, int max_level)
6165 {
6166         int level = wc->level;
6167         int ret;
6168
6169         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6170         while (level < max_level && path->nodes[level]) {
6171                 wc->level = level;
6172                 if (path->slots[level] + 1 <
6173                     btrfs_header_nritems(path->nodes[level])) {
6174                         path->slots[level]++;
6175                         return 0;
6176                 } else {
6177                         ret = walk_up_proc(trans, root, path, wc);
6178                         if (ret > 0)
6179                                 return 0;
6180
6181                         if (path->locks[level]) {
6182                                 btrfs_tree_unlock(path->nodes[level]);
6183                                 path->locks[level] = 0;
6184                         }
6185                         free_extent_buffer(path->nodes[level]);
6186                         path->nodes[level] = NULL;
6187                         level++;
6188                 }
6189         }
6190         return 1;
6191 }
6192
6193 /*
6194  * drop a subvolume tree.
6195  *
6196  * this function traverses the tree freeing any blocks that only
6197  * referenced by the tree.
6198  *
6199  * when a shared tree block is found. this function decreases its
6200  * reference count by one. if update_ref is true, this function
6201  * also make sure backrefs for the shared block and all lower level
6202  * blocks are properly updated.
6203  */
6204 int btrfs_drop_snapshot(struct btrfs_root *root,
6205                         struct btrfs_block_rsv *block_rsv, int update_ref)
6206 {
6207         struct btrfs_path *path;
6208         struct btrfs_trans_handle *trans;
6209         struct btrfs_root *tree_root = root->fs_info->tree_root;
6210         struct btrfs_root_item *root_item = &root->root_item;
6211         struct walk_control *wc;
6212         struct btrfs_key key;
6213         int err = 0;
6214         int ret;
6215         int level;
6216
6217         path = btrfs_alloc_path();
6218         BUG_ON(!path);
6219
6220         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6221         BUG_ON(!wc);
6222
6223         trans = btrfs_start_transaction(tree_root, 0);
6224         if (block_rsv)
6225                 trans->block_rsv = block_rsv;
6226
6227         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6228                 level = btrfs_header_level(root->node);
6229                 path->nodes[level] = btrfs_lock_root_node(root);
6230                 btrfs_set_lock_blocking(path->nodes[level]);
6231                 path->slots[level] = 0;
6232                 path->locks[level] = 1;
6233                 memset(&wc->update_progress, 0,
6234                        sizeof(wc->update_progress));
6235         } else {
6236                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6237                 memcpy(&wc->update_progress, &key,
6238                        sizeof(wc->update_progress));
6239
6240                 level = root_item->drop_level;
6241                 BUG_ON(level == 0);
6242                 path->lowest_level = level;
6243                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6244                 path->lowest_level = 0;
6245                 if (ret < 0) {
6246                         err = ret;
6247                         goto out;
6248                 }
6249                 WARN_ON(ret > 0);
6250
6251                 /*
6252                  * unlock our path, this is safe because only this
6253                  * function is allowed to delete this snapshot
6254                  */
6255                 btrfs_unlock_up_safe(path, 0);
6256
6257                 level = btrfs_header_level(root->node);
6258                 while (1) {
6259                         btrfs_tree_lock(path->nodes[level]);
6260                         btrfs_set_lock_blocking(path->nodes[level]);
6261
6262                         ret = btrfs_lookup_extent_info(trans, root,
6263                                                 path->nodes[level]->start,
6264                                                 path->nodes[level]->len,
6265                                                 &wc->refs[level],
6266                                                 &wc->flags[level]);
6267                         BUG_ON(ret);
6268                         BUG_ON(wc->refs[level] == 0);
6269
6270                         if (level == root_item->drop_level)
6271                                 break;
6272
6273                         btrfs_tree_unlock(path->nodes[level]);
6274                         WARN_ON(wc->refs[level] != 1);
6275                         level--;
6276                 }
6277         }
6278
6279         wc->level = level;
6280         wc->shared_level = -1;
6281         wc->stage = DROP_REFERENCE;
6282         wc->update_ref = update_ref;
6283         wc->keep_locks = 0;
6284         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6285
6286         while (1) {
6287                 ret = walk_down_tree(trans, root, path, wc);
6288                 if (ret < 0) {
6289                         err = ret;
6290                         break;
6291                 }
6292
6293                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6294                 if (ret < 0) {
6295                         err = ret;
6296                         break;
6297                 }
6298
6299                 if (ret > 0) {
6300                         BUG_ON(wc->stage != DROP_REFERENCE);
6301                         break;
6302                 }
6303
6304                 if (wc->stage == DROP_REFERENCE) {
6305                         level = wc->level;
6306                         btrfs_node_key(path->nodes[level],
6307                                        &root_item->drop_progress,
6308                                        path->slots[level]);
6309                         root_item->drop_level = level;
6310                 }
6311
6312                 BUG_ON(wc->level == 0);
6313                 if (btrfs_should_end_transaction(trans, tree_root)) {
6314                         ret = btrfs_update_root(trans, tree_root,
6315                                                 &root->root_key,
6316                                                 root_item);
6317                         BUG_ON(ret);
6318
6319                         btrfs_end_transaction_throttle(trans, tree_root);
6320                         trans = btrfs_start_transaction(tree_root, 0);
6321                         if (block_rsv)
6322                                 trans->block_rsv = block_rsv;
6323                 }
6324         }
6325         btrfs_release_path(root, path);
6326         BUG_ON(err);
6327
6328         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6329         BUG_ON(ret);
6330
6331         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6332                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6333                                            NULL, NULL);
6334                 BUG_ON(ret < 0);
6335                 if (ret > 0) {
6336                         /* if we fail to delete the orphan item this time
6337                          * around, it'll get picked up the next time.
6338                          *
6339                          * The most common failure here is just -ENOENT.
6340                          */
6341                         btrfs_del_orphan_item(trans, tree_root,
6342                                               root->root_key.objectid);
6343                 }
6344         }
6345
6346         if (root->in_radix) {
6347                 btrfs_free_fs_root(tree_root->fs_info, root);
6348         } else {
6349                 free_extent_buffer(root->node);
6350                 free_extent_buffer(root->commit_root);
6351                 kfree(root);
6352         }
6353 out:
6354         btrfs_end_transaction_throttle(trans, tree_root);
6355         kfree(wc);
6356         btrfs_free_path(path);
6357         return err;
6358 }
6359
6360 /*
6361  * drop subtree rooted at tree block 'node'.
6362  *
6363  * NOTE: this function will unlock and release tree block 'node'
6364  */
6365 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6366                         struct btrfs_root *root,
6367                         struct extent_buffer *node,
6368                         struct extent_buffer *parent)
6369 {
6370         struct btrfs_path *path;
6371         struct walk_control *wc;
6372         int level;
6373         int parent_level;
6374         int ret = 0;
6375         int wret;
6376
6377         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6378
6379         path = btrfs_alloc_path();
6380         BUG_ON(!path);
6381
6382         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6383         BUG_ON(!wc);
6384
6385         btrfs_assert_tree_locked(parent);
6386         parent_level = btrfs_header_level(parent);
6387         extent_buffer_get(parent);
6388         path->nodes[parent_level] = parent;
6389         path->slots[parent_level] = btrfs_header_nritems(parent);
6390
6391         btrfs_assert_tree_locked(node);
6392         level = btrfs_header_level(node);
6393         path->nodes[level] = node;
6394         path->slots[level] = 0;
6395         path->locks[level] = 1;
6396
6397         wc->refs[parent_level] = 1;
6398         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6399         wc->level = level;
6400         wc->shared_level = -1;
6401         wc->stage = DROP_REFERENCE;
6402         wc->update_ref = 0;
6403         wc->keep_locks = 1;
6404         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6405
6406         while (1) {
6407                 wret = walk_down_tree(trans, root, path, wc);
6408                 if (wret < 0) {
6409                         ret = wret;
6410                         break;
6411                 }
6412
6413                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6414                 if (wret < 0)
6415                         ret = wret;
6416                 if (wret != 0)
6417                         break;
6418         }
6419
6420         kfree(wc);
6421         btrfs_free_path(path);
6422         return ret;
6423 }
6424
6425 #if 0
6426 static unsigned long calc_ra(unsigned long start, unsigned long last,
6427                              unsigned long nr)
6428 {
6429         return min(last, start + nr - 1);
6430 }
6431
6432 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6433                                          u64 len)
6434 {
6435         u64 page_start;
6436         u64 page_end;
6437         unsigned long first_index;
6438         unsigned long last_index;
6439         unsigned long i;
6440         struct page *page;
6441         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6442         struct file_ra_state *ra;
6443         struct btrfs_ordered_extent *ordered;
6444         unsigned int total_read = 0;
6445         unsigned int total_dirty = 0;
6446         int ret = 0;
6447
6448         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6449
6450         mutex_lock(&inode->i_mutex);
6451         first_index = start >> PAGE_CACHE_SHIFT;
6452         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6453
6454         /* make sure the dirty trick played by the caller work */
6455         ret = invalidate_inode_pages2_range(inode->i_mapping,
6456                                             first_index, last_index);
6457         if (ret)
6458                 goto out_unlock;
6459
6460         file_ra_state_init(ra, inode->i_mapping);
6461
6462         for (i = first_index ; i <= last_index; i++) {
6463                 if (total_read % ra->ra_pages == 0) {
6464                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6465                                        calc_ra(i, last_index, ra->ra_pages));
6466                 }
6467                 total_read++;
6468 again:
6469                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6470                         BUG_ON(1);
6471                 page = grab_cache_page(inode->i_mapping, i);
6472                 if (!page) {
6473                         ret = -ENOMEM;
6474                         goto out_unlock;
6475                 }
6476                 if (!PageUptodate(page)) {
6477                         btrfs_readpage(NULL, page);
