fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <linux/slab.h>
  26 #include "compat.h"
  27 #include "hash.h"
  28 #include "ctree.h"
  29 #include "disk-io.h"
  30 #include "print-tree.h"
  31 #include "transaction.h"
  32 #include "volumes.h"
  33 #include "locking.h"
  34 #include "free-space-cache.h"
  35
  36 /* control flags for do_chunk_alloc's force field
  37  * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
  38  * if we really need one.
  39  *
  40  * CHUNK_ALLOC_FORCE means it must try to allocate one
  41  *
  42  * CHUNK_ALLOC_LIMITED means to only try and allocate one
  43  * if we have very few chunks already allocated.  This is
  44  * used as part of the clustering code to help make sure
  45  * we have a good pool of storage to cluster in, without
  46  * filling the FS with empty chunks
  47  *
  48  */
  49 enum {
  50         CHUNK_ALLOC_NO_FORCE = 0,
  51         CHUNK_ALLOC_FORCE = 1,
  52         CHUNK_ALLOC_LIMITED = 2,
  53 };
  54
  55 static int update_block_group(struct btrfs_trans_handle *trans,
  56                               struct btrfs_root *root,
  57                               u64 bytenr, u64 num_bytes, int alloc);
  58 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  59                                 struct btrfs_root *root,
  60                                 u64 bytenr, u64 num_bytes, u64 parent,
  61                                 u64 root_objectid, u64 owner_objectid,
  62                                 u64 owner_offset, int refs_to_drop,
  63                                 struct btrfs_delayed_extent_op *extra_op);
  64 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
  65                                     struct extent_buffer *leaf,
  66                                     struct btrfs_extent_item *ei);
  67 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
  68                                       struct btrfs_root *root,
  69                                       u64 parent, u64 root_objectid,
  70                                       u64 flags, u64 owner, u64 offset,
  71                                       struct btrfs_key *ins, int ref_mod);
  72 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
  73                                      struct btrfs_root *root,
  74                                      u64 parent, u64 root_objectid,
  75                                      u64 flags, struct btrfs_disk_key *key,
  76                                      int level, struct btrfs_key *ins);
  77 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  78                           struct btrfs_root *extent_root, u64 alloc_bytes,
  79                           u64 flags, int force);
  80 static int find_next_key(struct btrfs_path *path, int level,
  81                          struct btrfs_key *key);
  82 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  83                             int dump_block_groups);
  84
  85 static noinline int
  86 block_group_cache_done(struct btrfs_block_group_cache *cache)
  87 {
  88         smp_mb();
  89         return cache->cached == BTRFS_CACHE_FINISHED;
  90 }
  91
  92 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  93 {
  94         return (cache->flags & bits) == bits;
  95 }
  96
  97 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
  98 {
  99         atomic_inc(&cache->count);
 100 }
 101
 102 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 103 {
 104         if (atomic_dec_and_test(&cache->count)) {
 105                 WARN_ON(cache->pinned > 0);
 106                 WARN_ON(cache->reserved > 0);
 107                 WARN_ON(cache->reserved_pinned > 0);
 108                 kfree(cache);
 109         }
 110 }
 111
 112 /*
 113  * this adds the block group to the fs_info rb tree for the block group
 114  * cache
 115  */
 116 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
 117                                 struct btrfs_block_group_cache *block_group)
 118 {
 119         struct rb_node **p;
 120         struct rb_node *parent = NULL;
 121         struct btrfs_block_group_cache *cache;
 122
 123         spin_lock(&info->block_group_cache_lock);
 124         p = &info->block_group_cache_tree.rb_node;
 125
 126         while (*p) {
 127                 parent = *p;
 128                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 129                                  cache_node);
 130                 if (block_group->key.objectid < cache->key.objectid) {
 131                         p = &(*p)->rb_left;
 132                 } else if (block_group->key.objectid > cache->key.objectid) {
 133                         p = &(*p)->rb_right;
 134                 } else {
 135                         spin_unlock(&info->block_group_cache_lock);
 136                         return -EEXIST;
 137                 }
 138         }
 139
 140         rb_link_node(&block_group->cache_node, parent, p);
 141         rb_insert_color(&block_group->cache_node,
 142                         &info->block_group_cache_tree);
 143         spin_unlock(&info->block_group_cache_lock);
 144
 145         return 0;
 146 }
 147
 148 /*
 149  * This will return the block group at or after bytenr if contains is 0, else
 150  * it will return the block group that contains the bytenr
 151  */
 152 static struct btrfs_block_group_cache *
 153 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 154                               int contains)
 155 {
 156         struct btrfs_block_group_cache *cache, *ret = NULL;
 157         struct rb_node *n;
 158         u64 end, start;
 159
 160         spin_lock(&info->block_group_cache_lock);
 161         n = info->block_group_cache_tree.rb_node;
 162
 163         while (n) {
 164                 cache = rb_entry(n, struct btrfs_block_group_cache,
 165                                  cache_node);
 166                 end = cache->key.objectid + cache->key.offset - 1;
 167                 start = cache->key.objectid;
 168
 169                 if (bytenr < start) {
 170                         if (!contains && (!ret || start < ret->key.objectid))
 171                                 ret = cache;
 172                         n = n->rb_left;
 173                 } else if (bytenr > start) {
 174                         if (contains && bytenr <= end) {
 175                                 ret = cache;
 176                                 break;
 177                         }
 178                         n = n->rb_right;
 179                 } else {
 180                         ret = cache;
 181                         break;
 182                 }
 183         }
 184         if (ret)
 185                 btrfs_get_block_group(ret);
 186         spin_unlock(&info->block_group_cache_lock);
 187
 188         return ret;
 189 }
 190
 191 static int add_excluded_extent(struct btrfs_root *root,
 192                                u64 start, u64 num_bytes)
 193 {
 194         u64 end = start + num_bytes - 1;
 195         set_extent_bits(&root->fs_info->freed_extents[0],
 196                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 197         set_extent_bits(&root->fs_info->freed_extents[1],
 198                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 199         return 0;
 200 }
 201
 202 static void free_excluded_extents(struct btrfs_root *root,
 203                                   struct btrfs_block_group_cache *cache)
 204 {
 205         u64 start, end;
 206
 207         start = cache->key.objectid;
 208         end = start + cache->key.offset - 1;
 209
 210         clear_extent_bits(&root->fs_info->freed_extents[0],
 211                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 212         clear_extent_bits(&root->fs_info->freed_extents[1],
 213                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 214 }
 215
 216 static int exclude_super_stripes(struct btrfs_root *root,
 217                                  struct btrfs_block_group_cache *cache)
 218 {
 219         u64 bytenr;
 220         u64 *logical;
 221         int stripe_len;
 222         int i, nr, ret;
 223
 224         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
 225                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
 226                 cache->bytes_super += stripe_len;
 227                 ret = add_excluded_extent(root, cache->key.objectid,
 228                                           stripe_len);
 229                 BUG_ON(ret);
 230         }
 231
 232         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 233                 bytenr = btrfs_sb_offset(i);
 234                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 235                                        cache->key.objectid, bytenr,
 236                                        0, &logical, &nr, &stripe_len);
 237                 BUG_ON(ret);
 238
 239                 while (nr--) {
 240                         cache->bytes_super += stripe_len;
 241                         ret = add_excluded_extent(root, logical[nr],
 242                                                   stripe_len);
 243                         BUG_ON(ret);
 244                 }
 245
 246                 kfree(logical);
 247         }
 248         return 0;
 249 }
 250
 251 static struct btrfs_caching_control *
 252 get_caching_control(struct btrfs_block_group_cache *cache)
 253 {
 254         struct btrfs_caching_control *ctl;
 255
 256         spin_lock(&cache->lock);
 257         if (cache->cached != BTRFS_CACHE_STARTED) {
 258                 spin_unlock(&cache->lock);
 259                 return NULL;
 260         }
 261
 262         /* We're loading it the fast way, so we don't have a caching_ctl. */
 263         if (!cache->caching_ctl) {
 264                 spin_unlock(&cache->lock);
 265                 return NULL;
 266         }
 267
 268         ctl = cache->caching_ctl;
 269         atomic_inc(&ctl->count);
 270         spin_unlock(&cache->lock);
 271         return ctl;
 272 }
 273
 274 static void put_caching_control(struct btrfs_caching_control *ctl)
 275 {
 276         if (atomic_dec_and_test(&ctl->count))
 277                 kfree(ctl);
 278 }
 279
 280 /*
 281  * this is only called by cache_block_group, since we could have freed extents
 282  * we need to check the pinned_extents for any extents that can't be used yet
 283  * since their free space will be released as soon as the transaction commits.
 284  */
 285 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 286                               struct btrfs_fs_info *info, u64 start, u64 end)
 287 {
 288         u64 extent_start, extent_end, size, total_added = 0;
 289         int ret;
 290
 291         while (start < end) {
 292                 ret = find_first_extent_bit(info->pinned_extents, start,
 293                                             &extent_start, &extent_end,
 294                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 295                 if (ret)
 296                         break;
 297
 298                 if (extent_start <= start) {
 299                         start = extent_end + 1;
 300                 } else if (extent_start > start && extent_start < end) {
 301                         size = extent_start - start;
 302                         total_added += size;
 303                         ret = btrfs_add_free_space(block_group, start,
 304                                                    size);
 305                         BUG_ON(ret);
 306                         start = extent_end + 1;
 307                 } else {
 308                         break;
 309                 }
 310         }
 311
 312         if (start < end) {
 313                 size = end - start;
 314                 total_added += size;
 315                 ret = btrfs_add_free_space(block_group, start, size);
 316                 BUG_ON(ret);
 317         }
 318
 319         return total_added;
 320 }
 321
 322 static int caching_kthread(void *data)
 323 {
 324         struct btrfs_block_group_cache *block_group = data;
 325         struct btrfs_fs_info *fs_info = block_group->fs_info;
 326         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 327         struct btrfs_root *extent_root = fs_info->extent_root;
 328         struct btrfs_path *path;
 329         struct extent_buffer *leaf;
 330         struct btrfs_key key;
 331         u64 total_found = 0;
 332         u64 last = 0;
 333         u32 nritems;
 334         int ret = 0;
 335
 336         path = btrfs_alloc_path();
 337         if (!path)
 338                 return -ENOMEM;
 339
 340         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 341
 342         /*
 343          * We don't want to deadlock with somebody trying to allocate a new
 344          * extent for the extent root while also trying to search the extent
 345          * root to add free space.  So we skip locking and search the commit
 346          * root, since its read-only
 347          */
 348         path->skip_locking = 1;
 349         path->search_commit_root = 1;
 350         path->reada = 1;
 351
 352         key.objectid = last;
 353         key.offset = 0;
 354         key.type = BTRFS_EXTENT_ITEM_KEY;
 355 again:
 356         mutex_lock(&caching_ctl->mutex);
 357         /* need to make sure the commit_root doesn't disappear */
 358         down_read(&fs_info->extent_commit_sem);
 359
 360         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 361         if (ret < 0)
 362                 goto err;
 363
 364         leaf = path->nodes[0];
 365         nritems = btrfs_header_nritems(leaf);
 366
 367         while (1) {
 368                 smp_mb();
 369                 if (fs_info->closing > 1) {
 370                         last = (u64)-1;
 371                         break;
 372                 }
 373
 374                 if (path->slots[0] < nritems) {
 375                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 376                 } else {
 377                         ret = find_next_key(path, 0, &key);
 378                         if (ret)
 379                                 break;
 380
 381                         if (need_resched() ||
 382                             btrfs_next_leaf(extent_root, path)) {
 383                                 caching_ctl->progress = last;
 384                                 btrfs_release_path(extent_root, path);
 385                                 up_read(&fs_info->extent_commit_sem);
 386                                 mutex_unlock(&caching_ctl->mutex);
 387                                 cond_resched();
 388                                 goto again;
 389                         }
 390                         leaf = path->nodes[0];
 391                         nritems = btrfs_header_nritems(leaf);
 392                         continue;
 393                 }
 394
 395                 if (key.objectid < block_group->key.objectid) {
 396                         path->slots[0]++;
 397                         continue;
 398                 }
 399
 400                 if (key.objectid >= block_group->key.objectid +
 401                     block_group->key.offset)
 402                         break;
 403
 404                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 405                         total_found += add_new_free_space(block_group,
 406                                                           fs_info, last,
 407                                                           key.objectid);
 408                         last = key.objectid + key.offset;
 409
 410                         if (total_found > (1024 * 1024 * 2)) {
 411                                 total_found = 0;
 412                                 wake_up(&caching_ctl->wait);
 413                         }
 414                 }
 415                 path->slots[0]++;
 416         }
 417         ret = 0;
 418
 419         total_found += add_new_free_space(block_group, fs_info, last,
 420                                           block_group->key.objectid +
 421                                           block_group->key.offset);
 422         caching_ctl->progress = (u64)-1;
 423
 424         spin_lock(&block_group->lock);
 425         block_group->caching_ctl = NULL;
 426         block_group->cached = BTRFS_CACHE_FINISHED;
 427         spin_unlock(&block_group->lock);
 428
 429 err:
 430         btrfs_free_path(path);
 431         up_read(&fs_info->extent_commit_sem);
 432
 433         free_excluded_extents(extent_root, block_group);
 434
 435         mutex_unlock(&caching_ctl->mutex);
 436         wake_up(&caching_ctl->wait);
 437
 438         put_caching_control(caching_ctl);
 439         atomic_dec(&block_group->space_info->caching_threads);
 440         btrfs_put_block_group(block_group);
 441
 442         return 0;
 443 }
 444
 445 static int cache_block_group(struct btrfs_block_group_cache *cache,
 446                              struct btrfs_trans_handle *trans,
 447                              struct btrfs_root *root,
 448                              int load_cache_only)
 449 {
 450         struct btrfs_fs_info *fs_info = cache->fs_info;
 451         struct btrfs_caching_control *caching_ctl;
 452         struct task_struct *tsk;
 453         int ret = 0;
 454
 455         smp_mb();
 456         if (cache->cached != BTRFS_CACHE_NO)
 457                 return 0;
 458
 459         /*
 460          * We can't do the read from on-disk cache during a commit since we need
 461          * to have the normal tree locking.  Also if we are currently trying to
 462          * allocate blocks for the tree root we can't do the fast caching since
 463          * we likely hold important locks.