6478                         lock_page(page);
6479                         if (!PageUptodate(page)) {
6480                                 unlock_page(page);
6481                                 page_cache_release(page);
6482                                 ret = -EIO;
6483                                 goto out_unlock;
6484                         }
6485                 }
6486                 wait_on_page_writeback(page);
6487
6488                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6489                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6490                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6491
6492                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6493                 if (ordered) {
6494                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6495                         unlock_page(page);
6496                         page_cache_release(page);
6497                         btrfs_start_ordered_extent(inode, ordered, 1);
6498                         btrfs_put_ordered_extent(ordered);
6499                         goto again;
6500                 }
6501                 set_page_extent_mapped(page);
6502
6503                 if (i == first_index)
6504                         set_extent_bits(io_tree, page_start, page_end,
6505                                         EXTENT_BOUNDARY, GFP_NOFS);
6506                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6507
6508                 set_page_dirty(page);
6509                 total_dirty++;
6510
6511                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6512                 unlock_page(page);
6513                 page_cache_release(page);
6514         }
6515
6516 out_unlock:
6517         kfree(ra);
6518         mutex_unlock(&inode->i_mutex);
6519         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6520         return ret;
6521 }
6522
6523 static noinline int relocate_data_extent(struct inode *reloc_inode,
6524                                          struct btrfs_key *extent_key,
6525                                          u64 offset)
6526 {
6527         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6528         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6529         struct extent_map *em;
6530         u64 start = extent_key->objectid - offset;
6531         u64 end = start + extent_key->offset - 1;
6532
6533         em = alloc_extent_map(GFP_NOFS);
6534         BUG_ON(!em || IS_ERR(em));
6535
6536         em->start = start;
6537         em->len = extent_key->offset;
6538         em->block_len = extent_key->offset;
6539         em->block_start = extent_key->objectid;
6540         em->bdev = root->fs_info->fs_devices->latest_bdev;
6541         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6542
6543         /* setup extent map to cheat btrfs_readpage */
6544         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6545         while (1) {
6546                 int ret;
6547                 write_lock(&em_tree->lock);
6548                 ret = add_extent_mapping(em_tree, em);
6549                 write_unlock(&em_tree->lock);
6550                 if (ret != -EEXIST) {
6551                         free_extent_map(em);
6552                         break;
6553                 }
6554                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6555         }
6556         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6557
6558         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6559 }
6560
6561 struct btrfs_ref_path {
6562         u64 extent_start;
6563         u64 nodes[BTRFS_MAX_LEVEL];
6564         u64 root_objectid;
6565         u64 root_generation;
6566         u64 owner_objectid;
6567         u32 num_refs;
6568         int lowest_level;
6569         int current_level;
6570         int shared_level;
6571
6572         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6573         u64 new_nodes[BTRFS_MAX_LEVEL];
6574 };
6575
6576 struct disk_extent {
6577         u64 ram_bytes;
6578         u64 disk_bytenr;
6579         u64 disk_num_bytes;
6580         u64 offset;
6581         u64 num_bytes;
6582         u8 compression;
6583         u8 encryption;
6584         u16 other_encoding;
6585 };
6586
6587 static int is_cowonly_root(u64 root_objectid)
6588 {
6589         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6590             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6591             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6592             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6593             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6594             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6595                 return 1;
6596         return 0;
6597 }
6598
6599 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6600                                     struct btrfs_root *extent_root,
6601                                     struct btrfs_ref_path *ref_path,
6602                                     int first_time)
6603 {
6604         struct extent_buffer *leaf;
6605         struct btrfs_path *path;
6606         struct btrfs_extent_ref *ref;
6607         struct btrfs_key key;
6608         struct btrfs_key found_key;
6609         u64 bytenr;
6610         u32 nritems;
6611         int level;
6612         int ret = 1;
6613
6614         path = btrfs_alloc_path();
6615         if (!path)
6616                 return -ENOMEM;
6617
6618         if (first_time) {
6619                 ref_path->lowest_level = -1;
6620                 ref_path->current_level = -1;
6621                 ref_path->shared_level = -1;
6622                 goto walk_up;
6623         }
6624 walk_down:
6625         level = ref_path->current_level - 1;
6626         while (level >= -1) {
6627                 u64 parent;
6628                 if (level < ref_path->lowest_level)
6629                         break;
6630
6631                 if (level >= 0)
6632                         bytenr = ref_path->nodes[level];
6633                 else
6634                         bytenr = ref_path->extent_start;
6635                 BUG_ON(bytenr == 0);
6636
6637                 parent = ref_path->nodes[level + 1];
6638                 ref_path->nodes[level + 1] = 0;
6639                 ref_path->current_level = level;
6640                 BUG_ON(parent == 0);
6641
6642                 key.objectid = bytenr;
6643                 key.offset = parent + 1;
6644                 key.type = BTRFS_EXTENT_REF_KEY;
6645
6646                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6647                 if (ret < 0)
6648                         goto out;
6649                 BUG_ON(ret == 0);
6650
6651                 leaf = path->nodes[0];
6652                 nritems = btrfs_header_nritems(leaf);
6653                 if (path->slots[0] >= nritems) {
6654                         ret = btrfs_next_leaf(extent_root, path);
6655                         if (ret < 0)
6656                                 goto out;
6657                         if (ret > 0)
6658                                 goto next;
6659                         leaf = path->nodes[0];
6660                 }
6661
6662                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6663                 if (found_key.objectid == bytenr &&
6664                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6665                         if (level < ref_path->shared_level)
6666                                 ref_path->shared_level = level;
6667                         goto found;
6668                 }
6669 next:
6670                 level--;
6671                 btrfs_release_path(extent_root, path);
6672                 cond_resched();
6673         }
6674         /* reached lowest level */
6675         ret = 1;
6676         goto out;
6677 walk_up:
6678         level = ref_path->current_level;
6679         while (level < BTRFS_MAX_LEVEL - 1) {
6680                 u64 ref_objectid;
6681
6682                 if (level >= 0)
6683                         bytenr = ref_path->nodes[level];
6684                 else
6685                         bytenr = ref_path->extent_start;
6686
6687                 BUG_ON(bytenr == 0);
6688
6689                 key.objectid = bytenr;
6690                 key.offset = 0;
6691                 key.type = BTRFS_EXTENT_REF_KEY;
6692
6693                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6694                 if (ret < 0)
6695                         goto out;
6696
6697                 leaf = path->nodes[0];
6698                 nritems = btrfs_header_nritems(leaf);
6699                 if (path->slots[0] >= nritems) {
6700                         ret = btrfs_next_leaf(extent_root, path);
6701                         if (ret < 0)
6702                                 goto out;
6703                         if (ret > 0) {
6704                                 /* the extent was freed by someone */
6705                                 if (ref_path->lowest_level == level)
6706                                         goto out;
6707                                 btrfs_release_path(extent_root, path);
6708                                 goto walk_down;
6709                         }
6710                         leaf = path->nodes[0];
6711                 }
6712
6713                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6714                 if (found_key.objectid != bytenr ||
6715                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6716                         /* the extent was freed by someone */
6717                         if (ref_path->lowest_level == level) {
6718                                 ret = 1;
6719                                 goto out;
6720                         }
6721                         btrfs_release_path(extent_root, path);
6722                         goto walk_down;
6723                 }
6724 found:
6725                 ref = btrfs_item_ptr(leaf, path->slots[0],
6726                                 struct btrfs_extent_ref);
6727                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6728                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6729                         if (first_time) {
6730                                 level = (int)ref_objectid;
6731                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6732                                 ref_path->lowest_level = level;
6733                                 ref_path->current_level = level;
6734                                 ref_path->nodes[level] = bytenr;
6735                         } else {
6736                                 WARN_ON(ref_objectid != level);
6737                         }
6738                 } else {
6739                         WARN_ON(level != -1);
6740                 }
6741                 first_time = 0;
6742
6743                 if (ref_path->lowest_level == level) {
6744                         ref_path->owner_objectid = ref_objectid;
6745                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6746                 }
6747
6748                 /*
6749                  * the block is tree root or the block isn't in reference
6750                  * counted tree.
6751                  */
6752                 if (found_key.objectid == found_key.offset ||
6753                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6754                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6755                         ref_path->root_generation =
6756                                 btrfs_ref_generation(leaf, ref);
6757                         if (level < 0) {
6758                                 /* special reference from the tree log */
6759                                 ref_path->nodes[0] = found_key.offset;
6760                                 ref_path->current_level = 0;
6761                         }
6762                         ret = 0;
6763                         goto out;
6764                 }
6765
6766                 level++;
6767                 BUG_ON(ref_path->nodes[level] != 0);
6768                 ref_path->nodes[level] = found_key.offset;
6769                 ref_path->current_level = level;
6770
6771                 /*
6772                  * the reference was created in the running transaction,
6773                  * no need to continue walking up.
6774                  */
6775                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6776                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6777                         ref_path->root_generation =
6778                                 btrfs_ref_generation(leaf, ref);
6779                         ret = 0;
6780                         goto out;
6781                 }
6782
6783                 btrfs_release_path(extent_root, path);
6784                 cond_resched();
6785         }
6786         /* reached max tree level, but no tree root found. */
6787         BUG();
6788 out:
6789         btrfs_free_path(path);
6790         return ret;
6791 }
6792
6793 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6794                                 struct btrfs_root *extent_root,
6795                                 struct btrfs_ref_path *ref_path,
6796                                 u64 extent_start)
6797 {
6798         memset(ref_path, 0, sizeof(*ref_path));
6799         ref_path->extent_start = extent_start;
6800
6801         return __next_ref_path(trans, extent_root, ref_path, 1);
6802 }
6803
6804 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6805                                struct btrfs_root *extent_root,
6806                                struct btrfs_ref_path *ref_path)
6807 {
6808         return __next_ref_path(trans, extent_root, ref_path, 0);
6809 }
6810
6811 static noinline int get_new_locations(struct inode *reloc_inode,
6812                                       struct btrfs_key *extent_key,
6813                                       u64 offset, int no_fragment,
6814                                       struct disk_extent **extents,
6815                                       int *nr_extents)
6816 {
6817         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6818         struct btrfs_path *path;
6819         struct btrfs_file_extent_item *fi;
6820         struct extent_buffer *leaf;
6821         struct disk_extent *exts = *extents;
6822         struct btrfs_key found_key;
6823         u64 cur_pos;
6824         u64 last_byte;
6825         u32 nritems;
6826         int nr = 0;
6827         int max = *nr_extents;
6828         int ret;
6829
6830         WARN_ON(!no_fragment && *extents);
6831         if (!exts) {
6832                 max = 1;
6833                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6834                 if (!exts)
6835                         return -ENOMEM;
6836         }
6837
6838         path = btrfs_alloc_path();
6839         BUG_ON(!path);
6840
6841         cur_pos = extent_key->objectid - offset;
6842         last_byte = extent_key->objectid + extent_key->offset;
6843         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6844                                        cur_pos, 0);
6845         if (ret < 0)
6846                 goto out;
6847         if (ret > 0) {
6848                 ret = -ENOENT;
6849                 goto out;
6850         }
6851
6852         while (1) {
6853                 leaf = path->nodes[0];
6854                 nritems = btrfs_header_nritems(leaf);
6855                 if (path->slots[0] >= nritems) {
6856                         ret = btrfs_next_leaf(root, path);
6857                         if (ret < 0)
6858                                 goto out;
6859                         if (ret > 0)
6860                                 break;
6861                         leaf = path->nodes[0];
6862                 }
6863
6864                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6865                 if (found_key.offset != cur_pos ||
6866                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6867                     found_key.objectid != reloc_inode->i_ino)
6868                         break;
6869
6870                 fi = btrfs_item_ptr(leaf, path->slots[0],
6871                                     struct btrfs_file_extent_item);
6872                 if (btrfs_file_extent_type(leaf, fi) !=
6873                     BTRFS_FILE_EXTENT_REG ||
6874                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6875                         break;
6876
6877                 if (nr == max) {
6878                         struct disk_extent *old = exts;
6879                         max *= 2;
6880                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6881                         memcpy(exts, old, sizeof(*exts) * nr);
6882                         if (old != *extents)
6883                                 kfree(old);
6884                 }
6885
6886                 exts[nr].disk_bytenr =
6887                         btrfs_file_extent_disk_bytenr(leaf, fi);
6888                 exts[nr].disk_num_bytes =
6889                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6890                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6891                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6892                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6893                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6894                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6895                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6896                                                                            fi);
6897                 BUG_ON(exts[nr].offset > 0);
6898                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6899                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6900
6901                 cur_pos += exts[nr].num_bytes;
6902                 nr++;
6903
6904                 if (cur_pos + offset >= last_byte)
6905                         break;
6906
6907                 if (no_fragment) {
6908                         ret = 1;
6909                         goto out;
6910                 }
6911                 path->slots[0]++;
6912         }
6913
6914         BUG_ON(cur_pos + offset > last_byte);
6915         if (cur_pos + offset < last_byte) {
6916                 ret = -ENOENT;
6917                 goto out;
6918         }
6919         ret = 0;
6920 out:
6921         btrfs_free_path(path);
6922         if (ret) {
6923                 if (exts != *extents)
6924                         kfree(exts);
6925         } else {
6926                 *extents = exts;
6927                 *nr_extents = nr;
6928         }
6929         return ret;
6930 }
6931
6932 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6933                                         struct btrfs_root *root,
6934                                         struct btrfs_path *path,
6935                                         struct btrfs_key *extent_key,
6936                                         struct btrfs_key *leaf_key,
6937                                         struct btrfs_ref_path *ref_path,
6938                                         struct disk_extent *new_extents,
6939                                         int nr_extents)
6940 {
6941         struct extent_buffer *leaf;
6942         struct btrfs_file_extent_item *fi;
6943         struct inode *inode = NULL;
6944         struct btrfs_key key;
6945         u64 lock_start = 0;
6946         u64 lock_end = 0;
6947         u64 num_bytes;
6948         u64 ext_offset;
6949         u64 search_end = (u64)-1;
6950         u32 nritems;
6951         int nr_scaned = 0;
6952         int extent_locked = 0;
6953         int extent_type;
6954         int ret;
6955
6956         memcpy(&key, leaf_key, sizeof(key));
6957         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6958                 if (key.objectid < ref_path->owner_objectid ||
6959                     (key.objectid == ref_path->owner_objectid &&
6960                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6961                         key.objectid = ref_path->owner_objectid;
6962                         key.type = BTRFS_EXTENT_DATA_KEY;
6963                         key.offset = 0;
6964                 }
6965         }
6966
6967         while (1) {
6968                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6969                 if (ret < 0)
6970                         goto out;
6971
6972                 leaf = path->nodes[0];
6973                 nritems = btrfs_header_nritems(leaf);
6974 next:
6975                 if (extent_locked && ret > 0) {
6976                         /*
6977                          * the file extent item was modified by someone
6978                          * before the extent got locked.
6979                          */
6980                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6981                                       lock_end, GFP_NOFS);
6982                         extent_locked = 0;
6983                 }
6984
6985                 if (path->slots[0] >= nritems) {
6986                         if (++nr_scaned > 2)
6987                                 break;
6988
6989                         BUG_ON(extent_locked);
6990                         ret = btrfs_next_leaf(root, path);
6991                         if (ret < 0)
6992                                 goto out;
6993                         if (ret > 0)
6994                                 break;
6995                         leaf = path->nodes[0];
6996                         nritems = btrfs_header_nritems(leaf);
6997                 }
6998
6999                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7000
7001                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7002                         if ((key.objectid > ref_path->owner_objectid) ||
7003                             (key.objectid == ref_path->owner_objectid &&
7004                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7005                             key.offset >= search_end)
7006                                 break;
7007                 }
7008
7009                 if (inode && key.objectid != inode->i_ino) {
7010                         BUG_ON(extent_locked);
7011                         btrfs_release_path(root, path);
7012                         mutex_unlock(&inode->i_mutex);
7013                         iput(inode);
7014                         inode = NULL;
7015                         continue;
7016                 }
7017
7018                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7019                         path->slots[0]++;
7020                         ret = 1;
7021                         goto next;
7022                 }
7023                 fi = btrfs_item_ptr(leaf, path->slots[0],
7024                                     struct btrfs_file_extent_item);
7025                 extent_type = btrfs_file_extent_type(leaf, fi);
7026                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7027                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7028                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7029                      extent_key->objectid)) {
7030                         path->slots[0]++;
7031                         ret = 1;
7032                         goto next;
7033                 }
7034
7035                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7036                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7037
7038                 if (search_end == (u64)-1) {
7039                         search_end = key.offset - ext_offset +
7040                                 btrfs_file_extent_ram_bytes(leaf, fi);
7041                 }
7042
7043                 if (!extent_locked) {
7044                         lock_start = key.offset;
7045                         lock_end = lock_start + num_bytes - 1;
7046                 } else {
7047                         if (lock_start > key.offset ||
7048                             lock_end + 1 < key.offset + num_bytes) {
7049                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7050                                               lock_start, lock_end, GFP_NOFS);
7051                                 extent_locked = 0;
7052                         }
7053                 }
7054
7055                 if (!inode) {
7056                         btrfs_release_path(root, path);
7057
7058                         inode = btrfs_iget_locked(root->fs_info->sb,
7059                                                   key.objectid, root);
7060                         if (inode->i_state & I_NEW) {
7061                                 BTRFS_I(inode)->root = root;
7062                                 BTRFS_I(inode)->location.objectid =
7063                                         key.objectid;
7064                                 BTRFS_I(inode)->location.type =
7065                                         BTRFS_INODE_ITEM_KEY;
7066                                 BTRFS_I(inode)->location.offset = 0;
7067                                 btrfs_read_locked_inode(inode);
7068                                 unlock_new_inode(inode);
7069                         }
7070                         /*
7071                          * some code call btrfs_commit_transaction while
7072                          * holding the i_mutex, so we can't use mutex_lock
7073                          * here.