 464          */
 465         if (trans && (!trans->transaction->in_commit) &&
 466             (root && root != root->fs_info->tree_root)) {
 467                 spin_lock(&cache->lock);
 468                 if (cache->cached != BTRFS_CACHE_NO) {
 469                         spin_unlock(&cache->lock);
 470                         return 0;
 471                 }
 472                 cache->cached = BTRFS_CACHE_STARTED;
 473                 spin_unlock(&cache->lock);
 474
 475                 ret = load_free_space_cache(fs_info, cache);
 476
 477                 spin_lock(&cache->lock);
 478                 if (ret == 1) {
 479                         cache->cached = BTRFS_CACHE_FINISHED;
 480                         cache->last_byte_to_unpin = (u64)-1;
 481                 } else {
 482                         cache->cached = BTRFS_CACHE_NO;
 483                 }
 484                 spin_unlock(&cache->lock);
 485                 if (ret == 1) {
 486                         free_excluded_extents(fs_info->extent_root, cache);
 487                         return 0;
 488                 }
 489         }
 490
 491         if (load_cache_only)
 492                 return 0;
 493
 494         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
 495         BUG_ON(!caching_ctl);
 496
 497         INIT_LIST_HEAD(&caching_ctl->list);
 498         mutex_init(&caching_ctl->mutex);
 499         init_waitqueue_head(&caching_ctl->wait);
 500         caching_ctl->block_group = cache;
 501         caching_ctl->progress = cache->key.objectid;
 502         /* one for caching kthread, one for caching block group list */
 503         atomic_set(&caching_ctl->count, 2);
 504
 505         spin_lock(&cache->lock);
 506         if (cache->cached != BTRFS_CACHE_NO) {
 507                 spin_unlock(&cache->lock);
 508                 kfree(caching_ctl);
 509                 return 0;
 510         }
 511         cache->caching_ctl = caching_ctl;
 512         cache->cached = BTRFS_CACHE_STARTED;
 513         spin_unlock(&cache->lock);
 514
 515         down_write(&fs_info->extent_commit_sem);
 516         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 517         up_write(&fs_info->extent_commit_sem);
 518
 519         atomic_inc(&cache->space_info->caching_threads);
 520         btrfs_get_block_group(cache);
 521
 522         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 523                           cache->key.objectid);
 524         if (IS_ERR(tsk)) {
 525                 ret = PTR_ERR(tsk);
 526                 printk(KERN_ERR "error running thread %d\n", ret);
 527                 BUG();
 528         }
 529
 530         return ret;
 531 }
 532
 533 /*
 534  * return the block group that starts at or after bytenr
 535  */
 536 static struct btrfs_block_group_cache *
 537 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 538 {
 539         struct btrfs_block_group_cache *cache;
 540
 541         cache = block_group_cache_tree_search(info, bytenr, 0);
 542
 543         return cache;
 544 }
 545
 546 /*
 547  * return the block group that contains the given bytenr
 548  */
 549 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 550                                                  struct btrfs_fs_info *info,
 551                                                  u64 bytenr)
 552 {
 553         struct btrfs_block_group_cache *cache;
 554
 555         cache = block_group_cache_tree_search(info, bytenr, 1);
 556
 557         return cache;
 558 }
 559
 560 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 561                                                   u64 flags)
 562 {
 563         struct list_head *head = &info->space_info;
 564         struct btrfs_space_info *found;
 565
 566         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 567                  BTRFS_BLOCK_GROUP_METADATA;
 568
 569         rcu_read_lock();
 570         list_for_each_entry_rcu(found, head, list) {
 571                 if (found->flags & flags) {
 572                         rcu_read_unlock();
 573                         return found;
 574                 }
 575         }
 576         rcu_read_unlock();
 577         return NULL;
 578 }
 579
 580 /*
 581  * after adding space to the filesystem, we need to clear the full flags
 582  * on all the space infos.
 583  */
 584 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 585 {
 586         struct list_head *head = &info->space_info;
 587         struct btrfs_space_info *found;
 588
 589         rcu_read_lock();
 590         list_for_each_entry_rcu(found, head, list)
 591                 found->full = 0;
 592         rcu_read_unlock();
 593 }
 594
 595 static u64 div_factor(u64 num, int factor)
 596 {
 597         if (factor == 10)
 598                 return num;
 599         num *= factor;
 600         do_div(num, 10);
 601         return num;
 602 }
 603
 604 static u64 div_factor_fine(u64 num, int factor)
 605 {
 606         if (factor == 100)
 607                 return num;
 608         num *= factor;
 609         do_div(num, 100);
 610         return num;
 611 }
 612
 613 u64 btrfs_find_block_group(struct btrfs_root *root,
 614                            u64 search_start, u64 search_hint, int owner)
 615 {
 616         struct btrfs_block_group_cache *cache;
 617         u64 used;
 618         u64 last = max(search_hint, search_start);
 619         u64 group_start = 0;
 620         int full_search = 0;
 621         int factor = 9;
 622         int wrapped = 0;
 623 again:
 624         while (1) {
 625                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 626                 if (!cache)
 627                         break;
 628
 629                 spin_lock(&cache->lock);
 630                 last = cache->key.objectid + cache->key.offset;
 631                 used = btrfs_block_group_used(&cache->item);
 632
 633                 if ((full_search || !cache->ro) &&
 634                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 635                         if (used + cache->pinned + cache->reserved <
 636                             div_factor(cache->key.offset, factor)) {
 637                                 group_start = cache->key.objectid;
 638                                 spin_unlock(&cache->lock);
 639                                 btrfs_put_block_group(cache);
 640                                 goto found;
 641                         }
 642                 }
 643                 spin_unlock(&cache->lock);
 644                 btrfs_put_block_group(cache);
 645                 cond_resched();
 646         }
 647         if (!wrapped) {
 648                 last = search_start;
 649                 wrapped = 1;
 650                 goto again;
 651         }
 652         if (!full_search && factor < 10) {
 653                 last = search_start;
 654                 full_search = 1;
 655                 factor = 10;
 656                 goto again;
 657         }
 658 found:
 659         return group_start;
 660 }
 661
 662 /* simple helper to search for an existing extent at a given offset */
 663 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 664 {
 665         int ret;
 666         struct btrfs_key key;
 667         struct btrfs_path *path;
 668
 669         path = btrfs_alloc_path();
 670         BUG_ON(!path);
 671         key.objectid = start;
 672         key.offset = len;
 673         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 674         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 675                                 0, 0);
 676         btrfs_free_path(path);
 677         return ret;
 678 }
 679
 680 /*
 681  * helper function to lookup reference count and flags of extent.
 682  *
 683  * the head node for delayed ref is used to store the sum of all the
 684  * reference count modifications queued up in the rbtree. the head
 685  * node may also store the extent flags to set. This way you can check
 686  * to see what the reference count and extent flags would be if all of
 687  * the delayed refs are not processed.
 688  */
 689 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 690                              struct btrfs_root *root, u64 bytenr,
 691                              u64 num_bytes, u64 *refs, u64 *flags)
 692 {
 693         struct btrfs_delayed_ref_head *head;
 694         struct btrfs_delayed_ref_root *delayed_refs;
 695         struct btrfs_path *path;
 696         struct btrfs_extent_item *ei;
 697         struct extent_buffer *leaf;
 698         struct btrfs_key key;
 699         u32 item_size;
 700         u64 num_refs;
 701         u64 extent_flags;
 702         int ret;
 703
 704         path = btrfs_alloc_path();
 705         if (!path)
 706                 return -ENOMEM;
 707
 708         key.objectid = bytenr;
 709         key.type = BTRFS_EXTENT_ITEM_KEY;
 710         key.offset = num_bytes;
 711         if (!trans) {
 712                 path->skip_locking = 1;
 713                 path->search_commit_root = 1;
 714         }
 715 again:
 716         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 717                                 &key, path, 0, 0);
 718         if (ret < 0)
 719                 goto out_free;
 720
 721         if (ret == 0) {
 722                 leaf = path->nodes[0];
 723                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 724                 if (item_size >= sizeof(*ei)) {
 725                         ei = btrfs_item_ptr(leaf, path->slots[0],
 726                                             struct btrfs_extent_item);
 727                         num_refs = btrfs_extent_refs(leaf, ei);
 728                         extent_flags = btrfs_extent_flags(leaf, ei);
 729                 } else {
 730 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 731                         struct btrfs_extent_item_v0 *ei0;
 732                         BUG_ON(item_size != sizeof(*ei0));
 733                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 734                                              struct btrfs_extent_item_v0);
 735                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 736                         /* FIXME: this isn't correct for data */
 737                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 738 #else
 739                         BUG();
 740 #endif
 741                 }
 742                 BUG_ON(num_refs == 0);
 743         } else {
 744                 num_refs = 0;
 745                 extent_flags = 0;
 746                 ret = 0;
 747         }
 748
 749         if (!trans)
 750                 goto out;
 751
 752         delayed_refs = &trans->transaction->delayed_refs;
 753         spin_lock(&delayed_refs->lock);
 754         head = btrfs_find_delayed_ref_head(trans, bytenr);
 755         if (head) {
 756                 if (!mutex_trylock(&head->mutex)) {
 757                         atomic_inc(&head->node.refs);
 758                         spin_unlock(&delayed_refs->lock);
 759
 760                         btrfs_release_path(root->fs_info->extent_root, path);
 761
 762                         mutex_lock(&head->mutex);
 763                         mutex_unlock(&head->mutex);
 764                         btrfs_put_delayed_ref(&head->node);
 765                         goto again;
 766                 }
 767                 if (head->extent_op && head->extent_op->update_flags)
 768                         extent_flags |= head->extent_op->flags_to_set;
 769                 else
 770                         BUG_ON(num_refs == 0);
 771
 772                 num_refs += head->node.ref_mod;
 773                 mutex_unlock(&head->mutex);
 774         }
 775         spin_unlock(&delayed_refs->lock);
 776 out:
 777         WARN_ON(num_refs == 0);
 778         if (refs)
 779                 *refs = num_refs;
 780         if (flags)
 781                 *flags = extent_flags;
 782 out_free:
 783         btrfs_free_path(path);
 784         return ret;
 785 }
 786
 787 /*
 788  * Back reference rules.  Back refs have three main goals:
 789  *
 790  * 1) differentiate between all holders of references to an extent so that
 791  *    when a reference is dropped we can make sure it was a valid reference
 792  *    before freeing the extent.
 793  *
 794  * 2) Provide enough information to quickly find the holders of an extent
 795  *    if we notice a given block is corrupted or bad.
 796  *
 797  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 798  *    maintenance.  This is actually the same as #2, but with a slightly
 799  *    different use case.
 800  *
 801  * There are two kinds of back refs. The implicit back refs is optimized
 802  * for pointers in non-shared tree blocks. For a given pointer in a block,
 803  * back refs of this kind provide information about the block's owner tree
 804  * and the pointer's key. These information allow us to find the block by
 805  * b-tree searching. The full back refs is for pointers in tree blocks not
 806  * referenced by their owner trees. The location of tree block is recorded
 807  * in the back refs. Actually the full back refs is generic, and can be
 808  * used in all cases the implicit back refs is used. The major shortcoming
 809  * of the full back refs is its overhead. Every time a tree block gets
 810  * COWed, we have to update back refs entry for all pointers in it.
 811  *
 812  * For a newly allocated tree block, we use implicit back refs for
 813  * pointers in it. This means most tree related operations only involve
 814  * implicit back refs. For a tree block created in old transaction, the
 815  * only way to drop a reference to it is COW it. So we can detect the
 816  * event that tree block loses its owner tree's reference and do the
 817  * back refs conversion.
 818  *
 819  * When a tree block is COW'd through a tree, there are four cases:
 820  *
 821  * The reference count of the block is one and the tree is the block's
 822  * owner tree. Nothing to do in this case.
 823  *
 824  * The reference count of the block is one and the tree is not the
 825  * block's owner tree. In this case, full back refs is used for pointers
 826  * in the block. Remove these full back refs, add implicit back refs for
 827  * every pointers in the new block.
 828  *
 829  * The reference count of the block is greater than one and the tree is
 830  * the block's owner tree. In this case, implicit back refs is used for
 831  * pointers in the block. Add full back refs for every pointers in the
 832  * block, increase lower level extents' reference counts. The original
 833  * implicit back refs are entailed to the new block.
 834  *
 835  * The reference count of the block is greater than one and the tree is
 836  * not the block's owner tree. Add implicit back refs for every pointer in
 837  * the new block, increase lower level extents' reference count.
 838  *
 839  * Back Reference Key composing:
 840  *
 841  * The key objectid corresponds to the first byte in the extent,
 842  * The key type is used to differentiate between types of back refs.
 843  * There are different meanings of the key offset for different types
 844  * of back refs.
 845  *
 846  * File extents can be referenced by:
 847  *
 848  * - multiple snapshots, subvolumes, or different generations in one subvol
 849  * - different files inside a single subvolume
 850  * - different offsets inside a file (bookend extents in file.c)
 851  *
 852  * The extent ref structure for the implicit back refs has fields for:
 853  *
 854  * - Objectid of the subvolume root
 855  * - objectid of the file holding the reference
 856  * - original offset in the file
 857  * - how many bookend extents
 858  *
 859  * The key offset for the implicit back refs is hash of the first
 860  * three fields.
 861  *
 862  * The extent ref structure for the full back refs has field for:
 863  *
 864  * - number of pointers in the tree leaf
 865  *
 866  * The key offset for the implicit back refs is the first byte of
 867  * the tree leaf
 868  *
 869  * When a file extent is allocated, The implicit back refs is used.
 870  * the fields are filled in:
 871  *
 872  *     (root_key.objectid, inode objectid, offset in file, 1)
 873  *
 874  * When a file extent is removed file truncation, we find the
 875  * corresponding implicit back refs and check the following fields:
 876  *
 877  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 878  *
 879  * Btree extents can be referenced by:
 880  *
 881  * - Different subvolumes
 882  *
 883  * Both the implicit back refs and the full back refs for tree blocks
 884  * only consist of key. The key offset for the implicit back refs is
 885  * objectid of block's owner tree. The key offset for the full back refs
 886  * is the first byte of parent block.
 887  *
 888  * When implicit back refs is used, information about the lowest key and
 889  * level of the tree block are required. These information are stored in
 890  * tree block info structure.