7074                          */
7075                         if (is_bad_inode(inode) ||
7076                             !mutex_trylock(&inode->i_mutex)) {
7077                                 iput(inode);
7078                                 inode = NULL;
7079                                 key.offset = (u64)-1;
7080                                 goto skip;
7081                         }
7082                 }
7083
7084                 if (!extent_locked) {
7085                         struct btrfs_ordered_extent *ordered;
7086
7087                         btrfs_release_path(root, path);
7088
7089                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7090                                     lock_end, GFP_NOFS);
7091                         ordered = btrfs_lookup_first_ordered_extent(inode,
7092                                                                     lock_end);
7093                         if (ordered &&
7094                             ordered->file_offset <= lock_end &&
7095                             ordered->file_offset + ordered->len > lock_start) {
7096                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7097                                               lock_start, lock_end, GFP_NOFS);
7098                                 btrfs_start_ordered_extent(inode, ordered, 1);
7099                                 btrfs_put_ordered_extent(ordered);
7100                                 key.offset += num_bytes;
7101                                 goto skip;
7102                         }
7103                         if (ordered)
7104                                 btrfs_put_ordered_extent(ordered);
7105
7106                         extent_locked = 1;
7107                         continue;
7108                 }
7109
7110                 if (nr_extents == 1) {
7111                         /* update extent pointer in place */
7112                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7113                                                 new_extents[0].disk_bytenr);
7114                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7115                                                 new_extents[0].disk_num_bytes);
7116                         btrfs_mark_buffer_dirty(leaf);
7117
7118                         btrfs_drop_extent_cache(inode, key.offset,
7119                                                 key.offset + num_bytes - 1, 0);
7120
7121                         ret = btrfs_inc_extent_ref(trans, root,
7122                                                 new_extents[0].disk_bytenr,
7123                                                 new_extents[0].disk_num_bytes,
7124                                                 leaf->start,
7125                                                 root->root_key.objectid,
7126                                                 trans->transid,
7127                                                 key.objectid);
7128                         BUG_ON(ret);
7129
7130                         ret = btrfs_free_extent(trans, root,
7131                                                 extent_key->objectid,
7132                                                 extent_key->offset,
7133                                                 leaf->start,
7134                                                 btrfs_header_owner(leaf),
7135                                                 btrfs_header_generation(leaf),
7136                                                 key.objectid, 0);
7137                         BUG_ON(ret);
7138
7139                         btrfs_release_path(root, path);
7140                         key.offset += num_bytes;
7141                 } else {
7142                         BUG_ON(1);
7143 #if 0
7144                         u64 alloc_hint;
7145                         u64 extent_len;
7146                         int i;
7147                         /*
7148                          * drop old extent pointer at first, then insert the
7149                          * new pointers one bye one
7150                          */
7151                         btrfs_release_path(root, path);
7152                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7153                                                  key.offset + num_bytes,
7154                                                  key.offset, &alloc_hint);
7155                         BUG_ON(ret);
7156
7157                         for (i = 0; i < nr_extents; i++) {
7158                                 if (ext_offset >= new_extents[i].num_bytes) {
7159                                         ext_offset -= new_extents[i].num_bytes;
7160                                         continue;
7161                                 }
7162                                 extent_len = min(new_extents[i].num_bytes -
7163                                                  ext_offset, num_bytes);
7164
7165                                 ret = btrfs_insert_empty_item(trans, root,
7166                                                               path, &key,
7167                                                               sizeof(*fi));
7168                                 BUG_ON(ret);
7169
7170                                 leaf = path->nodes[0];
7171                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7172                                                 struct btrfs_file_extent_item);
7173                                 btrfs_set_file_extent_generation(leaf, fi,
7174                                                         trans->transid);
7175                                 btrfs_set_file_extent_type(leaf, fi,
7176                                                         BTRFS_FILE_EXTENT_REG);
7177                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7178                                                 new_extents[i].disk_bytenr);
7179                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7180                                                 new_extents[i].disk_num_bytes);
7181                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7182                                                 new_extents[i].ram_bytes);
7183
7184                                 btrfs_set_file_extent_compression(leaf, fi,
7185                                                 new_extents[i].compression);
7186                                 btrfs_set_file_extent_encryption(leaf, fi,
7187                                                 new_extents[i].encryption);
7188                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7189                                                 new_extents[i].other_encoding);
7190
7191                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7192                                                         extent_len);
7193                                 ext_offset += new_extents[i].offset;
7194                                 btrfs_set_file_extent_offset(leaf, fi,
7195                                                         ext_offset);
7196                                 btrfs_mark_buffer_dirty(leaf);
7197
7198                                 btrfs_drop_extent_cache(inode, key.offset,
7199                                                 key.offset + extent_len - 1, 0);
7200
7201                                 ret = btrfs_inc_extent_ref(trans, root,
7202                                                 new_extents[i].disk_bytenr,
7203                                                 new_extents[i].disk_num_bytes,
7204                                                 leaf->start,
7205                                                 root->root_key.objectid,
7206                                                 trans->transid, key.objectid);
7207                                 BUG_ON(ret);
7208                                 btrfs_release_path(root, path);
7209
7210                                 inode_add_bytes(inode, extent_len);
7211
7212                                 ext_offset = 0;
7213                                 num_bytes -= extent_len;
7214                                 key.offset += extent_len;
7215
7216                                 if (num_bytes == 0)
7217                                         break;
7218                         }
7219                         BUG_ON(i >= nr_extents);
7220 #endif
7221                 }
7222
7223                 if (extent_locked) {
7224                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7225                                       lock_end, GFP_NOFS);
7226                         extent_locked = 0;
7227                 }
7228 skip:
7229                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7230                     key.offset >= search_end)
7231                         break;
7232
7233                 cond_resched();
7234         }
7235         ret = 0;
7236 out:
7237         btrfs_release_path(root, path);
7238         if (inode) {
7239                 mutex_unlock(&inode->i_mutex);
7240                 if (extent_locked) {
7241                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7242                                       lock_end, GFP_NOFS);
7243                 }
7244                 iput(inode);
7245         }
7246         return ret;
7247 }
7248
7249 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7250                                struct btrfs_root *root,
7251                                struct extent_buffer *buf, u64 orig_start)
7252 {
7253         int level;
7254         int ret;
7255
7256         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7257         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7258
7259         level = btrfs_header_level(buf);
7260         if (level == 0) {
7261                 struct btrfs_leaf_ref *ref;
7262                 struct btrfs_leaf_ref *orig_ref;
7263
7264                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7265                 if (!orig_ref)
7266                         return -ENOENT;
7267
7268                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7269                 if (!ref) {
7270                         btrfs_free_leaf_ref(root, orig_ref);
7271                         return -ENOMEM;
7272                 }
7273
7274                 ref->nritems = orig_ref->nritems;
7275                 memcpy(ref->extents, orig_ref->extents,
7276                         sizeof(ref->extents[0]) * ref->nritems);
7277
7278                 btrfs_free_leaf_ref(root, orig_ref);
7279
7280                 ref->root_gen = trans->transid;
7281                 ref->bytenr = buf->start;
7282                 ref->owner = btrfs_header_owner(buf);
7283                 ref->generation = btrfs_header_generation(buf);
7284
7285                 ret = btrfs_add_leaf_ref(root, ref, 0);
7286                 WARN_ON(ret);
7287                 btrfs_free_leaf_ref(root, ref);
7288         }
7289         return 0;
7290 }
7291
7292 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7293                                         struct extent_buffer *leaf,
7294                                         struct btrfs_block_group_cache *group,
7295                                         struct btrfs_root *target_root)
7296 {
7297         struct btrfs_key key;
7298         struct inode *inode = NULL;
7299         struct btrfs_file_extent_item *fi;
7300         struct extent_state *cached_state = NULL;
7301         u64 num_bytes;
7302         u64 skip_objectid = 0;
7303         u32 nritems;
7304         u32 i;
7305
7306         nritems = btrfs_header_nritems(leaf);
7307         for (i = 0; i < nritems; i++) {
7308                 btrfs_item_key_to_cpu(leaf, &key, i);
7309                 if (key.objectid == skip_objectid ||
7310                     key.type != BTRFS_EXTENT_DATA_KEY)
7311                         continue;
7312                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7313                 if (btrfs_file_extent_type(leaf, fi) ==
7314                     BTRFS_FILE_EXTENT_INLINE)
7315                         continue;
7316                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7317                         continue;
7318                 if (!inode || inode->i_ino != key.objectid) {
7319                         iput(inode);
7320                         inode = btrfs_ilookup(target_root->fs_info->sb,
7321                                               key.objectid, target_root, 1);
7322                 }
7323                 if (!inode) {
7324                         skip_objectid = key.objectid;
7325                         continue;
7326                 }
7327                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7328
7329                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7330                                  key.offset + num_bytes - 1, 0, &cached_state,
7331                                  GFP_NOFS);
7332                 btrfs_drop_extent_cache(inode, key.offset,
7333                                         key.offset + num_bytes - 1, 1);
7334                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7335                                      key.offset + num_bytes - 1, &cached_state,
7336                                      GFP_NOFS);
7337                 cond_resched();
7338         }
7339         iput(inode);
7340         return 0;
7341 }
7342
7343 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7344                                         struct btrfs_root *root,
7345                                         struct extent_buffer *leaf,
7346                                         struct btrfs_block_group_cache *group,
7347                                         struct inode *reloc_inode)
7348 {
7349         struct btrfs_key key;
7350         struct btrfs_key extent_key;
7351         struct btrfs_file_extent_item *fi;
7352         struct btrfs_leaf_ref *ref;
7353         struct disk_extent *new_extent;
7354         u64 bytenr;
7355         u64 num_bytes;
7356         u32 nritems;
7357         u32 i;
7358         int ext_index;
7359         int nr_extent;
7360         int ret;
7361
7362         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7363         BUG_ON(!new_extent);
7364
7365         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7366         BUG_ON(!ref);