 891  */
 892
 893 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 894 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 895                                   struct btrfs_root *root,
 896                                   struct btrfs_path *path,
 897                                   u64 owner, u32 extra_size)
 898 {
 899         struct btrfs_extent_item *item;
 900         struct btrfs_extent_item_v0 *ei0;
 901         struct btrfs_extent_ref_v0 *ref0;
 902         struct btrfs_tree_block_info *bi;
 903         struct extent_buffer *leaf;
 904         struct btrfs_key key;
 905         struct btrfs_key found_key;
 906         u32 new_size = sizeof(*item);
 907         u64 refs;
 908         int ret;
 909
 910         leaf = path->nodes[0];
 911         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 912
 913         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 914         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 915                              struct btrfs_extent_item_v0);
 916         refs = btrfs_extent_refs_v0(leaf, ei0);
 917
 918         if (owner == (u64)-1) {
 919                 while (1) {
 920                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 921                                 ret = btrfs_next_leaf(root, path);
 922                                 if (ret < 0)
 923                                         return ret;
 924                                 BUG_ON(ret > 0);
 925                                 leaf = path->nodes[0];
 926                         }
 927                         btrfs_item_key_to_cpu(leaf, &found_key,
 928                                               path->slots[0]);
 929                         BUG_ON(key.objectid != found_key.objectid);
 930                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 931                                 path->slots[0]++;
 932                                 continue;
 933                         }
 934                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 935                                               struct btrfs_extent_ref_v0);
 936                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 937                         break;
 938                 }
 939         }
 940         btrfs_release_path(root, path);
 941
 942         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 943                 new_size += sizeof(*bi);
 944
 945         new_size -= sizeof(*ei0);
 946         ret = btrfs_search_slot(trans, root, &key, path,
 947                                 new_size + extra_size, 1);
 948         if (ret < 0)
 949                 return ret;
 950         BUG_ON(ret);
 951
 952         ret = btrfs_extend_item(trans, root, path, new_size);
 953         BUG_ON(ret);
 954
 955         leaf = path->nodes[0];
 956         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 957         btrfs_set_extent_refs(leaf, item, refs);
 958         /* FIXME: get real generation */
 959         btrfs_set_extent_generation(leaf, item, 0);
 960         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 961                 btrfs_set_extent_flags(leaf, item,
 962                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 963                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 964                 bi = (struct btrfs_tree_block_info *)(item + 1);
 965                 /* FIXME: get first key of the block */
 966                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 967                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 968         } else {
 969                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 970         }
 971         btrfs_mark_buffer_dirty(leaf);
 972         return 0;
 973 }
 974 #endif
 975
 976 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 977 {
 978         u32 high_crc = ~(u32)0;
 979         u32 low_crc = ~(u32)0;
 980         __le64 lenum;
 981
 982         lenum = cpu_to_le64(root_objectid);
 983         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 984         lenum = cpu_to_le64(owner);
 985         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 986         lenum = cpu_to_le64(offset);
 987         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 988
 989         return ((u64)high_crc << 31) ^ (u64)low_crc;
 990 }
 991
 992 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 993                                      struct btrfs_extent_data_ref *ref)
 994 {
 995         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 996                                     btrfs_extent_data_ref_objectid(leaf, ref),
 997                                     btrfs_extent_data_ref_offset(leaf, ref));
 998 }
 999
1000 static int match_extent_data_ref(struct extent_buffer *leaf,
1001                                  struct btrfs_extent_data_ref *ref,
1002                                  u64 root_objectid, u64 owner, u64 offset)
1003 {
1004         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1005             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1006             btrfs_extent_data_ref_offset(leaf, ref) != offset)
1007                 return 0;
1008         return 1;
1009 }
1010
1011 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1012                                            struct btrfs_root *root,
1013                                            struct btrfs_path *path,
1014                                            u64 bytenr, u64 parent,
1015                                            u64 root_objectid,
1016                                            u64 owner, u64 offset)
1017 {
1018         struct btrfs_key key;
1019         struct btrfs_extent_data_ref *ref;
1020         struct extent_buffer *leaf;
1021         u32 nritems;
1022         int ret;
1023         int recow;
1024         int err = -ENOENT;
1025
1026         key.objectid = bytenr;
1027         if (parent) {
1028                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1029                 key.offset = parent;
1030         } else {
1031                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1032                 key.offset = hash_extent_data_ref(root_objectid,
1033                                                   owner, offset);
1034         }
1035 again:
1036         recow = 0;
1037         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1038         if (ret < 0) {
1039                 err = ret;
1040                 goto fail;
1041         }
1042
1043         if (parent) {
1044                 if (!ret)
1045                         return 0;
1046 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1047                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1048                 btrfs_release_path(root, path);
1049                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1050                 if (ret < 0) {
1051                         err = ret;
1052                         goto fail;
1053                 }
1054                 if (!ret)
1055                         return 0;
1056 #endif
1057                 goto fail;
1058         }
1059
1060         leaf = path->nodes[0];
1061         nritems = btrfs_header_nritems(leaf);
1062         while (1) {
1063                 if (path->slots[0] >= nritems) {
1064                         ret = btrfs_next_leaf(root, path);
1065                         if (ret < 0)
1066                                 err = ret;
1067                         if (ret)
1068                                 goto fail;
1069
1070                         leaf = path->nodes[0];
1071                         nritems = btrfs_header_nritems(leaf);
1072                         recow = 1;
1073                 }
1074
1075                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1076                 if (key.objectid != bytenr ||
1077                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1078                         goto fail;
1079
1080                 ref = btrfs_item_ptr(leaf, path->slots[0],
1081                                      struct btrfs_extent_data_ref);
1082
1083                 if (match_extent_data_ref(leaf, ref, root_objectid,
1084                                           owner, offset)) {
1085                         if (recow) {
1086                                 btrfs_release_path(root, path);
1087                                 goto again;
1088                         }
1089                         err = 0;
1090                         break;
1091                 }
1092                 path->slots[0]++;
1093         }
1094 fail:
1095         return err;
1096 }
1097
1098 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1099                                            struct btrfs_root *root,
1100                                            struct btrfs_path *path,
1101                                            u64 bytenr, u64 parent,
1102                                            u64 root_objectid, u64 owner,
1103                                            u64 offset, int refs_to_add)
1104 {
1105         struct btrfs_key key;
1106         struct extent_buffer *leaf;
1107         u32 size;
1108         u32 num_refs;
1109         int ret;
1110
1111         key.objectid = bytenr;
1112         if (parent) {
1113                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1114                 key.offset = parent;
1115                 size = sizeof(struct btrfs_shared_data_ref);
1116         } else {
1117                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1118                 key.offset = hash_extent_data_ref(root_objectid,
1119                                                   owner, offset);
1120                 size = sizeof(struct btrfs_extent_data_ref);
1121         }
1122
1123         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1124         if (ret && ret != -EEXIST)
1125                 goto fail;
1126
1127         leaf = path->nodes[0];
1128         if (parent) {
1129                 struct btrfs_shared_data_ref *ref;
1130                 ref = btrfs_item_ptr(leaf, path->slots[0],
1131                                      struct btrfs_shared_data_ref);
1132                 if (ret == 0) {
1133                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1134                 } else {
1135                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1136                         num_refs += refs_to_add;
1137                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1138                 }
1139         } else {
1140                 struct btrfs_extent_data_ref *ref;
1141                 while (ret == -EEXIST) {
1142                         ref = btrfs_item_ptr(leaf, path->slots[0],
1143                                              struct btrfs_extent_data_ref);
1144                         if (match_extent_data_ref(leaf, ref, root_objectid,
1145                                                   owner, offset))
1146                                 break;
1147                         btrfs_release_path(root, path);
1148                         key.offset++;
1149                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1150                                                       size);
1151                         if (ret && ret != -EEXIST)
1152                                 goto fail;
1153
1154                         leaf = path->nodes[0];
1155                 }
1156                 ref = btrfs_item_ptr(leaf, path->slots[0],
1157                                      struct btrfs_extent_data_ref);
1158                 if (ret == 0) {
1159                         btrfs_set_extent_data_ref_root(leaf, ref,
1160                                                        root_objectid);
1161                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1162                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1163                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1164                 } else {
1165                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1166                         num_refs += refs_to_add;
1167                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1168                 }
1169         }
1170         btrfs_mark_buffer_dirty(leaf);
1171         ret = 0;
1172 fail:
1173         btrfs_release_path(root, path);
1174         return ret;
1175 }
1176
1177 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1178                                            struct btrfs_root *root,
1179                                            struct btrfs_path *path,
1180                                            int refs_to_drop)
1181 {
1182         struct btrfs_key key;
1183         struct btrfs_extent_data_ref *ref1 = NULL;
1184         struct btrfs_shared_data_ref *ref2 = NULL;
1185         struct extent_buffer *leaf;
1186         u32 num_refs = 0;
1187         int ret = 0;
1188
1189         leaf = path->nodes[0];
1190         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1191
1192         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1193                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1194                                       struct btrfs_extent_data_ref);
1195                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1196         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1197                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1198                                       struct btrfs_shared_data_ref);
1199                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1200 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1201         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1202                 struct btrfs_extent_ref_v0 *ref0;
1203                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1204                                       struct btrfs_extent_ref_v0);
1205                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1206 #endif
1207         } else {
1208                 BUG();
1209         }
1210
1211         BUG_ON(num_refs < refs_to_drop);
1212         num_refs -= refs_to_drop;
1213
1214         if (num_refs == 0) {
1215                 ret = btrfs_del_item(trans, root, path);
1216         } else {
1217                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1218                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1219                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1220                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1221 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1222                 else {
1223                         struct btrfs_extent_ref_v0 *ref0;
1224                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1225                                         struct btrfs_extent_ref_v0);
1226                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1227                 }
1228 #endif
1229                 btrfs_mark_buffer_dirty(leaf);
1230         }
1231         return ret;
1232 }
1233
1234 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1235                                           struct btrfs_path *path,
1236                                           struct btrfs_extent_inline_ref *iref)
1237 {
1238         struct btrfs_key key;
1239         struct extent_buffer *leaf;
1240         struct btrfs_extent_data_ref *ref1;
1241         struct btrfs_shared_data_ref *ref2;
1242         u32 num_refs = 0;
1243
1244         leaf = path->nodes[0];
1245         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1246         if (iref) {
1247                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1248                     BTRFS_EXTENT_DATA_REF_KEY) {
1249                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1250                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1251                 } else {
1252                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1253                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1254                 }
1255         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1256                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1257                                       struct btrfs_extent_data_ref);
1258                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1259         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1260                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1261                                       struct btrfs_shared_data_ref);
1262                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1263 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1264         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1265                 struct btrfs_extent_ref_v0 *ref0;
1266                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1267                                       struct btrfs_extent_ref_v0);
1268                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1269 #endif
1270         } else {
1271                 WARN_ON(1);
1272         }
1273         return num_refs;
1274 }
1275
1276 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1277                                           struct btrfs_root *root,
1278                                           struct btrfs_path *path,
1279                                           u64 bytenr, u64 parent,
1280                                           u64 root_objectid)
1281 {
1282         struct btrfs_key key;
1283         int ret;
1284
1285         key.objectid = bytenr;
1286         if (parent) {
1287                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1288                 key.offset = parent;
1289         } else {
1290                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1291                 key.offset = root_objectid;
1292         }
1293
1294         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1295         if (ret > 0)
1296                 ret = -ENOENT;
1297 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1298         if (ret == -ENOENT && parent) {
1299                 btrfs_release_path(root, path);
1300                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1301                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1302                 if (ret > 0)
1303                         ret = -ENOENT;
1304         }
1305 #endif
1306         return ret;
1307 }
1308
1309 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1310                                           struct btrfs_root *root,
1311                                           struct btrfs_path *path,
1312                                           u64 bytenr, u64 parent,
1313                                           u64 root_objectid)
1314 {
1315         struct btrfs_key key;
1316         int ret;
1317
1318         key.objectid = bytenr;
1319         if (parent) {
1320                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1321                 key.offset = parent;
1322         } else {
1323                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1324                 key.offset = root_objectid;
1325         }
1326
1327         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1328         btrfs_release_path(root, path);
1329         return ret;
1330 }
1331
1332 static inline int extent_ref_type(u64 parent, u64 owner)
1333 {
1334         int type;
1335         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1336                 if (parent > 0)
1337                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1338                 else
1339                         type = BTRFS_TREE_BLOCK_REF_KEY;
1340         } else {
1341                 if (parent > 0)
1342                         type = BTRFS_SHARED_DATA_REF_KEY;
1343                 else
1344                         type = BTRFS_EXTENT_DATA_REF_KEY;
1345         }
1346         return type;
1347 }
1348
1349 static int find_next_key(struct btrfs_path *path, int level,
1350                          struct btrfs_key *key)
1351
1352 {
1353         for (; level < BTRFS_MAX_LEVEL; level++) {
1354                 if (!path->nodes[level])
1355                         break;
1356                 if (path->slots[level] + 1 >=
1357                     btrfs_header_nritems(path->nodes[level]))
1358                         continue;
1359                 if (level == 0)
1360                         btrfs_item_key_to_cpu(path->nodes[level], key,
1361                                               path->slots[level] + 1);
1362                 else
1363                         btrfs_node_key_to_cpu(path->nodes[level], key,
1364                                               path->slots[level] + 1);
1365                 return 0;
1366         }
1367         return 1;
1368 }
1369
1370 /*
1371  * look for inline back ref. if back ref is found, *ref_ret is set
1372  * to the address of inline back ref, and 0 is returned.
1373  *
1374  * if back ref isn't found, *ref_ret is set to the address where it
1375  * should be inserted, and -ENOENT is returned.
1376  *
1377  * if insert is true and there are too many inline back refs, the path
1378  * points to the extent item, and -EAGAIN is returned.
1379  *
1380  * NOTE: inline back refs are ordered in the same way that back ref
1381  *       items in the tree are ordered.
1382  */
1383 static noinline_for_stack
1384 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1385                                  struct btrfs_root *root,
1386                                  struct btrfs_path *path,
1387                                  struct btrfs_extent_inline_ref **ref_ret,
1388                                  u64 bytenr, u64 num_bytes,
1389                                  u64 parent, u64 root_objectid,
1390                                  u64 owner, u64 offset, int insert)
1391 {
1392         struct btrfs_key key;
1393         struct extent_buffer *leaf;
1394         struct btrfs_extent_item *ei;
1395         struct btrfs_extent_inline_ref *iref;
1396         u64 flags;
1397         u64 item_size;
1398         unsigned long ptr;
1399         unsigned long end;
1400         int extra_size;
1401         int type;
1402         int want;
1403         int ret;
1404         int err = 0;
1405
1406         key.objectid = bytenr;
1407         key.type = BTRFS_EXTENT_ITEM_KEY;
1408         key.offset = num_bytes;
1409
1410         want = extent_ref_type(parent, owner);
1411         if (insert) {
1412                 extra_size = btrfs_extent_inline_ref_size(want);
1413                 path->keep_locks = 1;
1414         } else
1415                 extra_size = -1;
1416         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1417         if (ret < 0) {
1418                 err = ret;
1419                 goto out;
1420         }
1421         BUG_ON(ret);
1422
1423         leaf = path->nodes[0];
1424         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1425 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1426         if (item_size < sizeof(*ei)) {
1427                 if (!insert) {
1428                         err = -ENOENT;
1429                         goto out;
1430                 }
1431                 ret = convert_extent_item_v0(trans, root, path, owner,
1432                                              extra_size);
1433                 if (ret < 0) {
1434                         err = ret;
1435                         goto out;
1436                 }
1437                 leaf = path->nodes[0];
1438                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1439         }
1440 #endif
1441         BUG_ON(item_size < sizeof(*ei));
1442
1443         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1444         flags = btrfs_extent_flags(leaf, ei);
1445
1446         ptr = (unsigned long)(ei + 1);
1447         end = (unsigned long)ei + item_size;
1448
1449         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1450                 ptr += sizeof(struct btrfs_tree_block_info);
1451                 BUG_ON(ptr > end);
1452         } else {
1453                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1454         }
1455
1456         err = -ENOENT;
1457         while (1) {
1458                 if (ptr >= end) {
1459                         WARN_ON(ptr > end);
1460                         break;
1461                 }
1462                 iref = (struct btrfs_extent_inline_ref *)ptr;
1463                 type = btrfs_extent_inline_ref_type(leaf, iref);
1464                 if (want < type)
1465                         break;
1466                 if (want > type) {
1467                         ptr += btrfs_extent_inline_ref_size(type);
1468                         continue;
1469                 }
1470
1471                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1472                         struct btrfs_extent_data_ref *dref;
1473                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1474                         if (match_extent_data_ref(leaf, dref, root_objectid,
1475                                                   owner, offset)) {
1476                                 err = 0;
1477                                 break;
1478                         }
1479                         if (hash_extent_data_ref_item(leaf, dref) <
1480                             hash_extent_data_ref(root_objectid, owner, offset))
1481                                 break;
1482                 } else {
1483                         u64 ref_offset;
1484                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1485                         if (parent > 0) {
1486                                 if (parent == ref_offset) {
1487                                         err = 0;
1488                                         break;
1489                                 }
1490                                 if (ref_offset < parent)
1491                                         break;
1492                         } else {
1493                                 if (root_objectid == ref_offset) {
1494                                         err = 0;
1495                                         break;
1496                                 }
1497                                 if (ref_offset < root_objectid)
1498                                         break;
1499                         }
1500                 }
1501                 ptr += btrfs_extent_inline_ref_size(type);
1502         }
1503         if (err == -ENOENT && insert) {
1504                 if (item_size + extra_size >=
1505                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1506                         err = -EAGAIN;
1507                         goto out;
1508                 }
1509                 /*
1510                  * To add new inline back ref, we have to make sure
1511                  * there is no corresponding back ref item.