7367
7368         ext_index = -1;
7369         nritems = btrfs_header_nritems(leaf);
7370         for (i = 0; i < nritems; i++) {
7371                 btrfs_item_key_to_cpu(leaf, &key, i);
7372                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7373                         continue;
7374                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7375                 if (btrfs_file_extent_type(leaf, fi) ==
7376                     BTRFS_FILE_EXTENT_INLINE)
7377                         continue;
7378                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7379                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7380                 if (bytenr == 0)
7381                         continue;
7382
7383                 ext_index++;
7384                 if (bytenr >= group->key.objectid + group->key.offset ||
7385                     bytenr + num_bytes <= group->key.objectid)
7386                         continue;
7387
7388                 extent_key.objectid = bytenr;
7389                 extent_key.offset = num_bytes;
7390                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7391                 nr_extent = 1;
7392                 ret = get_new_locations(reloc_inode, &extent_key,
7393                                         group->key.objectid, 1,
7394                                         &new_extent, &nr_extent);
7395                 if (ret > 0)
7396                         continue;
7397                 BUG_ON(ret < 0);
7398
7399                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7400                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7401                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7402                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7403
7404                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7405                                                 new_extent->disk_bytenr);
7406                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7407                                                 new_extent->disk_num_bytes);
7408                 btrfs_mark_buffer_dirty(leaf);
7409
7410                 ret = btrfs_inc_extent_ref(trans, root,
7411                                         new_extent->disk_bytenr,
7412                                         new_extent->disk_num_bytes,
7413                                         leaf->start,
7414                                         root->root_key.objectid,
7415                                         trans->transid, key.objectid);
7416                 BUG_ON(ret);
7417
7418                 ret = btrfs_free_extent(trans, root,
7419                                         bytenr, num_bytes, leaf->start,
7420                                         btrfs_header_owner(leaf),
7421                                         btrfs_header_generation(leaf),
7422                                         key.objectid, 0);
7423                 BUG_ON(ret);
7424                 cond_resched();
7425         }
7426         kfree(new_extent);
7427         BUG_ON(ext_index + 1 != ref->nritems);
7428         btrfs_free_leaf_ref(root, ref);
7429         return 0;
7430 }
7431
7432 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7433                           struct btrfs_root *root)
7434 {
7435         struct btrfs_root *reloc_root;
7436         int ret;
7437
7438         if (root->reloc_root) {
7439                 reloc_root = root->reloc_root;
7440                 root->reloc_root = NULL;
7441                 list_add(&reloc_root->dead_list,
7442                          &root->fs_info->dead_reloc_roots);
7443
7444                 btrfs_set_root_bytenr(&reloc_root->root_item,
7445                                       reloc_root->node->start);
7446                 btrfs_set_root_level(&root->root_item,
7447                                      btrfs_header_level(reloc_root->node));
7448                 memset(&reloc_root->root_item.drop_progress, 0,
7449                         sizeof(struct btrfs_disk_key));
7450                 reloc_root->root_item.drop_level = 0;
7451
7452                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7453                                         &reloc_root->root_key,
7454                                         &reloc_root->root_item);
7455                 BUG_ON(ret);
7456         }
7457         return 0;
7458 }
7459
7460 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7461 {
7462         struct btrfs_trans_handle *trans;
7463         struct btrfs_root *reloc_root;
7464         struct btrfs_root *prev_root = NULL;
7465         struct list_head dead_roots;
7466         int ret;
7467         unsigned long nr;
7468
7469         INIT_LIST_HEAD(&dead_roots);
7470         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7471
7472         while (!list_empty(&dead_roots)) {
7473                 reloc_root = list_entry(dead_roots.prev,
7474                                         struct btrfs_root, dead_list);
7475                 list_del_init(&reloc_root->dead_list);
7476
7477                 BUG_ON(reloc_root->commit_root != NULL);
7478                 while (1) {
7479                         trans = btrfs_join_transaction(root, 1);
7480                         BUG_ON(!trans);
7481
7482                         mutex_lock(&root->fs_info->drop_mutex);
7483                         ret = btrfs_drop_snapshot(trans, reloc_root);
7484                         if (ret != -EAGAIN)
7485                                 break;
7486                         mutex_unlock(&root->fs_info->drop_mutex);
7487
7488                         nr = trans->blocks_used;
7489                         ret = btrfs_end_transaction(trans, root);
7490                         BUG_ON(ret);
7491                         btrfs_btree_balance_dirty(root, nr);
7492                 }
7493
7494                 free_extent_buffer(reloc_root->node);
7495
7496                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7497                                      &reloc_root->root_key);
7498                 BUG_ON(ret);
7499                 mutex_unlock(&root->fs_info->drop_mutex);
7500
7501                 nr = trans->blocks_used;
7502                 ret = btrfs_end_transaction(trans, root);
7503                 BUG_ON(ret);
7504                 btrfs_btree_balance_dirty(root, nr);
7505
7506                 kfree(prev_root);
7507                 prev_root = reloc_root;
7508         }
7509         if (prev_root) {
7510                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7511                 kfree(prev_root);
7512         }
7513         return 0;
7514 }
7515
7516 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7517 {
7518         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7519         return 0;
7520 }
7521
7522 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7523 {
7524         struct btrfs_root *reloc_root;
7525         struct btrfs_trans_handle *trans;
7526         struct btrfs_key location;
7527         int found;
7528         int ret;
7529
7530         mutex_lock(&root->fs_info->tree_reloc_mutex);
7531         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7532         BUG_ON(ret);
7533         found = !list_empty(&root->fs_info->dead_reloc_roots);
7534         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7535
7536         if (found) {
7537                 trans = btrfs_start_transaction(root, 1);
7538                 BUG_ON(!trans);
7539                 ret = btrfs_commit_transaction(trans, root);
7540                 BUG_ON(ret);
7541         }
7542
7543         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7544         location.offset = (u64)-1;
7545         location.type = BTRFS_ROOT_ITEM_KEY;
7546
7547         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7548         BUG_ON(!reloc_root);
7549         btrfs_orphan_cleanup(reloc_root);
7550         return 0;
7551 }
7552
7553 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7554                                     struct btrfs_root *root)
7555 {
7556         struct btrfs_root *reloc_root;
7557         struct extent_buffer *eb;
7558         struct btrfs_root_item *root_item;
7559         struct btrfs_key root_key;
7560         int ret;
7561
7562         BUG_ON(!root->ref_cows);
7563         if (root->reloc_root)
7564                 return 0;
7565
7566         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7567         BUG_ON(!root_item);
7568
7569         ret = btrfs_copy_root(trans, root, root->commit_root,
7570                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7571         BUG_ON(ret);
7572
7573         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7574         root_key.offset = root->root_key.objectid;
7575         root_key.type = BTRFS_ROOT_ITEM_KEY;
7576
7577         memcpy(root_item, &root->root_item, sizeof(root_item));
7578         btrfs_set_root_refs(root_item, 0);
7579         btrfs_set_root_bytenr(root_item, eb->start);
7580         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7581         btrfs_set_root_generation(root_item, trans->transid);
7582
7583         btrfs_tree_unlock(eb);
7584         free_extent_buffer(eb);
7585
7586         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7587                                 &root_key, root_item);
7588         BUG_ON(ret);
7589         kfree(root_item);
7590
7591         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7592                                                  &root_key);
7593         BUG_ON(!reloc_root);
7594         reloc_root->last_trans = trans->transid;
7595         reloc_root->commit_root = NULL;
7596         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7597
7598         root->reloc_root = reloc_root;
7599         return 0;
7600 }
7601
7602 /*
7603  * Core function of space balance.
7604  *
7605  * The idea is using reloc trees to relocate tree blocks in reference
7606  * counted roots. There is one reloc tree for each subvol, and all
7607  * reloc trees share same root key objectid. Reloc trees are snapshots
7608  * of the latest committed roots of subvols (root->commit_root).
7609  *
7610  * To relocate a tree block referenced by a subvol, there are two steps.
7611  * COW the block through subvol's reloc tree, then update block pointer
7612  * in the subvol to point to the new block. Since all reloc trees share
7613  * same root key objectid, doing special handing for tree blocks owned
7614  * by them is easy. Once a tree block has been COWed in one reloc tree,
7615  * we can use the resulting new block directly when the same block is
7616  * required to COW again through other reloc trees. By this way, relocated
7617  * tree blocks are shared between reloc trees, so they are also shared
7618  * between subvols.
7619  */
7620 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7621                                       struct btrfs_root *root,
7622                                       struct btrfs_path *path,
7623                                       struct btrfs_key *first_key,
7624                                       struct btrfs_ref_path *ref_path,
7625                                       struct btrfs_block_group_cache *group,
7626                                       struct inode *reloc_inode)
7627 {
7628         struct btrfs_root *reloc_root;
7629         struct extent_buffer *eb = NULL;
7630         struct btrfs_key *keys;
7631         u64 *nodes;
7632         int level;
7633         int shared_level;
7634         int lowest_level = 0;
7635         int ret;
7636
7637         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7638                 lowest_level = ref_path->owner_objectid;
7639
7640         if (!root->ref_cows) {
7641                 path->lowest_level = lowest_level;
7642                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7643                 BUG_ON(ret < 0);
7644                 path->lowest_level = 0;
7645                 btrfs_release_path(root, path);
7646                 return 0;
7647         }
7648
7649         mutex_lock(&root->fs_info->tree_reloc_mutex);
7650         ret = init_reloc_tree(trans, root);
7651         BUG_ON(ret);
7652         reloc_root = root->reloc_root;
7653
7654         shared_level = ref_path->shared_level;
7655         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7656
7657         keys = ref_path->node_keys;
7658         nodes = ref_path->new_nodes;
7659         memset(&keys[shared_level + 1], 0,
7660                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7661         memset(&nodes[shared_level + 1], 0,
7662                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7663
7664         if (nodes[lowest_level] == 0) {
7665                 path->lowest_level = lowest_level;
7666                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7667                                         0, 1);
7668                 BUG_ON(ret);
7669                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7670                         eb = path->nodes[level];
7671                         if (!eb || eb == reloc_root->node)
7672                                 break;
7673                         nodes[level] = eb->start;
7674                         if (level == 0)
7675                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7676                         else
7677                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7678                 }
7679                 if (nodes[0] &&
7680                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7681                         eb = path->nodes[0];
7682                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7683                                                       group, reloc_inode);
7684                         BUG_ON(ret);
7685                 }
7686                 btrfs_release_path(reloc_root, path);
7687         } else {
7688                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7689                                        lowest_level);
7690                 BUG_ON(ret);
7691         }
7692
7693         /*
7694          * replace tree blocks in the fs tree with tree blocks in
7695          * the reloc tree.