1512                  * For simplicity, we just do not add new inline back
1513                  * ref if there is any kind of item for this block
1514                  */
1515                 if (find_next_key(path, 0, &key) == 0 &&
1516                     key.objectid == bytenr &&
1517                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1518                         err = -EAGAIN;
1519                         goto out;
1520                 }
1521         }
1522         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1523 out:
1524         if (insert) {
1525                 path->keep_locks = 0;
1526                 btrfs_unlock_up_safe(path, 1);
1527         }
1528         return err;
1529 }
1530
1531 /*
1532  * helper to add new inline back ref
1533  */
1534 static noinline_for_stack
1535 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1536                                 struct btrfs_root *root,
1537                                 struct btrfs_path *path,
1538                                 struct btrfs_extent_inline_ref *iref,
1539                                 u64 parent, u64 root_objectid,
1540                                 u64 owner, u64 offset, int refs_to_add,
1541                                 struct btrfs_delayed_extent_op *extent_op)
1542 {
1543         struct extent_buffer *leaf;
1544         struct btrfs_extent_item *ei;
1545         unsigned long ptr;
1546         unsigned long end;
1547         unsigned long item_offset;
1548         u64 refs;
1549         int size;
1550         int type;
1551         int ret;
1552
1553         leaf = path->nodes[0];
1554         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1555         item_offset = (unsigned long)iref - (unsigned long)ei;
1556
1557         type = extent_ref_type(parent, owner);
1558         size = btrfs_extent_inline_ref_size(type);
1559
1560         ret = btrfs_extend_item(trans, root, path, size);
1561         BUG_ON(ret);
1562
1563         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1564         refs = btrfs_extent_refs(leaf, ei);
1565         refs += refs_to_add;
1566         btrfs_set_extent_refs(leaf, ei, refs);
1567         if (extent_op)
1568                 __run_delayed_extent_op(extent_op, leaf, ei);
1569
1570         ptr = (unsigned long)ei + item_offset;
1571         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1572         if (ptr < end - size)
1573                 memmove_extent_buffer(leaf, ptr + size, ptr,
1574                                       end - size - ptr);
1575
1576         iref = (struct btrfs_extent_inline_ref *)ptr;
1577         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1578         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1579                 struct btrfs_extent_data_ref *dref;
1580                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1581                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1582                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1583                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1584                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1585         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1586                 struct btrfs_shared_data_ref *sref;
1587                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1588                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1589                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1590         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1591                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1592         } else {
1593                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1594         }
1595         btrfs_mark_buffer_dirty(leaf);
1596         return 0;
1597 }
1598
1599 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1600                                  struct btrfs_root *root,
1601                                  struct btrfs_path *path,
1602                                  struct btrfs_extent_inline_ref **ref_ret,
1603                                  u64 bytenr, u64 num_bytes, u64 parent,
1604                                  u64 root_objectid, u64 owner, u64 offset)
1605 {
1606         int ret;
1607
1608         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1609                                            bytenr, num_bytes, parent,
1610                                            root_objectid, owner, offset, 0);
1611         if (ret != -ENOENT)
1612                 return ret;
1613
1614         btrfs_release_path(root, path);
1615         *ref_ret = NULL;
1616
1617         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1618                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1619                                             root_objectid);
1620         } else {
1621                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1622                                              root_objectid, owner, offset);
1623         }
1624         return ret;
1625 }
1626
1627 /*
1628  * helper to update/remove inline back ref
1629  */
1630 static noinline_for_stack
1631 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1632                                  struct btrfs_root *root,
1633                                  struct btrfs_path *path,
1634                                  struct btrfs_extent_inline_ref *iref,
1635                                  int refs_to_mod,
1636                                  struct btrfs_delayed_extent_op *extent_op)
1637 {
1638         struct extent_buffer *leaf;
1639         struct btrfs_extent_item *ei;
1640         struct btrfs_extent_data_ref *dref = NULL;
1641         struct btrfs_shared_data_ref *sref = NULL;
1642         unsigned long ptr;
1643         unsigned long end;
1644         u32 item_size;
1645         int size;
1646         int type;
1647         int ret;
1648         u64 refs;
1649
1650         leaf = path->nodes[0];
1651         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1652         refs = btrfs_extent_refs(leaf, ei);
1653         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1654         refs += refs_to_mod;
1655         btrfs_set_extent_refs(leaf, ei, refs);
1656         if (extent_op)
1657                 __run_delayed_extent_op(extent_op, leaf, ei);
1658
1659         type = btrfs_extent_inline_ref_type(leaf, iref);
1660
1661         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1662                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1663                 refs = btrfs_extent_data_ref_count(leaf, dref);
1664         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1665                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1666                 refs = btrfs_shared_data_ref_count(leaf, sref);
1667         } else {
1668                 refs = 1;
1669                 BUG_ON(refs_to_mod != -1);
1670         }
1671
1672         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1673         refs += refs_to_mod;
1674
1675         if (refs > 0) {
1676                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1677                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1678                 else
1679                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1680         } else {
1681                 size =  btrfs_extent_inline_ref_size(type);
1682                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1683                 ptr = (unsigned long)iref;
1684                 end = (unsigned long)ei + item_size;
1685                 if (ptr + size < end)
1686                         memmove_extent_buffer(leaf, ptr, ptr + size,
1687                                               end - ptr - size);
1688                 item_size -= size;
1689                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1690                 BUG_ON(ret);
1691         }
1692         btrfs_mark_buffer_dirty(leaf);
1693         return 0;
1694 }
1695
1696 static noinline_for_stack
1697 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1698                                  struct btrfs_root *root,
1699                                  struct btrfs_path *path,
1700                                  u64 bytenr, u64 num_bytes, u64 parent,
1701                                  u64 root_objectid, u64 owner,
1702                                  u64 offset, int refs_to_add,
1703                                  struct btrfs_delayed_extent_op *extent_op)
1704 {
1705         struct btrfs_extent_inline_ref *iref;
1706         int ret;
1707
1708         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1709                                            bytenr, num_bytes, parent,
1710                                            root_objectid, owner, offset, 1);
1711         if (ret == 0) {
1712                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1713                 ret = update_inline_extent_backref(trans, root, path, iref,
1714                                                    refs_to_add, extent_op);
1715         } else if (ret == -ENOENT) {
1716                 ret = setup_inline_extent_backref(trans, root, path, iref,
1717                                                   parent, root_objectid,
1718                                                   owner, offset, refs_to_add,
1719                                                   extent_op);
1720         }
1721         return ret;
1722 }
1723
1724 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1725                                  struct btrfs_root *root,
1726                                  struct btrfs_path *path,
1727                                  u64 bytenr, u64 parent, u64 root_objectid,
1728                                  u64 owner, u64 offset, int refs_to_add)
1729 {
1730         int ret;
1731         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1732                 BUG_ON(refs_to_add != 1);
1733                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1734                                             parent, root_objectid);
1735         } else {
1736                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1737                                              parent, root_objectid,
1738                                              owner, offset, refs_to_add);
1739         }
1740         return ret;
1741 }
1742
1743 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1744                                  struct btrfs_root *root,
1745                                  struct btrfs_path *path,
1746                                  struct btrfs_extent_inline_ref *iref,
1747                                  int refs_to_drop, int is_data)
1748 {
1749         int ret;
1750
1751         BUG_ON(!is_data && refs_to_drop != 1);
1752         if (iref) {
1753                 ret = update_inline_extent_backref(trans, root, path, iref,
1754                                                    -refs_to_drop, NULL);
1755         } else if (is_data) {
1756                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1757         } else {
1758                 ret = btrfs_del_item(trans, root, path);
1759         }
1760         return ret;
1761 }
1762
1763 static int btrfs_issue_discard(struct block_device *bdev,
1764                                 u64 start, u64 len)
1765 {
1766         return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1767 }
1768
1769 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1770                                 u64 num_bytes, u64 *actual_bytes)
1771 {
1772         int ret;
1773         u64 discarded_bytes = 0;
1774         struct btrfs_multi_bio *multi = NULL;
1775
1776
1777         /* Tell the block device(s) that the sectors can be discarded */
1778         ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
1779                               bytenr, &num_bytes, &multi, 0);
1780         if (!ret) {
1781                 struct btrfs_bio_stripe *stripe = multi->stripes;
1782                 int i;
1783
1784
1785                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1786                         ret = btrfs_issue_discard(stripe->dev->bdev,
1787                                                   stripe->physical,
1788                                                   stripe->length);
1789                         if (!ret)
1790                                 discarded_bytes += stripe->length;
1791                         else if (ret != -EOPNOTSUPP)
1792                                 break;
1793                 }
1794                 kfree(multi);
1795         }
1796         if (discarded_bytes && ret == -EOPNOTSUPP)
1797                 ret = 0;
1798
1799         if (actual_bytes)
1800                 *actual_bytes = discarded_bytes;
1801
1802
1803         return ret;
1804 }
1805
1806 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1807                          struct btrfs_root *root,
1808                          u64 bytenr, u64 num_bytes, u64 parent,
1809                          u64 root_objectid, u64 owner, u64 offset)
1810 {
1811         int ret;
1812         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1813                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1814
1815         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1816                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1817                                         parent, root_objectid, (int)owner,
1818                                         BTRFS_ADD_DELAYED_REF, NULL);
1819         } else {
1820                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1821                                         parent, root_objectid, owner, offset,
1822                                         BTRFS_ADD_DELAYED_REF, NULL);
1823         }
1824         return ret;
1825 }
1826
1827 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1828                                   struct btrfs_root *root,
1829                                   u64 bytenr, u64 num_bytes,
1830                                   u64 parent, u64 root_objectid,
1831                                   u64 owner, u64 offset, int refs_to_add,
1832                                   struct btrfs_delayed_extent_op *extent_op)
1833 {
1834         struct btrfs_path *path;
1835         struct extent_buffer *leaf;
1836         struct btrfs_extent_item *item;
1837         u64 refs;
1838         int ret;
1839         int err = 0;
1840
1841         path = btrfs_alloc_path();
1842         if (!path)
1843                 return -ENOMEM;
1844
1845         path->reada = 1;
1846         path->leave_spinning = 1;
1847         /* this will setup the path even if it fails to insert the back ref */
1848         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1849                                            path, bytenr, num_bytes, parent,
1850                                            root_objectid, owner, offset,
1851                                            refs_to_add, extent_op);
1852         if (ret == 0)
1853                 goto out;
1854
1855         if (ret != -EAGAIN) {
1856                 err = ret;
1857                 goto out;
1858         }
1859
1860         leaf = path->nodes[0];
1861         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1862         refs = btrfs_extent_refs(leaf, item);
1863         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1864         if (extent_op)
1865                 __run_delayed_extent_op(extent_op, leaf, item);
1866
1867         btrfs_mark_buffer_dirty(leaf);
1868         btrfs_release_path(root->fs_info->extent_root, path);
1869
1870         path->reada = 1;
1871         path->leave_spinning = 1;
1872
1873         /* now insert the actual backref */
1874         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1875                                     path, bytenr, parent, root_objectid,
1876                                     owner, offset, refs_to_add);
1877         BUG_ON(ret);
1878 out:
1879         btrfs_free_path(path);
1880         return err;
1881 }
1882
1883 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1884                                 struct btrfs_root *root,
1885                                 struct btrfs_delayed_ref_node *node,
1886                                 struct btrfs_delayed_extent_op *extent_op,
1887                                 int insert_reserved)
1888 {
1889         int ret = 0;
1890         struct btrfs_delayed_data_ref *ref;
1891         struct btrfs_key ins;
1892         u64 parent = 0;
1893         u64 ref_root = 0;
1894         u64 flags = 0;
1895
1896         ins.objectid = node->bytenr;
1897         ins.offset = node->num_bytes;
1898         ins.type = BTRFS_EXTENT_ITEM_KEY;
1899
1900         ref = btrfs_delayed_node_to_data_ref(node);
1901         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1902                 parent = ref->parent;
1903         else
1904                 ref_root = ref->root;
1905
1906         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1907                 if (extent_op) {
1908                         BUG_ON(extent_op->update_key);
1909                         flags |= extent_op->flags_to_set;
1910                 }
1911                 ret = alloc_reserved_file_extent(trans, root,
1912                                                  parent, ref_root, flags,
1913                                                  ref->objectid, ref->offset,
1914                                                  &ins, node->ref_mod);
1915         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1916                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1917                                              node->num_bytes, parent,
1918                                              ref_root, ref->objectid,
1919                                              ref->offset, node->ref_mod,
1920                                              extent_op);
1921         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1922                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1923                                           node->num_bytes, parent,
1924                                           ref_root, ref->objectid,
1925                                           ref->offset, node->ref_mod,
1926                                           extent_op);
1927         } else {
1928                 BUG();
1929         }
1930         return ret;
1931 }
1932
1933 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1934                                     struct extent_buffer *leaf,
1935                                     struct btrfs_extent_item *ei)
1936 {
1937         u64 flags = btrfs_extent_flags(leaf, ei);
1938         if (extent_op->update_flags) {
1939                 flags |= extent_op->flags_to_set;
1940                 btrfs_set_extent_flags(leaf, ei, flags);
1941         }
1942
1943         if (extent_op->update_key) {
1944                 struct btrfs_tree_block_info *bi;
1945                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1946                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1947                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1948         }
1949 }
1950
1951 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1952                                  struct btrfs_root *root,
1953                                  struct btrfs_delayed_ref_node *node,
1954                                  struct btrfs_delayed_extent_op *extent_op)
1955 {
1956         struct btrfs_key key;
1957         struct btrfs_path *path;
1958         struct btrfs_extent_item *ei;
1959         struct extent_buffer *leaf;
1960         u32 item_size;
1961         int ret;
1962         int err = 0;
1963
1964         path = btrfs_alloc_path();
1965         if (!path)
1966                 return -ENOMEM;
1967
1968         key.objectid = node->bytenr;
1969         key.type = BTRFS_EXTENT_ITEM_KEY;
1970         key.offset = node->num_bytes;
1971
1972         path->reada = 1;
1973         path->leave_spinning = 1;
1974         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1975                                 path, 0, 1);
1976         if (ret < 0) {
1977                 err = ret;
1978                 goto out;
1979         }
1980         if (ret > 0) {
1981                 err = -EIO;
1982                 goto out;
1983         }
1984
1985         leaf = path->nodes[0];
1986         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1987 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1988         if (item_size < sizeof(*ei)) {
1989                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1990                                              path, (u64)-1, 0);
1991                 if (ret < 0) {
1992                         err = ret;
1993                         goto out;
1994                 }
1995                 leaf = path->nodes[0];
1996                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1997         }
1998 #endif
1999         BUG_ON(item_size < sizeof(*ei));
2000         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2001         __run_delayed_extent_op(extent_op, leaf, ei);
2002
2003         btrfs_mark_buffer_dirty(leaf);
2004 out:
2005         btrfs_free_path(path);
2006         return err;
2007 }
2008
2009 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2010                                 struct btrfs_root *root,
2011                                 struct btrfs_delayed_ref_node *node,
2012                                 struct btrfs_delayed_extent_op *extent_op,
2013                                 int insert_reserved)
2014 {
2015         int ret = 0;
2016         struct btrfs_delayed_tree_ref *ref;
2017         struct btrfs_key ins;
2018         u64 parent = 0;
2019         u64 ref_root = 0;
2020
2021         ins.objectid = node->bytenr;
2022         ins.offset = node->num_bytes;
2023         ins.type = BTRFS_EXTENT_ITEM_KEY;
2024
2025         ref = btrfs_delayed_node_to_tree_ref(node);
2026         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2027                 parent = ref->parent;
2028         else
2029                 ref_root = ref->root;
2030
2031         BUG_ON(node->ref_mod != 1);
2032         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2033                 BUG_ON(!extent_op || !extent_op->update_flags ||
2034                        !extent_op->update_key);
2035                 ret = alloc_reserved_tree_block(trans, root,
2036                                                 parent, ref_root,
2037                                                 extent_op->flags_to_set,
2038                                                 &extent_op->key,
2039                                                 ref->level, &ins);
2040         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2041                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2042                                              node->num_bytes, parent, ref_root,
2043                                              ref->level, 0, 1, extent_op);
2044         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2045                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2046                                           node->num_bytes, parent, ref_root,
2047                                           ref->level, 0, 1, extent_op);
2048         } else {
2049                 BUG();
2050         }
2051         return ret;
2052 }
2053
2054 /* helper function to actually process a single delayed ref entry */
2055 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2056                                struct btrfs_root *root,
2057                                struct btrfs_delayed_ref_node *node,
2058                                struct btrfs_delayed_extent_op *extent_op,
2059                                int insert_reserved)
2060 {
2061         int ret;
2062         if (btrfs_delayed_ref_is_head(node)) {
2063                 struct btrfs_delayed_ref_head *head;
2064                 /*
2065                  * we've hit the end of the chain and we were supposed
2066                  * to insert this extent into the tree.  But, it got
2067                  * deleted before we ever needed to insert it, so all
2068                  * we have to do is clean up the accounting
2069                  */
2070                 BUG_ON(extent_op);
2071                 head = btrfs_delayed_node_to_head(node);
2072                 if (insert_reserved) {
2073                         btrfs_pin_extent(root, node->bytenr,
2074                                          node->num_bytes, 1);
2075                         if (head->is_data) {
2076                                 ret = btrfs_del_csums(trans, root,
2077                                                       node->bytenr,
2078                                                       node->num_bytes);
2079                                 BUG_ON(ret);
2080                         }
2081                 }
2082                 mutex_unlock(&head->mutex);
2083                 return 0;
2084         }
2085
2086         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2087             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2088                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2089                                            insert_reserved);
2090         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2091                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2092                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2093                                            insert_reserved);
2094         else
2095                 BUG();
2096         return ret;
2097 }
2098
2099 static noinline struct btrfs_delayed_ref_node *
2100 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2101 {
2102         struct rb_node *node;
2103         struct btrfs_delayed_ref_node *ref;
2104         int action = BTRFS_ADD_DELAYED_REF;
2105 again:
2106         /*
2107          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2108          * this prevents ref count from going down to zero when
2109          * there still are pending delayed ref.
2110          */
2111         node = rb_prev(&head->node.rb_node);
2112         while (1) {
2113                 if (!node)
2114                         break;
2115                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2116                                 rb_node);
2117                 if (ref->bytenr != head->node.bytenr)
2118                         break;
2119                 if (ref->action == action)
2120                         return ref;
2121                 node = rb_prev(node);
2122         }
2123         if (action == BTRFS_ADD_DELAYED_REF) {
2124                 action = BTRFS_DROP_DELAYED_REF;
2125                 goto again;
2126         }
2127         return NULL;
2128 }
2129
2130 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2131                                        struct btrfs_root *root,
2132                                        struct list_head *cluster)
2133 {
2134         struct btrfs_delayed_ref_root *delayed_refs;
2135         struct btrfs_delayed_ref_node *ref;
2136         struct btrfs_delayed_ref_head *locked_ref = NULL;
2137         struct btrfs_delayed_extent_op *extent_op;
2138         int ret;
2139         int count = 0;
2140         int must_insert_reserved = 0;
2141
2142         delayed_refs = &trans->transaction->delayed_refs;
2143         while (1) {
2144                 if (!locked_ref) {
2145                         /* pick a new head ref from the cluster list */
2146                         if (list_empty(cluster))
2147                                 break;
2148
2149                         locked_ref = list_entry(cluster->next,
2150                                      struct btrfs_delayed_ref_head, cluster);
2151
2152                         /* grab the lock that says we are going to process
2153                          * all the refs for this head */
2154                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2155
2156                         /*
2157                          * we may have dropped the spin lock to get the head
2158                          * mutex lock, and that might have given someone else
2159                          * time to free the head.  If that's true, it has been
2160                          * removed from our list and we can move on.
2161                          */
2162                         if (ret == -EAGAIN) {
2163                                 locked_ref = NULL;
2164                                 count++;
2165                                 continue;
2166                         }
2167                 }
2168
2169                 /*
2170                  * record the must insert reserved flag before we
2171                  * drop the spin lock.
2172                  */
2173                 must_insert_reserved = locked_ref->must_insert_reserved;
2174                 locked_ref->must_insert_reserved = 0;
2175
2176                 extent_op = locked_ref->extent_op;
2177                 locked_ref->extent_op = NULL;
2178
2179                 /*
2180                  * locked_ref is the head node, so we have to go one
2181                  * node back for any delayed ref updates
2182                  */
2183                 ref = select_delayed_ref(locked_ref);
2184                 if (!ref) {
2185                         /* All delayed refs have been processed, Go ahead
2186                          * and send the head node to run_one_delayed_ref,
2187                          * so that any accounting fixes can happen
2188                          */
2189                         ref = &locked_ref->node;
2190
2191                         if (extent_op && must_insert_reserved) {
2192                                 kfree(extent_op);
2193                                 extent_op = NULL;
2194                         }
2195
2196                         if (extent_op) {
2197                                 spin_unlock(&delayed_refs->lock);
2198
2199                                 ret = run_delayed_extent_op(trans, root,
2200                                                             ref, extent_op);
2201                                 BUG_ON(ret);
2202                                 kfree(extent_op);
2203
2204                                 cond_resched();
2205                                 spin_lock(&delayed_refs->lock);
2206                                 continue;
2207                         }
2208
2209                         list_del_init(&locked_ref->cluster);
2210                         locked_ref = NULL;
2211                 }
2212
2213                 ref->in_tree = 0;
2214                 rb_erase(&ref->rb_node, &delayed_refs->root);
2215                 delayed_refs->num_entries--;
2216
2217                 spin_unlock(&delayed_refs->lock);
2218
2219                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2220                                           must_insert_reserved);
2221                 BUG_ON(ret);
2222
2223                 btrfs_put_delayed_ref(ref);
2224                 kfree(extent_op);
2225                 count++;
2226
2227                 cond_resched();
2228                 spin_lock(&delayed_refs->lock);
2229         }
2230         return count;
2231 }
2232
2233 /*
2234  * this starts processing the delayed reference count updates and
2235  * extent insertions we have queued up so far.  count can be
2236  * 0, which means to process everything in the tree at the start
2237  * of the run (but not newly added entries), or it can be some target
2238  * number you'd like to process.
2239  */
2240 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2241                            struct btrfs_root *root, unsigned long count)
2242 {
2243         struct rb_node *node;
2244         struct btrfs_delayed_ref_root *delayed_refs;
2245         struct btrfs_delayed_ref_node *ref;
2246         struct list_head cluster;
2247         int ret;
2248         int run_all = count == (unsigned long)-1;
2249         int run_most = 0;
2250
2251         if (root == root->fs_info->extent_root)
2252                 root = root->fs_info->tree_root;
2253
2254         delayed_refs = &trans->transaction->delayed_refs;
2255         INIT_LIST_HEAD(&cluster);
2256 again:
2257         spin_lock(&delayed_refs->lock);
2258         if (count == 0) {
2259                 count = delayed_refs->num_entries * 2;
2260                 run_most = 1;
2261         }
2262         while (1) {
2263                 if (!(run_all || run_most) &&
2264                     delayed_refs->num_heads_ready < 64)
2265                         break;
2266
2267                 /*
2268                  * go find something we can process in the rbtree.  We start at
2269                  * the beginning of the tree, and then build a cluster
2270                  * of refs to process starting at the first one we are able to
2271                  * lock
2272                  */
2273                 ret = btrfs_find_ref_cluster(trans, &cluster,
2274                                              delayed_refs->run_delayed_start);
2275                 if (ret)
2276                         break;
2277
2278                 ret = run_clustered_refs(trans, root, &cluster);
2279                 BUG_ON(ret < 0);
2280
2281                 count -= min_t(unsigned long, ret, count);
2282
2283                 if (count == 0)
2284                         break;
2285         }
2286
2287         if (run_all) {
2288                 node = rb_first(&delayed_refs->root);
2289                 if (!node)
2290                         goto out;
2291                 count = (unsigned long)-1;
2292
2293                 while (node) {
2294                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2295                                        rb_node);
2296                         if (btrfs_delayed_ref_is_head(ref)) {
2297                                 struct btrfs_delayed_ref_head *head;
2298
2299                                 head = btrfs_delayed_node_to_head(ref);
2300                                 atomic_inc(&ref->refs);
2301
2302                                 spin_unlock(&delayed_refs->lock);
2303                                 mutex_lock(&head->mutex);
2304                                 mutex_unlock(&head->mutex);
2305
2306                                 btrfs_put_delayed_ref(ref);
2307                                 cond_resched();
2308                                 goto again;
2309                         }
2310                         node = rb_next(node);
2311                 }
2312                 spin_unlock(&delayed_refs->lock);
2313                 schedule_timeout(1);
2314                 goto again;
2315         }
2316 out:
2317         spin_unlock(&delayed_refs->lock);
2318         return 0;
2319 }
2320
2321 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2322                                 struct btrfs_root *root,
2323                                 u64 bytenr, u64 num_bytes, u64 flags,
2324                                 int is_data)
2325 {
2326         struct btrfs_delayed_extent_op *extent_op;
2327         int ret;
2328
2329         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2330         if (!extent_op)
2331                 return -ENOMEM;
2332
2333         extent_op->flags_to_set = flags;
2334         extent_op->update_flags = 1;
2335         extent_op->update_key = 0;
2336         extent_op->is_data = is_data ? 1 : 0;
2337
2338         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2339         if (ret)
2340                 kfree(extent_op);
2341         return ret;
2342 }
2343
2344 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2345                                       struct btrfs_root *root,
2346                                       struct btrfs_path *path,
2347                                       u64 objectid, u64 offset, u64 bytenr)
2348 {
2349         struct btrfs_delayed_ref_head *head;
2350         struct btrfs_delayed_ref_node *ref;
2351         struct btrfs_delayed_data_ref *data_ref;
2352         struct btrfs_delayed_ref_root *delayed_refs;
2353         struct rb_node *node;
2354         int ret = 0;
2355
2356         ret = -ENOENT;
2357         delayed_refs = &trans->transaction->delayed_refs;
2358         spin_lock(&delayed_refs->lock);
2359         head = btrfs_find_delayed_ref_head(trans, bytenr);
2360         if (!head)
2361                 goto out;
2362
2363         if (!mutex_trylock(&head->mutex)) {
2364                 atomic_inc(&head->node.refs);
2365                 spin_unlock(&delayed_refs->lock);
2366
2367                 btrfs_release_path(root->fs_info->extent_root, path);
2368
2369                 mutex_lock(&head->mutex);
2370                 mutex_unlock(&head->mutex);
2371                 btrfs_put_delayed_ref(&head->node);
2372                 return -EAGAIN;
2373         }
2374
2375         node = rb_prev(&head->node.rb_node);
2376         if (!node)
2377                 goto out_unlock;
2378
2379         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2380
2381         if (ref->bytenr != bytenr)
2382                 goto out_unlock;
2383
2384         ret = 1;
2385         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2386                 goto out_unlock;
2387
2388         data_ref = btrfs_delayed_node_to_data_ref(ref);
2389
2390         node = rb_prev(node);
2391         if (node) {
2392                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2393                 if (ref->bytenr == bytenr)
2394                         goto out_unlock;
2395         }
2396
2397         if (data_ref->root != root->root_key.objectid ||
2398             data_ref->objectid != objectid || data_ref->offset != offset)
2399                 goto out_unlock;
2400
2401         ret = 0;
2402 out_unlock:
2403         mutex_unlock(&head->mutex);
2404 out:
2405         spin_unlock(&delayed_refs->lock);
2406         return ret;
2407 }
2408
2409 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2410                                         struct btrfs_root *root,
2411                                         struct btrfs_path *path,
2412                                         u64 objectid, u64 offset, u64 bytenr)
2413 {
2414         struct btrfs_root *extent_root = root->fs_info->extent_root;
2415         struct extent_buffer *leaf;
2416         struct btrfs_extent_data_ref *ref;
2417         struct btrfs_extent_inline_ref *iref;
2418         struct btrfs_extent_item *ei;
2419         struct btrfs_key key;
2420         u32 item_size;
2421         int ret;
2422
2423         key.objectid = bytenr;
2424         key.offset = (u64)-1;
2425         key.type = BTRFS_EXTENT_ITEM_KEY;
2426
2427         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2428         if (ret < 0)
2429                 goto out;
2430         BUG_ON(ret == 0);
2431
2432         ret = -ENOENT;
2433         if (path->slots[0] == 0)
2434                 goto out;
2435
2436         path->slots[0]--;
2437         leaf = path->nodes[0];
2438         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2439
2440         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2441                 goto out;
2442
2443         ret = 1;
2444         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2445 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2446         if (item_size < sizeof(*ei)) {
2447                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2448                 goto out;
2449         }
2450 #endif
2451         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2452
2453         if (item_size != sizeof(*ei) +
2454             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2455                 goto out;
2456
2457         if (btrfs_extent_generation(leaf, ei) <=
2458             btrfs_root_last_snapshot(&root->root_item))
2459                 goto out;
2460
2461         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2462         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2463             BTRFS_EXTENT_DATA_REF_KEY)
2464                 goto out;
2465
2466         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2467         if (btrfs_extent_refs(leaf, ei) !=
2468             btrfs_extent_data_ref_count(leaf, ref) ||
2469             btrfs_extent_data_ref_root(leaf, ref) !=
2470             root->root_key.objectid ||
2471             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2472             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2473                 goto out;
2474
2475         ret = 0;
2476 out:
2477         return ret;
2478 }
2479
2480 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2481                           struct btrfs_root *root,
2482                           u64 objectid, u64 offset, u64 bytenr)
2483 {
2484         struct btrfs_path *path;
2485         int ret;
2486         int ret2;
2487
2488         path = btrfs_alloc_path();
2489         if (!path)
2490                 return -ENOENT;
2491
2492         do {
2493                 ret = check_committed_ref(trans, root, path, objectid,
2494                                           offset, bytenr);
2495                 if (ret && ret != -ENOENT)
2496                         goto out;
2497
2498                 ret2 = check_delayed_ref(trans, root, path, objectid,
2499                                          offset, bytenr);
2500         } while (ret2 == -EAGAIN);
2501
2502         if (ret2 && ret2 != -ENOENT) {
2503                 ret = ret2;
2504                 goto out;
2505         }
2506
2507         if (ret != -ENOENT || ret2 != -ENOENT)
2508                 ret = 0;
2509 out:
2510         btrfs_free_path(path);
2511         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2512                 WARN_ON(ret > 0);
2513         return ret;
2514 }
2515
2516 #if 0
2517 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2518                     struct extent_buffer *buf, u32 nr_extents)
2519 {
2520         struct btrfs_key key;
2521         struct btrfs_file_extent_item *fi;
2522         u64 root_gen;
2523         u32 nritems;
2524         int i;
2525         int level;
2526         int ret = 0;
2527         int shared = 0;
2528
2529         if (!root->ref_cows)
2530                 return 0;
2531
2532         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2533                 shared = 0;
2534                 root_gen = root->root_key.offset;
2535         } else {
2536                 shared = 1;
2537                 root_gen = trans->transid - 1;
2538         }
2539
2540         level = btrfs_header_level(buf);
2541         nritems = btrfs_header_nritems(buf);
2542
2543         if (level == 0) {
2544                 struct btrfs_leaf_ref *ref;
2545                 struct btrfs_extent_info *info;
2546
2547                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2548                 if (!ref) {
2549                         ret = -ENOMEM;
2550                         goto out;
2551                 }
2552
2553                 ref->root_gen = root_gen;
2554                 ref->bytenr = buf->start;
2555                 ref->owner = btrfs_header_owner(buf);
2556                 ref->generation = btrfs_header_generation(buf);
2557                 ref->nritems = nr_extents;
2558                 info = ref->extents;
2559
2560                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2561                         u64 disk_bytenr;
2562                         btrfs_item_key_to_cpu(buf, &key, i);
2563                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2564                                 continue;
2565                         fi = btrfs_item_ptr(buf, i,
2566                                             struct btrfs_file_extent_item);
2567                         if (btrfs_file_extent_type(buf, fi) ==
2568                             BTRFS_FILE_EXTENT_INLINE)
2569                                 continue;
2570                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2571                         if (disk_bytenr == 0)
2572                                 continue;
2573
2574                         info->bytenr = disk_bytenr;
2575                         info->num_bytes =
2576                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2577                         info->objectid = key.objectid;
2578                         info->offset = key.offset;
2579                         info++;
2580                 }
2581
2582                 ret = btrfs_add_leaf_ref(root, ref, shared);
2583                 if (ret == -EEXIST && shared) {
2584                         struct btrfs_leaf_ref *old;
2585                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2586                         BUG_ON(!old);
2587                         btrfs_remove_leaf_ref(root, old);
2588                         btrfs_free_leaf_ref(root, old);
2589                         ret = btrfs_add_leaf_ref(root, ref, shared);
2590                 }
2591                 WARN_ON(ret);
2592                 btrfs_free_leaf_ref(root, ref);
2593         }
2594 out:
2595         return ret;
2596 }
2597
2598 /* when a block goes through cow, we update the reference counts of
2599  * everything that block points to.  The internal pointers of the block
2600  * can be in just about any order, and it is likely to have clusters of
2601  * things that are close together and clusters of things that are not.