7696          */
7697         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7698         BUG_ON(ret < 0);
7699
7700         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7701                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7702                                         0, 0);
7703                 BUG_ON(ret);
7704                 extent_buffer_get(path->nodes[0]);
7705                 eb = path->nodes[0];
7706                 btrfs_release_path(reloc_root, path);
7707                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7708                 BUG_ON(ret);
7709                 free_extent_buffer(eb);
7710         }
7711
7712         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7713         path->lowest_level = 0;
7714         return 0;
7715 }
7716
7717 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7718                                         struct btrfs_root *root,
7719                                         struct btrfs_path *path,
7720                                         struct btrfs_key *first_key,
7721                                         struct btrfs_ref_path *ref_path)
7722 {
7723         int ret;
7724
7725         ret = relocate_one_path(trans, root, path, first_key,
7726                                 ref_path, NULL, NULL);
7727         BUG_ON(ret);
7728
7729         return 0;
7730 }
7731
7732 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7733                                     struct btrfs_root *extent_root,
7734                                     struct btrfs_path *path,
7735                                     struct btrfs_key *extent_key)
7736 {
7737         int ret;
7738
7739         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7740         if (ret)
7741                 goto out;
7742         ret = btrfs_del_item(trans, extent_root, path);
7743 out:
7744         btrfs_release_path(extent_root, path);
7745         return ret;
7746 }
7747
7748 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7749                                                 struct btrfs_ref_path *ref_path)
7750 {
7751         struct btrfs_key root_key;
7752
7753         root_key.objectid = ref_path->root_objectid;
7754         root_key.type = BTRFS_ROOT_ITEM_KEY;
7755         if (is_cowonly_root(ref_path->root_objectid))
7756                 root_key.offset = 0;
7757         else
7758                 root_key.offset = (u64)-1;
7759
7760         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7761 }
7762
7763 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7764                                         struct btrfs_path *path,
7765                                         struct btrfs_key *extent_key,
7766                                         struct btrfs_block_group_cache *group,
7767                                         struct inode *reloc_inode, int pass)
7768 {
7769         struct btrfs_trans_handle *trans;
7770         struct btrfs_root *found_root;
7771         struct btrfs_ref_path *ref_path = NULL;
7772         struct disk_extent *new_extents = NULL;
7773         int nr_extents = 0;
7774         int loops;
7775         int ret;
7776         int level;
7777         struct btrfs_key first_key;
7778         u64 prev_block = 0;
7779
7780
7781         trans = btrfs_start_transaction(extent_root, 1);
7782         BUG_ON(!trans);
7783
7784         if (extent_key->objectid == 0) {
7785                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7786                 goto out;
7787         }
7788
7789         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7790         if (!ref_path) {
7791                 ret = -ENOMEM;
7792                 goto out;
7793         }
7794
7795         for (loops = 0; ; loops++) {
7796                 if (loops == 0) {
7797                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7798                                                    extent_key->objectid);
7799                 } else {
7800                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7801                 }
7802                 if (ret < 0)
7803                         goto out;
7804                 if (ret > 0)
7805                         break;
7806
7807                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7808                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7809                         continue;
7810
7811                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7812                 BUG_ON(!found_root);
7813                 /*
7814                  * for reference counted tree, only process reference paths
7815                  * rooted at the latest committed root.
7816                  */
7817                 if (found_root->ref_cows &&
7818                     ref_path->root_generation != found_root->root_key.offset)
7819                         continue;
7820
7821                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7822                         if (pass == 0) {
7823                                 /*
7824                                  * copy data extents to new locations
7825                                  */
7826                                 u64 group_start = group->key.objectid;
7827                                 ret = relocate_data_extent(reloc_inode,
7828                                                            extent_key,
7829                                                            group_start);
7830                                 if (ret < 0)
7831                                         goto out;
7832                                 break;
7833                         }
7834                         level = 0;
7835                 } else {
7836                         level = ref_path->owner_objectid;
7837                 }
7838
7839                 if (prev_block != ref_path->nodes[level]) {
7840                         struct extent_buffer *eb;
7841                         u64 block_start = ref_path->nodes[level];
7842                         u64 block_size = btrfs_level_size(found_root, level);
7843
7844                         eb = read_tree_block(found_root, block_start,
7845                                              block_size, 0);
7846                         btrfs_tree_lock(eb);
7847                         BUG_ON(level != btrfs_header_level(eb));
7848
7849                         if (level == 0)
7850                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7851                         else
7852                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7853
7854                         btrfs_tree_unlock(eb);
7855                         free_extent_buffer(eb);
7856                         prev_block = block_start;
7857                 }
7858
7859                 mutex_lock(&extent_root->fs_info->trans_mutex);
7860                 btrfs_record_root_in_trans(found_root);
7861                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7862                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7863                         /*
7864                          * try to update data extent references while
7865                          * keeping metadata shared between snapshots.
7866                          */
7867                         if (pass == 1) {
7868                                 ret = relocate_one_path(trans, found_root,
7869                                                 path, &first_key, ref_path,
7870                                                 group, reloc_inode);
7871                                 if (ret < 0)
7872                                         goto out;
7873                                 continue;
7874                         }
7875                         /*
7876                          * use fallback method to process the remaining
7877                          * references.
7878                          */
7879                         if (!new_extents) {
7880                                 u64 group_start = group->key.objectid;
7881                                 new_extents = kmalloc(sizeof(*new_extents),
7882                                                       GFP_NOFS);
7883                                 nr_extents = 1;
7884                                 ret = get_new_locations(reloc_inode,
7885                                                         extent_key,
7886                                                         group_start, 1,
7887                                                         &new_extents,
7888                                                         &nr_extents);
7889                                 if (ret)
7890                                         goto out;
7891                         }
7892                         ret = replace_one_extent(trans, found_root,
7893                                                 path, extent_key,
7894                                                 &first_key, ref_path,
7895                                                 new_extents, nr_extents);
7896                 } else {
7897                         ret = relocate_tree_block(trans, found_root, path,
7898                                                   &first_key, ref_path);
7899                 }
7900                 if (ret < 0)
7901                         goto out;
7902         }
7903         ret = 0;
7904 out:
7905         btrfs_end_transaction(trans, extent_root);
7906         kfree(new_extents);
7907         kfree(ref_path);
7908         return ret;
7909 }
7910 #endif
7911
7912 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7913 {
7914         u64 num_devices;
7915         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7916                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7917
7918         /*
7919          * we add in the count of missing devices because we want
7920          * to make sure that any RAID levels on a degraded FS
7921          * continue to be honored.
7922          */
7923         num_devices = root->fs_info->fs_devices->rw_devices +
7924                 root->fs_info->fs_devices->missing_devices;
7925
7926         if (num_devices == 1) {
7927                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7928                 stripped = flags & ~stripped;
7929
7930                 /* turn raid0 into single device chunks */
7931                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7932                         return stripped;
7933
7934                 /* turn mirroring into duplication */
7935                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7936                              BTRFS_BLOCK_GROUP_RAID10))
7937                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7938                 return flags;
7939         } else {
7940                 /* they already had raid on here, just return */
7941                 if (flags & stripped)
7942                         return flags;
7943
7944                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7945                 stripped = flags & ~stripped;
7946
7947                 /* switch duplicated blocks with raid1 */
7948                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7949                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7950
7951                 /* turn single device chunks into raid0 */
7952                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7953         }
7954         return flags;
7955 }
7956
7957 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7958 {
7959         struct btrfs_space_info *sinfo = cache->space_info;
7960         u64 num_bytes;
7961         int ret = -ENOSPC;
7962
7963         if (cache->ro)
7964                 return 0;
7965
7966         spin_lock(&sinfo->lock);
7967         spin_lock(&cache->lock);
7968         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7969                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7970
7971         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7972             sinfo->bytes_may_use + sinfo->bytes_readonly +
7973             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
7974                 sinfo->bytes_readonly += num_bytes;
7975                 sinfo->bytes_reserved += cache->reserved_pinned;
7976                 cache->reserved_pinned = 0;
7977                 cache->ro = 1;
7978                 ret = 0;
7979         }
7980
7981         spin_unlock(&cache->lock);
7982         spin_unlock(&sinfo->lock);
7983         return ret;
7984 }
7985
7986 int btrfs_set_block_group_ro(struct btrfs_root *root,
7987                              struct btrfs_block_group_cache *cache)
7988
7989 {
7990         struct btrfs_trans_handle *trans;
7991         u64 alloc_flags;
7992         int ret;
7993
7994         BUG_ON(cache->ro);
7995
7996         trans = btrfs_join_transaction(root, 1);
7997         BUG_ON(IS_ERR(trans));
7998
7999         alloc_flags = update_block_group_flags(root, cache->flags);
8000         if (alloc_flags != cache->flags)
8001                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8002
8003         ret = set_block_group_ro(cache);
8004         if (!ret)
8005                 goto out;
8006         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8007         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8008         if (ret < 0)
8009                 goto out;
8010         ret = set_block_group_ro(cache);
8011 out:
8012         btrfs_end_transaction(trans, root);
8013         return ret;
8014 }
8015
8016 /*
8017  * helper to account the unused space of all the readonly block group in the
8018  * list. takes mirrors into account.
8019  */
8020 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8021 {
8022         struct btrfs_block_group_cache *block_group;
8023         u64 free_bytes = 0;
8024         int factor;
8025
8026         list_for_each_entry(block_group, groups_list, list) {
8027                 spin_lock(&block_group->lock);
8028
8029                 if (!block_group->ro) {
8030                         spin_unlock(&block_group->lock);
8031                         continue;
8032                 }
8033
8034                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8035                                           BTRFS_BLOCK_GROUP_RAID10 |
8036                                           BTRFS_BLOCK_GROUP_DUP))
8037                         factor = 2;
8038                 else
8039                         factor = 1;
8040
8041                 free_bytes += (block_group->key.offset -
8042                                btrfs_block_group_used(&block_group->item)) *
8043                                factor;
8044
8045                 spin_unlock(&block_group->lock);
8046         }
8047
8048         return free_bytes;
8049 }
8050
8051 /*
8052  * helper to account the unused space of all the readonly block group in the
8053  * space_info. takes mirrors into account.