2602  *
2603  * To help reduce the seeks that come with updating all of these reference
2604  * counts, sort them by byte number before actual updates are done.
2605  *
2606  * struct refsort is used to match byte number to slot in the btree block.
2607  * we sort based on the byte number and then use the slot to actually
2608  * find the item.
2609  *
2610  * struct refsort is smaller than strcut btrfs_item and smaller than
2611  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2612  * for a btree block, there's no way for a kmalloc of refsorts for a
2613  * single node to be bigger than a page.
2614  */
2615 struct refsort {
2616         u64 bytenr;
2617         u32 slot;
2618 };
2619
2620 /*
2621  * for passing into sort()
2622  */
2623 static int refsort_cmp(const void *a_void, const void *b_void)
2624 {
2625         const struct refsort *a = a_void;
2626         const struct refsort *b = b_void;
2627
2628         if (a->bytenr < b->bytenr)
2629                 return -1;
2630         if (a->bytenr > b->bytenr)
2631                 return 1;
2632         return 0;
2633 }
2634 #endif
2635
2636 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2637                            struct btrfs_root *root,
2638                            struct extent_buffer *buf,
2639                            int full_backref, int inc)
2640 {
2641         u64 bytenr;
2642         u64 num_bytes;
2643         u64 parent;
2644         u64 ref_root;
2645         u32 nritems;
2646         struct btrfs_key key;
2647         struct btrfs_file_extent_item *fi;
2648         int i;
2649         int level;
2650         int ret = 0;
2651         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2652                             u64, u64, u64, u64, u64, u64);
2653
2654         ref_root = btrfs_header_owner(buf);
2655         nritems = btrfs_header_nritems(buf);
2656         level = btrfs_header_level(buf);
2657
2658         if (!root->ref_cows && level == 0)
2659                 return 0;
2660
2661         if (inc)
2662                 process_func = btrfs_inc_extent_ref;
2663         else
2664                 process_func = btrfs_free_extent;
2665
2666         if (full_backref)
2667                 parent = buf->start;
2668         else
2669                 parent = 0;
2670
2671         for (i = 0; i < nritems; i++) {
2672                 if (level == 0) {
2673                         btrfs_item_key_to_cpu(buf, &key, i);
2674                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2675                                 continue;
2676                         fi = btrfs_item_ptr(buf, i,
2677                                             struct btrfs_file_extent_item);
2678                         if (btrfs_file_extent_type(buf, fi) ==
2679                             BTRFS_FILE_EXTENT_INLINE)
2680                                 continue;
2681                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2682                         if (bytenr == 0)
2683                                 continue;
2684
2685                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2686                         key.offset -= btrfs_file_extent_offset(buf, fi);
2687                         ret = process_func(trans, root, bytenr, num_bytes,
2688                                            parent, ref_root, key.objectid,
2689                                            key.offset);
2690                         if (ret)
2691                                 goto fail;
2692                 } else {
2693                         bytenr = btrfs_node_blockptr(buf, i);
2694                         num_bytes = btrfs_level_size(root, level - 1);
2695                         ret = process_func(trans, root, bytenr, num_bytes,
2696                                            parent, ref_root, level - 1, 0);
2697                         if (ret)
2698                                 goto fail;
2699                 }
2700         }
2701         return 0;
2702 fail:
2703         BUG();
2704         return ret;
2705 }
2706
2707 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2708                   struct extent_buffer *buf, int full_backref)
2709 {
2710         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2711 }
2712
2713 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2714                   struct extent_buffer *buf, int full_backref)
2715 {
2716         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2717 }
2718
2719 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2720                                  struct btrfs_root *root,
2721                                  struct btrfs_path *path,
2722                                  struct btrfs_block_group_cache *cache)
2723 {
2724         int ret;
2725         struct btrfs_root *extent_root = root->fs_info->extent_root;
2726         unsigned long bi;
2727         struct extent_buffer *leaf;
2728
2729         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2730         if (ret < 0)
2731                 goto fail;
2732         BUG_ON(ret);
2733
2734         leaf = path->nodes[0];
2735         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2736         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2737         btrfs_mark_buffer_dirty(leaf);
2738         btrfs_release_path(extent_root, path);
2739 fail:
2740         if (ret)
2741                 return ret;
2742         return 0;
2743
2744 }
2745
2746 static struct btrfs_block_group_cache *
2747 next_block_group(struct btrfs_root *root,
2748                  struct btrfs_block_group_cache *cache)
2749 {
2750         struct rb_node *node;
2751         spin_lock(&root->fs_info->block_group_cache_lock);
2752         node = rb_next(&cache->cache_node);
2753         btrfs_put_block_group(cache);
2754         if (node) {
2755                 cache = rb_entry(node, struct btrfs_block_group_cache,
2756                                  cache_node);
2757                 btrfs_get_block_group(cache);
2758         } else
2759                 cache = NULL;
2760         spin_unlock(&root->fs_info->block_group_cache_lock);
2761         return cache;
2762 }
2763
2764 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2765                             struct btrfs_trans_handle *trans,
2766                             struct btrfs_path *path)
2767 {
2768         struct btrfs_root *root = block_group->fs_info->tree_root;
2769         struct inode *inode = NULL;
2770         u64 alloc_hint = 0;
2771         int dcs = BTRFS_DC_ERROR;
2772         int num_pages = 0;
2773         int retries = 0;
2774         int ret = 0;
2775
2776         /*
2777          * If this block group is smaller than 100 megs don't bother caching the
2778          * block group.
2779          */
2780         if (block_group->key.offset < (100 * 1024 * 1024)) {
2781                 spin_lock(&block_group->lock);
2782                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2783                 spin_unlock(&block_group->lock);
2784                 return 0;
2785         }
2786
2787 again:
2788         inode = lookup_free_space_inode(root, block_group, path);
2789         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2790                 ret = PTR_ERR(inode);
2791                 btrfs_release_path(root, path);
2792                 goto out;
2793         }
2794
2795         if (IS_ERR(inode)) {
2796                 BUG_ON(retries);
2797                 retries++;
2798
2799                 if (block_group->ro)
2800                         goto out_free;
2801
2802                 ret = create_free_space_inode(root, trans, block_group, path);
2803                 if (ret)
2804                         goto out_free;
2805                 goto again;
2806         }
2807
2808         /*
2809          * We want to set the generation to 0, that way if anything goes wrong
2810          * from here on out we know not to trust this cache when we load up next
2811          * time.
2812          */
2813         BTRFS_I(inode)->generation = 0;
2814         ret = btrfs_update_inode(trans, root, inode);
2815         WARN_ON(ret);
2816
2817         if (i_size_read(inode) > 0) {
2818                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2819                                                       inode);
2820                 if (ret)
2821                         goto out_put;
2822         }
2823
2824         spin_lock(&block_group->lock);
2825         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2826                 /* We're not cached, don't bother trying to write stuff out */
2827                 dcs = BTRFS_DC_WRITTEN;
2828                 spin_unlock(&block_group->lock);
2829                 goto out_put;
2830         }
2831         spin_unlock(&block_group->lock);
2832
2833         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2834         if (!num_pages)
2835                 num_pages = 1;
2836
2837         /*
2838          * Just to make absolutely sure we have enough space, we're going to
2839          * preallocate 12 pages worth of space for each block group.  In
2840          * practice we ought to use at most 8, but we need extra space so we can
2841          * add our header and have a terminator between the extents and the
2842          * bitmaps.
2843          */
2844         num_pages *= 16;
2845         num_pages *= PAGE_CACHE_SIZE;
2846
2847         ret = btrfs_check_data_free_space(inode, num_pages);
2848         if (ret)
2849                 goto out_put;
2850
2851         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2852                                               num_pages, num_pages,
2853                                               &alloc_hint);
2854         if (!ret)
2855                 dcs = BTRFS_DC_SETUP;
2856         btrfs_free_reserved_data_space(inode, num_pages);
2857 out_put:
2858         iput(inode);
2859 out_free:
2860         btrfs_release_path(root, path);
2861 out:
2862         spin_lock(&block_group->lock);
2863         block_group->disk_cache_state = dcs;
2864         spin_unlock(&block_group->lock);
2865
2866         return ret;
2867 }
2868
2869 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2870                                    struct btrfs_root *root)
2871 {
2872         struct btrfs_block_group_cache *cache;
2873         int err = 0;
2874         struct btrfs_path *path;
2875         u64 last = 0;
2876
2877         path = btrfs_alloc_path();
2878         if (!path)
2879                 return -ENOMEM;
2880
2881 again:
2882         while (1) {
2883                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2884                 while (cache) {
2885                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2886                                 break;
2887                         cache = next_block_group(root, cache);
2888                 }
2889                 if (!cache) {
2890                         if (last == 0)
2891                                 break;
2892                         last = 0;
2893                         continue;
2894                 }
2895                 err = cache_save_setup(cache, trans, path);
2896                 last = cache->key.objectid + cache->key.offset;
2897                 btrfs_put_block_group(cache);
2898         }
2899
2900         while (1) {
2901                 if (last == 0) {
2902                         err = btrfs_run_delayed_refs(trans, root,
2903                                                      (unsigned long)-1);
2904                         BUG_ON(err);
2905                 }
2906
2907                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2908                 while (cache) {
2909                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2910                                 btrfs_put_block_group(cache);
2911                                 goto again;
2912                         }
2913
2914                         if (cache->dirty)
2915                                 break;
2916                         cache = next_block_group(root, cache);
2917                 }
2918                 if (!cache) {
2919                         if (last == 0)
2920                                 break;
2921                         last = 0;
2922                         continue;
2923                 }
2924
2925                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2926                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2927                 cache->dirty = 0;
2928                 last = cache->key.objectid + cache->key.offset;
2929
2930                 err = write_one_cache_group(trans, root, path, cache);
2931                 BUG_ON(err);
2932                 btrfs_put_block_group(cache);
2933         }
2934
2935         while (1) {
2936                 /*
2937                  * I don't think this is needed since we're just marking our
2938                  * preallocated extent as written, but just in case it can't
2939                  * hurt.
2940                  */
2941                 if (last == 0) {
2942                         err = btrfs_run_delayed_refs(trans, root,
2943                                                      (unsigned long)-1);
2944                         BUG_ON(err);
2945                 }
2946
2947                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2948                 while (cache) {
2949                         /*
2950                          * Really this shouldn't happen, but it could if we
2951                          * couldn't write the entire preallocated extent and
2952                          * splitting the extent resulted in a new block.
2953                          */
2954                         if (cache->dirty) {
2955                                 btrfs_put_block_group(cache);
2956                                 goto again;
2957                         }
2958                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2959                                 break;
2960                         cache = next_block_group(root, cache);
2961                 }
2962                 if (!cache) {
2963                         if (last == 0)
2964                                 break;
2965                         last = 0;
2966                         continue;
2967                 }
2968
2969                 btrfs_write_out_cache(root, trans, cache, path);
2970
2971                 /*
2972                  * If we didn't have an error then the cache state is still
2973                  * NEED_WRITE, so we can set it to WRITTEN.
2974                  */
2975                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2976                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2977                 last = cache->key.objectid + cache->key.offset;
2978                 btrfs_put_block_group(cache);
2979         }
2980
2981         btrfs_free_path(path);
2982         return 0;
2983 }
2984
2985 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2986 {
2987         struct btrfs_block_group_cache *block_group;
2988         int readonly = 0;
2989
2990         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2991         if (!block_group || block_group->ro)
2992                 readonly = 1;
2993         if (block_group)
2994                 btrfs_put_block_group(block_group);
2995         return readonly;
2996 }
2997
2998 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2999                              u64 total_bytes, u64 bytes_used,
3000                              struct btrfs_space_info **space_info)
3001 {
3002         struct btrfs_space_info *found;
3003         int i;
3004         int factor;
3005
3006         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3007                      BTRFS_BLOCK_GROUP_RAID10))
3008                 factor = 2;
3009         else
3010                 factor = 1;
3011
3012         found = __find_space_info(info, flags);
3013         if (found) {
3014                 spin_lock(&found->lock);
3015                 found->total_bytes += total_bytes;
3016                 found->disk_total += total_bytes * factor;
3017                 found->bytes_used += bytes_used;
3018                 found->disk_used += bytes_used * factor;
3019                 found->full = 0;
3020                 spin_unlock(&found->lock);
3021                 *space_info = found;
3022                 return 0;
3023         }
3024         found = kzalloc(sizeof(*found), GFP_NOFS);
3025         if (!found)
3026                 return -ENOMEM;
3027
3028         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3029                 INIT_LIST_HEAD(&found->block_groups[i]);
3030         init_rwsem(&found->groups_sem);
3031         spin_lock_init(&found->lock);
3032         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3033                                 BTRFS_BLOCK_GROUP_SYSTEM |
3034                                 BTRFS_BLOCK_GROUP_METADATA);
3035         found->total_bytes = total_bytes;
3036         found->disk_total = total_bytes * factor;
3037         found->bytes_used = bytes_used;
3038         found->disk_used = bytes_used * factor;
3039         found->bytes_pinned = 0;
3040         found->bytes_reserved = 0;
3041         found->bytes_readonly = 0;
3042         found->bytes_may_use = 0;
3043         found->full = 0;
3044         found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3045         found->chunk_alloc = 0;
3046         *space_info = found;
3047         list_add_rcu(&found->list, &info->space_info);
3048         atomic_set(&found->caching_threads, 0);
3049         return 0;
3050 }
3051
3052 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3053 {
3054         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3055                                    BTRFS_BLOCK_GROUP_RAID1 |
3056                                    BTRFS_BLOCK_GROUP_RAID10 |
3057                                    BTRFS_BLOCK_GROUP_DUP);
3058         if (extra_flags) {
3059                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3060                         fs_info->avail_data_alloc_bits |= extra_flags;
3061                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3062                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3063                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3064                         fs_info->avail_system_alloc_bits |= extra_flags;
3065         }
3066 }
3067
3068 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3069 {
3070         /*
3071          * we add in the count of missing devices because we want
3072          * to make sure that any RAID levels on a degraded FS
3073          * continue to be honored.
3074          */
3075         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3076                 root->fs_info->fs_devices->missing_devices;
3077
3078         if (num_devices == 1)
3079                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3080         if (num_devices < 4)
3081                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3082
3083         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3084             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3085                       BTRFS_BLOCK_GROUP_RAID10))) {
3086                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3087         }
3088
3089         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3090             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3091                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3092         }
3093
3094         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3095             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3096              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3097              (flags & BTRFS_BLOCK_GROUP_DUP)))
3098                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3099         return flags;
3100 }
3101
3102 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3103 {
3104         if (flags & BTRFS_BLOCK_GROUP_DATA)
3105                 flags |= root->fs_info->avail_data_alloc_bits &
3106                          root->fs_info->data_alloc_profile;
3107         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3108                 flags |= root->fs_info->avail_system_alloc_bits &
3109                          root->fs_info->system_alloc_profile;
3110         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3111                 flags |= root->fs_info->avail_metadata_alloc_bits &
3112                          root->fs_info->metadata_alloc_profile;
3113         return btrfs_reduce_alloc_profile(root, flags);
3114 }
3115
3116 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3117 {
3118         u64 flags;
3119
3120         if (data)
3121                 flags = BTRFS_BLOCK_GROUP_DATA;
3122         else if (root == root->fs_info->chunk_root)
3123                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3124         else
3125                 flags = BTRFS_BLOCK_GROUP_METADATA;
3126
3127         return get_alloc_profile(root, flags);
3128 }
3129
3130 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3131 {
3132         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3133                                                        BTRFS_BLOCK_GROUP_DATA);
3134 }
3135
3136 /*
3137  * This will check the space that the inode allocates from to make sure we have
3138  * enough space for bytes.
3139  */
3140 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3141 {
3142         struct btrfs_space_info *data_sinfo;
3143         struct btrfs_root *root = BTRFS_I(inode)->root;
3144         u64 used;
3145         int ret = 0, committed = 0, alloc_chunk = 1;
3146
3147         /* make sure bytes are sectorsize aligned */
3148         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3149
3150         if (root == root->fs_info->tree_root) {
3151                 alloc_chunk = 0;
3152                 committed = 1;
3153         }
3154
3155         data_sinfo = BTRFS_I(inode)->space_info;
3156         if (!data_sinfo)
3157                 goto alloc;
3158
3159 again:
3160         /* make sure we have enough space to handle the data first */
3161         spin_lock(&data_sinfo->lock);
3162         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3163                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3164                 data_sinfo->bytes_may_use;
3165
3166         if (used + bytes > data_sinfo->total_bytes) {
3167                 struct btrfs_trans_handle *trans;
3168
3169                 /*
3170                  * if we don't have enough free bytes in this space then we need
3171                  * to alloc a new chunk.
3172                  */
3173                 if (!data_sinfo->full && alloc_chunk) {
3174                         u64 alloc_target;
3175
3176                         data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3177                         spin_unlock(&data_sinfo->lock);
3178 alloc:
3179                         alloc_target = btrfs_get_alloc_profile(root, 1);
3180                         trans = btrfs_join_transaction(root);
3181                         if (IS_ERR(trans))
3182                                 return PTR_ERR(trans);
3183
3184                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3185                                              bytes + 2 * 1024 * 1024,
3186                                              alloc_target,
3187                                              CHUNK_ALLOC_NO_FORCE);
3188                         btrfs_end_transaction(trans, root);
3189                         if (ret < 0) {
3190                                 if (ret != -ENOSPC)
3191                                         return ret;
3192                                 else
3193                                         goto commit_trans;
3194                         }
3195
3196                         if (!data_sinfo) {
3197                                 btrfs_set_inode_space_info(root, inode);
3198                                 data_sinfo = BTRFS_I(inode)->space_info;
3199                         }
3200                         goto again;
3201                 }
3202                 spin_unlock(&data_sinfo->lock);
3203
3204                 /* commit the current transaction and try again */
3205 commit_trans:
3206                 if (!committed &&
3207                     !atomic_read(&root->fs_info->open_ioctl_trans)) {
3208                         committed = 1;
3209                         trans = btrfs_join_transaction(root);
3210                         if (IS_ERR(trans))
3211                                 return PTR_ERR(trans);
3212                         ret = btrfs_commit_transaction(trans, root);
3213                         if (ret)
3214                                 return ret;
3215                         goto again;
3216                 }
3217
3218 #if 0 /* I hope we never need this code again, just in case */
3219                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3220                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3221                        "%llu bytes_readonly, %llu may use %llu total\n",
3222                        (unsigned long long)bytes,
3223                        (unsigned long long)data_sinfo->bytes_used,
3224                        (unsigned long long)data_sinfo->bytes_reserved,
3225                        (unsigned long long)data_sinfo->bytes_pinned,
3226                        (unsigned long long)data_sinfo->bytes_readonly,
3227                        (unsigned long long)data_sinfo->bytes_may_use,
3228                        (unsigned long long)data_sinfo->total_bytes);
3229 #endif
3230                 return -ENOSPC;
3231         }
3232         data_sinfo->bytes_may_use += bytes;
3233         BTRFS_I(inode)->reserved_bytes += bytes;
3234         spin_unlock(&data_sinfo->lock);
3235
3236         return 0;
3237 }
3238
3239 /*
3240  * called when we are clearing an delalloc extent from the
3241  * inode's io_tree or there was an error for whatever reason
3242  * after calling btrfs_check_data_free_space
3243  */
3244 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3245 {
3246         struct btrfs_root *root = BTRFS_I(inode)->root;
3247         struct btrfs_space_info *data_sinfo;
3248
3249         /* make sure bytes are sectorsize aligned */
3250         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3251
3252         data_sinfo = BTRFS_I(inode)->space_info;
3253         spin_lock(&data_sinfo->lock);
3254         data_sinfo->bytes_may_use -= bytes;
3255         BTRFS_I(inode)->reserved_bytes -= bytes;
3256         spin_unlock(&data_sinfo->lock);
3257 }
3258
3259 static void force_metadata_allocation(struct btrfs_fs_info *info)
3260 {
3261         struct list_head *head = &info->space_info;
3262         struct btrfs_space_info *found;
3263
3264         rcu_read_lock();
3265         list_for_each_entry_rcu(found, head, list) {
3266                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3267                         found->force_alloc = CHUNK_ALLOC_FORCE;
3268         }
3269         rcu_read_unlock();
3270 }
3271
3272 static int should_alloc_chunk(struct btrfs_root *root,
3273                               struct btrfs_space_info *sinfo, u64 alloc_bytes,
3274                               int force)
3275 {
3276         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3277         u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3278         u64 thresh;
3279
3280         if (force == CHUNK_ALLOC_FORCE)
3281                 return 1;
3282
3283         /*
3284          * in limited mode, we want to have some free space up to
3285          * about 1% of the FS size.
3286          */
3287         if (force == CHUNK_ALLOC_LIMITED) {
3288                 thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3289                 thresh = max_t(u64, 64 * 1024 * 1024,
3290                                div_factor_fine(thresh, 1));
3291
3292                 if (num_bytes - num_allocated < thresh)
3293                         return 1;
3294         }
3295
3296         /*
3297          * we have two similar checks here, one based on percentage
3298          * and once based on a hard number of 256MB.  The idea
3299          * is that if we have a good amount of free
3300          * room, don't allocate a chunk.  A good mount is
3301          * less than 80% utilized of the chunks we have allocated,
3302          * or more than 256MB free
3303          */
3304         if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3305                 return 0;
3306
3307         if (num_allocated + alloc_bytes < div_factor(num_bytes, 8))
3308                 return 0;
3309
3310         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3311
3312         /* 256MB or 5% of the FS */
3313         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3314
3315         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3316                 return 0;
3317         return 1;
3318 }
3319
3320 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3321                           struct btrfs_root *extent_root, u64 alloc_bytes,
3322                           u64 flags, int force)
3323 {
3324         struct btrfs_space_info *space_info;
3325         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3326         int wait_for_alloc = 0;
3327         int ret = 0;
3328
3329         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3330
3331         space_info = __find_space_info(extent_root->fs_info, flags);
3332         if (!space_info) {
3333                 ret = update_space_info(extent_root->fs_info, flags,
3334                                         0, 0, &space_info);
3335                 BUG_ON(ret);
3336         }
3337         BUG_ON(!space_info);
3338
3339 again:
3340         spin_lock(&space_info->lock);
3341         if (space_info->force_alloc)
3342                 force = space_info->force_alloc;
3343         if (space_info->full) {
3344                 spin_unlock(&space_info->lock);
3345                 return 0;
3346         }
3347
3348         if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) {
3349                 spin_unlock(&space_info->lock);
3350                 return 0;
3351         } else if (space_info->chunk_alloc) {
3352                 wait_for_alloc = 1;
3353         } else {
3354                 space_info->chunk_alloc = 1;
3355         }
3356
3357         spin_unlock(&space_info->lock);
3358
3359         mutex_lock(&fs_info->chunk_mutex);
3360
3361         /*
3362          * The chunk_mutex is held throughout the entirety of a chunk
3363          * allocation, so once we've acquired the chunk_mutex we know that the
3364          * other guy is done and we need to recheck and see if we should
3365          * allocate.
3366          */
3367         if (wait_for_alloc) {
3368                 mutex_unlock(&fs_info->chunk_mutex);
3369                 wait_for_alloc = 0;
3370                 goto again;
3371         }
3372
3373         /*
3374          * If we have mixed data/metadata chunks we want to make sure we keep
3375          * allocating mixed chunks instead of individual chunks.
3376          */
3377         if (btrfs_mixed_space_info(space_info))
3378                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3379
3380         /*
3381          * if we're doing a data chunk, go ahead and make sure that
3382          * we keep a reasonable number of metadata chunks allocated in the
3383          * FS as well.
3384          */
3385         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3386                 fs_info->data_chunk_allocations++;
3387                 if (!(fs_info->data_chunk_allocations %
3388                       fs_info->metadata_ratio))
3389                         force_metadata_allocation(fs_info);
3390         }
3391
3392         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3393         spin_lock(&space_info->lock);
3394         if (ret)
3395                 space_info->full = 1;
3396         else
3397                 ret = 1;
3398
3399         space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3400         space_info->chunk_alloc = 0;
3401         spin_unlock(&space_info->lock);
3402         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3403         return ret;
3404 }
3405
3406 /*
3407  * shrink metadata reservation for delalloc
3408  */
3409 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3410                            struct btrfs_root *root, u64 to_reclaim, int sync)
3411 {
3412         struct btrfs_block_rsv *block_rsv;
3413         struct btrfs_space_info *space_info;
3414         u64 reserved;
3415         u64 max_reclaim;
3416         u64 reclaimed = 0;
3417         long time_left;
3418         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3419         int loops = 0;
3420         unsigned long progress;
3421
3422         block_rsv = &root->fs_info->delalloc_block_rsv;
3423         space_info = block_rsv->space_info;
3424
3425         smp_mb();
3426         reserved = space_info->bytes_reserved;
3427         progress = space_info->reservation_progress;
3428
3429         if (reserved == 0)
3430                 return 0;
3431
3432         max_reclaim = min(reserved, to_reclaim);
3433
3434         while (loops < 1024) {
3435                 /* have the flusher threads jump in and do some IO */
3436                 smp_mb();
3437                 nr_pages = min_t(unsigned long, nr_pages,
3438                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3439                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3440
3441                 spin_lock(&space_info->lock);
3442                 if (reserved > space_info->bytes_reserved)
3443                         reclaimed += reserved - space_info->bytes_reserved;
3444                 reserved = space_info->bytes_reserved;
3445                 spin_unlock(&space_info->lock);
3446
3447                 loops++;
3448
3449                 if (reserved == 0 || reclaimed >= max_reclaim)
3450                         break;
3451
3452                 if (trans && trans->transaction->blocked)
3453                         return -EAGAIN;
3454
3455                 time_left = schedule_timeout_interruptible(1);
3456
3457                 /* We were interrupted, exit */
3458                 if (time_left)
3459                         break;
3460
3461                 /* we've kicked the IO a few times, if anything has been freed,
3462                  * exit.  There is no sense in looping here for a long time
3463                  * when we really need to commit the transaction, or there are
3464                  * just too many writers without enough free space
3465                  */
3466
3467                 if (loops > 3) {
3468                         smp_mb();
3469                         if (progress != space_info->reservation_progress)
3470                                 break;
3471                 }
3472
3473         }
3474         return reclaimed >= to_reclaim;
3475 }
3476
3477 /*
3478  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3479  * idea is if this is the first call (retries == 0) then we will add to our
3480  * reserved count if we can't make the allocation in order to hold our place
3481  * while we go and try and free up space.  That way for retries > 1 we don't try
3482  * and add space, we just check to see if the amount of unused space is >= the
3483  * total space, meaning that our reservation is valid.
3484  *
3485  * However if we don't intend to retry this reservation, pass -1 as retries so
3486  * that it short circuits this logic.
3487  */
3488 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3489                                   struct btrfs_root *root,
3490                                   struct btrfs_block_rsv *block_rsv,
3491                                   u64 orig_bytes, int flush)
3492 {
3493         struct btrfs_space_info *space_info = block_rsv->space_info;
3494         u64 unused;
3495         u64 num_bytes = orig_bytes;
3496         int retries = 0;
3497         int ret = 0;
3498         bool reserved = false;
3499         bool committed = false;
3500
3501 again:
3502         ret = -ENOSPC;
3503         if (reserved)
3504                 num_bytes = 0;
3505
3506         spin_lock(&space_info->lock);
3507         unused = space_info->bytes_used + space_info->bytes_reserved +
3508                  space_info->bytes_pinned + space_info->bytes_readonly +
3509                  space_info->bytes_may_use;
3510
3511         /*
3512          * The idea here is that we've not already over-reserved the block group
3513          * then we can go ahead and save our reservation first and then start
3514          * flushing if we need to.  Otherwise if we've already overcommitted
3515          * lets start flushing stuff first and then come back and try to make
3516          * our reservation.
3517          */
3518         if (unused <= space_info->total_bytes) {
3519                 unused = space_info->total_bytes - unused;
3520                 if (unused >= num_bytes) {
3521                         if (!reserved)
3522                                 space_info->bytes_reserved += orig_bytes;
3523                         ret = 0;
3524                 } else {
3525                         /*
3526                          * Ok set num_bytes to orig_bytes since we aren't
3527                          * overocmmitted, this way we only try and reclaim what
3528                          * we need.
3529                          */
3530                         num_bytes = orig_bytes;
3531                 }
3532         } else {
3533                 /*
3534                  * Ok we're over committed, set num_bytes to the overcommitted
3535                  * amount plus the amount of bytes that we need for this
3536                  * reservation.
3537                  */
3538                 num_bytes = unused - space_info->total_bytes +
3539                         (orig_bytes * (retries + 1));
3540         }
3541
3542         /*
3543          * Couldn't make our reservation, save our place so while we're trying
3544          * to reclaim space we can actually use it instead of somebody else
3545          * stealing it from us.
3546          */
3547         if (ret && !reserved) {
3548                 space_info->bytes_reserved += orig_bytes;
3549                 reserved = true;
3550         }
3551
3552         spin_unlock(&space_info->lock);
3553
3554         if (!ret)
3555                 return 0;
3556
3557         if (!flush)
3558                 goto out;
3559
3560         /*
3561          * We do synchronous shrinking since we don't actually unreserve
3562          * metadata until after the IO is completed.
3563          */
3564         ret = shrink_delalloc(trans, root, num_bytes, 1);
3565         if (ret > 0)
3566                 return 0;
3567         else if (ret < 0)
3568                 goto out;
3569
3570         /*
3571          * So if we were overcommitted it's possible that somebody else flushed
3572          * out enough space and we simply didn't have enough space to reclaim,
3573          * so go back around and try again.