8054  */
8055 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8056 {
8057         int i;
8058         u64 free_bytes = 0;
8059
8060         spin_lock(&sinfo->lock);
8061
8062         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8063                 if (!list_empty(&sinfo->block_groups[i]))
8064                         free_bytes += __btrfs_get_ro_block_group_free_space(
8065                                                 &sinfo->block_groups[i]);
8066
8067         spin_unlock(&sinfo->lock);
8068
8069         return free_bytes;
8070 }
8071
8072 int btrfs_set_block_group_rw(struct btrfs_root *root,
8073                               struct btrfs_block_group_cache *cache)
8074 {
8075         struct btrfs_space_info *sinfo = cache->space_info;
8076         u64 num_bytes;
8077
8078         BUG_ON(!cache->ro);
8079
8080         spin_lock(&sinfo->lock);
8081         spin_lock(&cache->lock);
8082         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8083                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8084         sinfo->bytes_readonly -= num_bytes;
8085         cache->ro = 0;
8086         spin_unlock(&cache->lock);
8087         spin_unlock(&sinfo->lock);
8088         return 0;
8089 }
8090
8091 /*
8092  * checks to see if its even possible to relocate this block group.
8093  *
8094  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8095  * ok to go ahead and try.
8096  */
8097 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8098 {
8099         struct btrfs_block_group_cache *block_group;
8100         struct btrfs_space_info *space_info;
8101         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8102         struct btrfs_device *device;
8103         int full = 0;
8104         int ret = 0;
8105
8106         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8107
8108         /* odd, couldn't find the block group, leave it alone */
8109         if (!block_group)
8110                 return -1;
8111
8112         /* no bytes used, we're good */
8113         if (!btrfs_block_group_used(&block_group->item))
8114                 goto out;
8115
8116         space_info = block_group->space_info;
8117         spin_lock(&space_info->lock);
8118
8119         full = space_info->full;
8120
8121         /*
8122          * if this is the last block group we have in this space, we can't
8123          * relocate it unless we're able to allocate a new chunk below.
8124          *
8125          * Otherwise, we need to make sure we have room in the space to handle
8126          * all of the extents from this block group.  If we can, we're good
8127          */
8128         if ((space_info->total_bytes != block_group->key.offset) &&
8129            (space_info->bytes_used + space_info->bytes_reserved +
8130             space_info->bytes_pinned + space_info->bytes_readonly +
8131             btrfs_block_group_used(&block_group->item) <
8132             space_info->total_bytes)) {
8133                 spin_unlock(&space_info->lock);
8134                 goto out;
8135         }
8136         spin_unlock(&space_info->lock);
8137
8138         /*
8139          * ok we don't have enough space, but maybe we have free space on our
8140          * devices to allocate new chunks for relocation, so loop through our
8141          * alloc devices and guess if we have enough space.  However, if we
8142          * were marked as full, then we know there aren't enough chunks, and we
8143          * can just return.
8144          */
8145         ret = -1;
8146         if (full)
8147                 goto out;
8148
8149         mutex_lock(&root->fs_info->chunk_mutex);
8150         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8151                 u64 min_free = btrfs_block_group_used(&block_group->item);
8152                 u64 dev_offset;
8153
8154                 /*
8155                  * check to make sure we can actually find a chunk with enough
8156                  * space to fit our block group in.
8157                  */
8158                 if (device->total_bytes > device->bytes_used + min_free) {
8159                         ret = find_free_dev_extent(NULL, device, min_free,
8160                                                    &dev_offset, NULL);
8161                         if (!ret)
8162                                 break;
8163                         ret = -1;
8164                 }
8165         }
8166         mutex_unlock(&root->fs_info->chunk_mutex);
8167 out:
8168         btrfs_put_block_group(block_group);
8169         return ret;
8170 }
8171
8172 static int find_first_block_group(struct btrfs_root *root,
8173                 struct btrfs_path *path, struct btrfs_key *key)
8174 {
8175         int ret = 0;
8176         struct btrfs_key found_key;
8177         struct extent_buffer *leaf;
8178         int slot;
8179
8180         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8181         if (ret < 0)
8182                 goto out;
8183
8184         while (1) {
8185                 slot = path->slots[0];
8186                 leaf = path->nodes[0];
8187                 if (slot >= btrfs_header_nritems(leaf)) {
8188                         ret = btrfs_next_leaf(root, path);
8189                         if (ret == 0)
8190                                 continue;
8191                         if (ret < 0)
8192                                 goto out;
8193                         break;
8194                 }
8195                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8196
8197                 if (found_key.objectid >= key->objectid &&
8198                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8199                         ret = 0;
8200                         goto out;
8201                 }
8202                 path->slots[0]++;
8203         }
8204 out:
8205         return ret;
8206 }
8207
8208 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8209 {
8210         struct btrfs_block_group_cache *block_group;
8211         u64 last = 0;
8212
8213         while (1) {
8214                 struct inode *inode;
8215
8216                 block_group = btrfs_lookup_first_block_group(info, last);
8217                 while (block_group) {
8218                         spin_lock(&block_group->lock);
8219                         if (block_group->iref)
8220                                 break;
8221                         spin_unlock(&block_group->lock);
8222                         block_group = next_block_group(info->tree_root,
8223                                                        block_group);
8224                 }
8225                 if (!block_group) {
8226                         if (last == 0)
8227                                 break;
8228                         last = 0;
8229                         continue;
8230                 }
8231
8232                 inode = block_group->inode;
8233                 block_group->iref = 0;
8234                 block_group->inode = NULL;
8235                 spin_unlock(&block_group->lock);
8236                 iput(inode);
8237                 last = block_group->key.objectid + block_group->key.offset;
8238                 btrfs_put_block_group(block_group);
8239         }
8240 }
8241
8242 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8243 {
8244         struct btrfs_block_group_cache *block_group;
8245         struct btrfs_space_info *space_info;
8246         struct btrfs_caching_control *caching_ctl;
8247         struct rb_node *n;
8248
8249         down_write(&info->extent_commit_sem);
8250         while (!list_empty(&info->caching_block_groups)) {
8251                 caching_ctl = list_entry(info->caching_block_groups.next,
8252                                          struct btrfs_caching_control, list);
8253                 list_del(&caching_ctl->list);
8254                 put_caching_control(caching_ctl);
8255         }
8256         up_write(&info->extent_commit_sem);
8257
8258         spin_lock(&info->block_group_cache_lock);
8259         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8260                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8261                                        cache_node);
8262                 rb_erase(&block_group->cache_node,
8263                          &info->block_group_cache_tree);
8264                 spin_unlock(&info->block_group_cache_lock);
8265
8266                 down_write(&block_group->space_info->groups_sem);
8267                 list_del(&block_group->list);
8268                 up_write(&block_group->space_info->groups_sem);
8269
8270                 if (block_group->cached == BTRFS_CACHE_STARTED)
8271                         wait_block_group_cache_done(block_group);
8272
8273                 btrfs_remove_free_space_cache(block_group);
8274                 btrfs_put_block_group(block_group);
8275
8276                 spin_lock(&info->block_group_cache_lock);
8277         }
8278         spin_unlock(&info->block_group_cache_lock);
8279
8280         /* now that all the block groups are freed, go through and
8281          * free all the space_info structs.  This is only called during
8282          * the final stages of unmount, and so we know nobody is
8283          * using them.  We call synchronize_rcu() once before we start,
8284          * just to be on the safe side.
8285          */
8286         synchronize_rcu();
8287
8288         release_global_block_rsv(info);
8289
8290         while(!list_empty(&info->space_info)) {
8291                 space_info = list_entry(info->space_info.next,
8292                                         struct btrfs_space_info,
8293                                         list);
8294                 if (space_info->bytes_pinned > 0 ||
8295                     space_info->bytes_reserved > 0) {
8296                         WARN_ON(1);
8297                         dump_space_info(space_info, 0, 0);
8298                 }
8299                 list_del(&space_info->list);
8300                 kfree(space_info);
8301         }
8302         return 0;
8303 }
8304
8305 static void __link_block_group(struct btrfs_space_info *space_info,
8306                                struct btrfs_block_group_cache *cache)
8307 {
8308         int index = get_block_group_index(cache);
8309
8310         down_write(&space_info->groups_sem);
8311         list_add_tail(&cache->list, &space_info->block_groups[index]);
8312         up_write(&space_info->groups_sem);
8313 }
8314
8315 int btrfs_read_block_groups(struct btrfs_root *root)
8316 {
8317         struct btrfs_path *path;
8318         int ret;
8319         struct btrfs_block_group_cache *cache;
8320         struct btrfs_fs_info *info = root->fs_info;
8321         struct btrfs_space_info *space_info;
8322         struct btrfs_key key;
8323         struct btrfs_key found_key;
8324         struct extent_buffer *leaf;
8325         int need_clear = 0;
8326         u64 cache_gen;
8327
8328         root = info->extent_root;
8329         key.objectid = 0;
8330         key.offset = 0;
8331         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8332         path = btrfs_alloc_path();
8333         if (!path)
8334                 return -ENOMEM;
8335
8336         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8337         if (cache_gen != 0 &&
8338             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8339                 need_clear = 1;
8340         if (btrfs_test_opt(root, CLEAR_CACHE))
8341                 need_clear = 1;
8342         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8343                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8344
8345         while (1) {
8346                 ret = find_first_block_group(root, path, &key);
8347                 if (ret > 0)
8348                         break;
8349                 if (ret != 0)
8350                         goto error;
8351                 leaf = path->nodes[0];
8352                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8353                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8354                 if (!cache) {
8355                         ret = -ENOMEM;
8356                         goto error;
8357                 }
8358
8359                 atomic_set(&cache->count, 1);
8360                 spin_lock_init(&cache->lock);
8361                 spin_lock_init(&cache->tree_lock);
8362                 cache->fs_info = info;
8363                 INIT_LIST_HEAD(&cache->list);
8364                 INIT_LIST_HEAD(&cache->cluster_list);
8365
8366                 if (need_clear)
8367                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8368
8369                 /*
8370                  * we only want to have 32k of ram per block group for keeping
8371                  * track of free space, and if we pass 1/2 of that we want to
8372                  * start converting things over to using bitmaps
8373                  */
8374                 cache->extents_thresh = ((1024 * 32) / 2) /
8375                         sizeof(struct btrfs_free_space);
8376
8377                 read_extent_buffer(leaf, &cache->item,
8378                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8379                                    sizeof(cache->item));
8380                 memcpy(&cache->key, &found_key, sizeof(found_key));
8381
8382                 key.objectid = found_key.objectid + found_key.offset;
8383                 btrfs_release_path(root, path);
8384                 cache->flags = btrfs_block_group_flags(&cache->item);
8385                 cache->sectorsize = root->sectorsize;
8386
8387                 /*
8388                  * check for two cases, either we are full, and therefore
8389                  * don't need to bother with the caching work since we won't
8390                  * find any space, or we are empty, and we can just add all
8391                  * the space in and be done with it.  This saves us _alot_ of
8392                  * time, particularly in the full case.
8393                  */
8394                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8395                         exclude_super_stripes(root, cache);
8396                         cache->last_byte_to_unpin = (u64)-1;
8397                         cache->cached = BTRFS_CACHE_FINISHED;
8398                         free_excluded_extents(root, cache);
8399                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8400                         exclude_super_stripes(root, cache);
8401                         cache->last_byte_to_unpin = (u64)-1;
8402                         cache->cached = BTRFS_CACHE_FINISHED;
8403                         add_new_free_space(cache, root->fs_info,
8404                                            found_key.objectid,
8405                                            found_key.objectid +
8406                                            found_key.offset);
8407                         free_excluded_extents(root, cache);
8408                 }
8409
8410                 ret = update_space_info(info, cache->flags, found_key.offset,
8411                                         btrfs_block_group_used(&cache->item),
8412                                         &space_info);
8413                 BUG_ON(ret);
8414                 cache->space_info = space_info;
8415                 spin_lock(&cache->space_info->lock);
8416                 cache->space_info->bytes_readonly += cache->bytes_super;
8417                 spin_unlock(&cache->space_info->lock);
8418
8419                 __link_block_group(space_info, cache);
8420
8421                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8422                 BUG_ON(ret);
8423
8424                 set_avail_alloc_bits(root->fs_info, cache->flags);
8425                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8426                         set_block_group_ro(cache);
8427         }
8428
8429         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8430                 if (!(get_alloc_profile(root, space_info->flags) &
8431                       (BTRFS_BLOCK_GROUP_RAID10 |
8432                        BTRFS_BLOCK_GROUP_RAID1 |
8433                        BTRFS_BLOCK_GROUP_DUP)))
8434                         continue;
8435                 /*
8436                  * avoid allocating from un-mirrored block group if there are
8437                  * mirrored block groups.
8438                  */
8439                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8440                         set_block_group_ro(cache);
8441                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8442                         set_block_group_ro(cache);
8443         }
8444
8445         init_global_block_rsv(info);
8446         ret = 0;
8447 error:
8448         btrfs_free_path(path);
8449         return ret;
8450 }
8451
8452 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8453                            struct btrfs_root *root, u64 bytes_used,
8454                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8455                            u64 size)
8456 {
8457         int ret;
8458         struct btrfs_root *extent_root;
8459         struct btrfs_block_group_cache *cache;
8460
8461         extent_root = root->fs_info->extent_root;
8462
8463         root->fs_info->last_trans_log_full_commit = trans->transid;
8464
8465         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8466         if (!cache)
8467                 return -ENOMEM;
8468
8469         cache->key.objectid = chunk_offset;
8470         cache->key.offset = size;
8471         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8472         cache->sectorsize = root->sectorsize;
8473         cache->fs_info = root->fs_info;
8474
8475         /*
8476          * we only want to have 32k of ram per block group for keeping track
8477          * of free space, and if we pass 1/2 of that we want to start
8478          * converting things over to using bitmaps
8479          */
8480         cache->extents_thresh = ((1024 * 32) / 2) /
8481                 sizeof(struct btrfs_free_space);
8482         atomic_set(&cache->count, 1);
8483         spin_lock_init(&cache->lock);
8484         spin_lock_init(&cache->tree_lock);
8485         INIT_LIST_HEAD(&cache->list);
8486         INIT_LIST_HEAD(&cache->cluster_list);
8487
8488         btrfs_set_block_group_used(&cache->item, bytes_used);
8489         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8490         cache->flags = type;
8491         btrfs_set_block_group_flags(&cache->item, type);
8492
8493         cache->last_byte_to_unpin = (u64)-1;
8494         cache->cached = BTRFS_CACHE_FINISHED;
8495         exclude_super_stripes(root, cache);
8496
8497         add_new_free_space(cache, root->fs_info, chunk_offset,
8498                            chunk_offset + size);
8499
8500         free_excluded_extents(root, cache);
8501
8502         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8503                                 &cache->space_info);
8504         BUG_ON(ret);
8505
8506         spin_lock(&cache->space_info->lock);
8507         cache->space_info->bytes_readonly += cache->bytes_super;
8508         spin_unlock(&cache->space_info->lock);
8509
8510         __link_block_group(cache->space_info, cache);
8511
8512         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8513         BUG_ON(ret);
8514
8515         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8516                                 sizeof(cache->item));
8517         BUG_ON(ret);
8518
8519         set_avail_alloc_bits(extent_root->fs_info, type);
8520
8521         return 0;
8522 }
8523
8524 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8525                              struct btrfs_root *root, u64 group_start)
8526 {
8527         struct btrfs_path *path;
8528         struct btrfs_block_group_cache *block_group;
8529         struct btrfs_free_cluster *cluster;
8530         struct btrfs_root *tree_root = root->fs_info->tree_root;
8531         struct btrfs_key key;
8532         struct inode *inode;
8533         int ret;
8534         int factor;
8535
8536         root = root->fs_info->extent_root;
8537
8538         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8539         BUG_ON(!block_group);
8540         BUG_ON(!block_group->ro);
8541
8542         memcpy(&key, &block_group->key, sizeof(key));
8543         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8544                                   BTRFS_BLOCK_GROUP_RAID1 |
8545                                   BTRFS_BLOCK_GROUP_RAID10))
8546                 factor = 2;
8547         else
8548                 factor = 1;
8549
8550         /* make sure this block group isn't part of an allocation cluster */
8551         cluster = &root->fs_info->data_alloc_cluster;
8552         spin_lock(&cluster->refill_lock);
8553         btrfs_return_cluster_to_free_space(block_group, cluster);
8554         spin_unlock(&cluster->refill_lock);
8555
8556         /*
8557          * make sure this block group isn't part of a metadata
8558          * allocation cluster
8559          */
8560         cluster = &root->fs_info->meta_alloc_cluster;
8561         spin_lock(&cluster->refill_lock);
8562         btrfs_return_cluster_to_free_space(block_group, cluster);
8563         spin_unlock(&cluster->refill_lock);
8564
8565         path = btrfs_alloc_path();
8566         BUG_ON(!path);
8567
8568         inode = lookup_free_space_inode(root, block_group, path);
8569         if (!IS_ERR(inode)) {
8570                 btrfs_orphan_add(trans, inode);
8571                 clear_nlink(inode);
8572                 /* One for the block groups ref */
8573                 spin_lock(&block_group->lock);
8574                 if (block_group->iref) {
8575                         block_group->iref = 0;
8576                         block_group->inode = NULL;
8577                         spin_unlock(&block_group->lock);
8578                         iput(inode);
8579                 } else {
8580                         spin_unlock(&block_group->lock);
8581                 }
8582                 /* One for our lookup ref */
8583                 iput(inode);
8584         }
8585
8586         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8587         key.offset = block_group->key.objectid;
8588         key.type = 0;
8589
8590         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8591         if (ret < 0)
8592                 goto out;
8593         if (ret > 0)
8594                 btrfs_release_path(tree_root, path);
8595         if (ret == 0) {
8596                 ret = btrfs_del_item(trans, tree_root, path);
8597                 if (ret)
8598                         goto out;
8599                 btrfs_release_path(tree_root, path);
8600         }
8601
8602         spin_lock(&root->fs_info->block_group_cache_lock);
8603         rb_erase(&block_group->cache_node,
8604                  &root->fs_info->block_group_cache_tree);
8605         spin_unlock(&root->fs_info->block_group_cache_lock);
8606
8607         down_write(&block_group->space_info->groups_sem);
8608         /*
8609          * we must use list_del_init so people can check to see if they
8610          * are still on the list after taking the semaphore
8611          */
8612         list_del_init(&block_group->list);
8613         up_write(&block_group->space_info->groups_sem);
8614
8615         if (block_group->cached == BTRFS_CACHE_STARTED)
8616                 wait_block_group_cache_done(block_group);
8617
8618         btrfs_remove_free_space_cache(block_group);
8619
8620         spin_lock(&block_group->space_info->lock);
8621         block_group->space_info->total_bytes -= block_group->key.offset;
8622         block_group->space_info->bytes_readonly -= block_group->key.offset;
8623         block_group->space_info->disk_total -= block_group->key.offset * factor;
8624         spin_unlock(&block_group->space_info->lock);
8625
8626         memcpy(&key, &block_group->key, sizeof(key));
8627
8628         btrfs_clear_space_info_full(root->fs_info);
8629
8630         btrfs_put_block_group(block_group);
8631         btrfs_put_block_group(block_group);
8632
8633         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8634         if (ret > 0)
8635                 ret = -EIO;
8636         if (ret < 0)
8637                 goto out;
8638
8639         ret = btrfs_del_item(trans, root, path);
8640 out:
8641         btrfs_free_path(path);
8642         return ret;
8643 }
8644
8645 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8646 {
8647         return unpin_extent_range(root, start, end);
8648 }
8649
8650 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8651                                u64 num_bytes)
8652 {
8653         return btrfs_discard_extent(root, bytenr, num_bytes);
8654 }