3574          */
3575         if (retries < 2) {
3576                 retries++;
3577                 goto again;
3578         }
3579
3580         spin_lock(&space_info->lock);
3581         /*
3582          * Not enough space to be reclaimed, don't bother committing the
3583          * transaction.
3584          */
3585         if (space_info->bytes_pinned < orig_bytes)
3586                 ret = -ENOSPC;
3587         spin_unlock(&space_info->lock);
3588         if (ret)
3589                 goto out;
3590
3591         ret = -EAGAIN;
3592         if (trans || committed)
3593                 goto out;
3594
3595         ret = -ENOSPC;
3596         trans = btrfs_join_transaction(root);
3597         if (IS_ERR(trans))
3598                 goto out;
3599         ret = btrfs_commit_transaction(trans, root);
3600         if (!ret) {
3601                 trans = NULL;
3602                 committed = true;
3603                 goto again;
3604         }
3605
3606 out:
3607         if (reserved) {
3608                 spin_lock(&space_info->lock);
3609                 space_info->bytes_reserved -= orig_bytes;
3610                 spin_unlock(&space_info->lock);
3611         }
3612
3613         return ret;
3614 }
3615
3616 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3617                                              struct btrfs_root *root)
3618 {
3619         struct btrfs_block_rsv *block_rsv;
3620         if (root->ref_cows)
3621                 block_rsv = trans->block_rsv;
3622         else
3623                 block_rsv = root->block_rsv;
3624
3625         if (!block_rsv)
3626                 block_rsv = &root->fs_info->empty_block_rsv;
3627
3628         return block_rsv;
3629 }
3630
3631 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3632                                u64 num_bytes)
3633 {
3634         int ret = -ENOSPC;
3635         spin_lock(&block_rsv->lock);
3636         if (block_rsv->reserved >= num_bytes) {
3637                 block_rsv->reserved -= num_bytes;
3638                 if (block_rsv->reserved < block_rsv->size)
3639                         block_rsv->full = 0;
3640                 ret = 0;
3641         }
3642         spin_unlock(&block_rsv->lock);
3643         return ret;
3644 }
3645
3646 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3647                                 u64 num_bytes, int update_size)
3648 {
3649         spin_lock(&block_rsv->lock);
3650         block_rsv->reserved += num_bytes;
3651         if (update_size)
3652                 block_rsv->size += num_bytes;
3653         else if (block_rsv->reserved >= block_rsv->size)
3654                 block_rsv->full = 1;
3655         spin_unlock(&block_rsv->lock);
3656 }
3657
3658 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3659                              struct btrfs_block_rsv *dest, u64 num_bytes)
3660 {
3661         struct btrfs_space_info *space_info = block_rsv->space_info;
3662
3663         spin_lock(&block_rsv->lock);
3664         if (num_bytes == (u64)-1)
3665                 num_bytes = block_rsv->size;
3666         block_rsv->size -= num_bytes;
3667         if (block_rsv->reserved >= block_rsv->size) {
3668                 num_bytes = block_rsv->reserved - block_rsv->size;
3669                 block_rsv->reserved = block_rsv->size;
3670                 block_rsv->full = 1;
3671         } else {
3672                 num_bytes = 0;
3673         }
3674         spin_unlock(&block_rsv->lock);
3675
3676         if (num_bytes > 0) {
3677                 if (dest) {
3678                         spin_lock(&dest->lock);
3679                         if (!dest->full) {
3680                                 u64 bytes_to_add;
3681
3682                                 bytes_to_add = dest->size - dest->reserved;
3683                                 bytes_to_add = min(num_bytes, bytes_to_add);
3684                                 dest->reserved += bytes_to_add;
3685                                 if (dest->reserved >= dest->size)
3686                                         dest->full = 1;
3687                                 num_bytes -= bytes_to_add;
3688                         }
3689                         spin_unlock(&dest->lock);
3690                 }
3691                 if (num_bytes) {
3692                         spin_lock(&space_info->lock);
3693                         space_info->bytes_reserved -= num_bytes;
3694                         space_info->reservation_progress++;
3695                         spin_unlock(&space_info->lock);
3696                 }
3697         }
3698 }
3699
3700 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3701                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3702 {
3703         int ret;
3704
3705         ret = block_rsv_use_bytes(src, num_bytes);
3706         if (ret)
3707                 return ret;
3708
3709         block_rsv_add_bytes(dst, num_bytes, 1);
3710         return 0;
3711 }
3712
3713 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3714 {
3715         memset(rsv, 0, sizeof(*rsv));
3716         spin_lock_init(&rsv->lock);
3717         atomic_set(&rsv->usage, 1);
3718         rsv->priority = 6;
3719         INIT_LIST_HEAD(&rsv->list);
3720 }
3721
3722 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3723 {
3724         struct btrfs_block_rsv *block_rsv;
3725         struct btrfs_fs_info *fs_info = root->fs_info;
3726
3727         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3728         if (!block_rsv)
3729                 return NULL;
3730
3731         btrfs_init_block_rsv(block_rsv);
3732         block_rsv->space_info = __find_space_info(fs_info,
3733                                                   BTRFS_BLOCK_GROUP_METADATA);
3734         return block_rsv;
3735 }
3736
3737 void btrfs_free_block_rsv(struct btrfs_root *root,
3738                           struct btrfs_block_rsv *rsv)
3739 {
3740         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3741                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3742                 if (!rsv->durable)
3743                         kfree(rsv);
3744         }
3745 }
3746
3747 /*
3748  * make the block_rsv struct be able to capture freed space.
3749  * the captured space will re-add to the the block_rsv struct
3750  * after transaction commit
3751  */
3752 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3753                                  struct btrfs_block_rsv *block_rsv)
3754 {
3755         block_rsv->durable = 1;
3756         mutex_lock(&fs_info->durable_block_rsv_mutex);
3757         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3758         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3759 }
3760
3761 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3762                         struct btrfs_root *root,
3763                         struct btrfs_block_rsv *block_rsv,
3764                         u64 num_bytes)
3765 {
3766         int ret;
3767
3768         if (num_bytes == 0)
3769                 return 0;
3770
3771         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3772         if (!ret) {
3773                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3774                 return 0;
3775         }
3776
3777         return ret;
3778 }
3779
3780 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3781                           struct btrfs_root *root,
3782                           struct btrfs_block_rsv *block_rsv,
3783                           u64 min_reserved, int min_factor)
3784 {
3785         u64 num_bytes = 0;
3786         int commit_trans = 0;
3787         int ret = -ENOSPC;
3788
3789         if (!block_rsv)
3790                 return 0;
3791
3792         spin_lock(&block_rsv->lock);
3793         if (min_factor > 0)
3794                 num_bytes = div_factor(block_rsv->size, min_factor);
3795         if (min_reserved > num_bytes)
3796                 num_bytes = min_reserved;
3797
3798         if (block_rsv->reserved >= num_bytes) {
3799                 ret = 0;
3800         } else {
3801                 num_bytes -= block_rsv->reserved;
3802                 if (block_rsv->durable &&
3803                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3804                         commit_trans = 1;
3805         }
3806         spin_unlock(&block_rsv->lock);
3807         if (!ret)
3808                 return 0;
3809
3810         if (block_rsv->refill_used) {
3811                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3812                                              num_bytes, 0);
3813                 if (!ret) {
3814                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3815                         return 0;
3816                 }
3817         }
3818
3819         if (commit_trans) {
3820                 if (trans)
3821                         return -EAGAIN;
3822
3823                 trans = btrfs_join_transaction(root);
3824                 BUG_ON(IS_ERR(trans));
3825                 ret = btrfs_commit_transaction(trans, root);
3826                 return 0;
3827         }
3828
3829         return -ENOSPC;
3830 }
3831
3832 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3833                             struct btrfs_block_rsv *dst_rsv,
3834                             u64 num_bytes)
3835 {
3836         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3837 }
3838
3839 void btrfs_block_rsv_release(struct btrfs_root *root,
3840                              struct btrfs_block_rsv *block_rsv,
3841                              u64 num_bytes)
3842 {
3843         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3844         if (global_rsv->full || global_rsv == block_rsv ||
3845             block_rsv->space_info != global_rsv->space_info)
3846                 global_rsv = NULL;
3847         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3848 }
3849
3850 /*
3851  * helper to calculate size of global block reservation.
3852  * the desired value is sum of space used by extent tree,
3853  * checksum tree and root tree
3854  */
3855 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3856 {
3857         struct btrfs_space_info *sinfo;
3858         u64 num_bytes;
3859         u64 meta_used;
3860         u64 data_used;
3861         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3862 #if 0
3863         /*
3864          * per tree used space accounting can be inaccuracy, so we
3865          * can't rely on it.
3866          */
3867         spin_lock(&fs_info->extent_root->accounting_lock);
3868         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3869         spin_unlock(&fs_info->extent_root->accounting_lock);
3870
3871         spin_lock(&fs_info->csum_root->accounting_lock);
3872         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3873         spin_unlock(&fs_info->csum_root->accounting_lock);
3874
3875         spin_lock(&fs_info->tree_root->accounting_lock);
3876         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3877         spin_unlock(&fs_info->tree_root->accounting_lock);
3878 #endif
3879         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3880         spin_lock(&sinfo->lock);
3881         data_used = sinfo->bytes_used;
3882         spin_unlock(&sinfo->lock);
3883
3884         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3885         spin_lock(&sinfo->lock);
3886         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3887                 data_used = 0;
3888         meta_used = sinfo->bytes_used;
3889         spin_unlock(&sinfo->lock);
3890
3891         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3892                     csum_size * 2;
3893         num_bytes += div64_u64(data_used + meta_used, 50);
3894
3895         if (num_bytes * 3 > meta_used)
3896                 num_bytes = div64_u64(meta_used, 3);
3897
3898         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3899 }
3900
3901 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3902 {
3903         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3904         struct btrfs_space_info *sinfo = block_rsv->space_info;
3905         u64 num_bytes;
3906
3907         num_bytes = calc_global_metadata_size(fs_info);
3908
3909         spin_lock(&block_rsv->lock);
3910         spin_lock(&sinfo->lock);
3911
3912         block_rsv->size = num_bytes;
3913
3914         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3915                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3916                     sinfo->bytes_may_use;
3917
3918         if (sinfo->total_bytes > num_bytes) {
3919                 num_bytes = sinfo->total_bytes - num_bytes;
3920                 block_rsv->reserved += num_bytes;
3921                 sinfo->bytes_reserved += num_bytes;
3922         }
3923
3924         if (block_rsv->reserved >= block_rsv->size) {
3925                 num_bytes = block_rsv->reserved - block_rsv->size;
3926                 sinfo->bytes_reserved -= num_bytes;
3927                 sinfo->reservation_progress++;
3928                 block_rsv->reserved = block_rsv->size;
3929                 block_rsv->full = 1;
3930         }
3931 #if 0
3932         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3933                 block_rsv->size, block_rsv->reserved);
3934 #endif
3935         spin_unlock(&sinfo->lock);
3936         spin_unlock(&block_rsv->lock);
3937 }
3938
3939 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3940 {
3941         struct btrfs_space_info *space_info;
3942
3943         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3944         fs_info->chunk_block_rsv.space_info = space_info;
3945         fs_info->chunk_block_rsv.priority = 10;
3946
3947         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3948         fs_info->global_block_rsv.space_info = space_info;
3949         fs_info->global_block_rsv.priority = 10;
3950         fs_info->global_block_rsv.refill_used = 1;
3951         fs_info->delalloc_block_rsv.space_info = space_info;
3952         fs_info->trans_block_rsv.space_info = space_info;
3953         fs_info->empty_block_rsv.space_info = space_info;
3954         fs_info->empty_block_rsv.priority = 10;
3955
3956         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3957         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3958         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3959         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3960         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3961
3962         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3963
3964         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3965
3966         update_global_block_rsv(fs_info);
3967 }
3968
3969 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3970 {
3971         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3972         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3973         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3974         WARN_ON(fs_info->trans_block_rsv.size > 0);
3975         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3976         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3977         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3978 }
3979
3980 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3981 {
3982         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3983                 3 * num_items;
3984 }
3985
3986 int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle *trans,
3987                                     struct btrfs_root *root,
3988                                     struct btrfs_block_rsv *rsv)
3989 {
3990         struct btrfs_block_rsv *trans_rsv = &root->fs_info->trans_block_rsv;
3991         u64 num_bytes;
3992         int ret;
3993
3994         /*
3995          * Truncate should be freeing data, but give us 2 items just in case it
3996          * needs to use some space.  We may want to be smarter about this in the
3997          * future.
3998          */
3999         num_bytes = calc_trans_metadata_size(root, 2);
4000
4001         /* We already have enough bytes, just return */
4002         if (rsv->reserved >= num_bytes)
4003                 return 0;
4004
4005         num_bytes -= rsv->reserved;
4006
4007         /*
4008          * You should have reserved enough space before hand to do this, so this
4009          * should not fail.
4010          */
4011         ret = block_rsv_migrate_bytes(trans_rsv, rsv, num_bytes);
4012         BUG_ON(ret);
4013
4014         return 0;
4015 }
4016
4017 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
4018                                  struct btrfs_root *root,
4019                                  int num_items)
4020 {
4021         u64 num_bytes;
4022         int ret;
4023
4024         if (num_items == 0 || root->fs_info->chunk_root == root)
4025                 return 0;
4026
4027         num_bytes = calc_trans_metadata_size(root, num_items);
4028         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
4029                                   num_bytes);
4030         if (!ret) {
4031                 trans->bytes_reserved += num_bytes;
4032                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4033         }
4034         return ret;
4035 }
4036
4037 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4038                                   struct btrfs_root *root)
4039 {
4040         if (!trans->bytes_reserved)
4041                 return;
4042
4043         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
4044         btrfs_block_rsv_release(root, trans->block_rsv,
4045                                 trans->bytes_reserved);
4046         trans->bytes_reserved = 0;
4047 }
4048
4049 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4050                                   struct inode *inode)
4051 {
4052         struct btrfs_root *root = BTRFS_I(inode)->root;
4053         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4054         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4055
4056         /*
4057          * We need to hold space in order to delete our orphan item once we've
4058          * added it, so this takes the reservation so we can release it later
4059          * when we are truly done with the orphan item.
4060          */
4061         u64 num_bytes = calc_trans_metadata_size(root, 1);
4062         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4063 }
4064
4065 void btrfs_orphan_release_metadata(struct inode *inode)
4066 {
4067         struct btrfs_root *root = BTRFS_I(inode)->root;
4068         u64 num_bytes = calc_trans_metadata_size(root, 1);
4069         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4070 }
4071
4072 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
4073                                 struct btrfs_pending_snapshot *pending)
4074 {
4075         struct btrfs_root *root = pending->root;
4076         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4077         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
4078         /*
4079          * two for root back/forward refs, two for directory entries
4080          * and one for root of the snapshot.
4081          */
4082         u64 num_bytes = calc_trans_metadata_size(root, 5);
4083         dst_rsv->space_info = src_rsv->space_info;
4084         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4085 }
4086
4087 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
4088 {
4089         return num_bytes >>= 3;
4090 }
4091
4092 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
4093 {
4094         struct btrfs_root *root = BTRFS_I(inode)->root;
4095         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
4096         u64 to_reserve;
4097         int nr_extents;
4098         int reserved_extents;
4099         int ret;
4100
4101         if (btrfs_transaction_in_commit(root->fs_info))
4102                 schedule_timeout(1);
4103
4104         num_bytes = ALIGN(num_bytes, root->sectorsize);
4105
4106         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4107         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4108
4109         if (nr_extents > reserved_extents) {
4110                 nr_extents -= reserved_extents;
4111                 to_reserve = calc_trans_metadata_size(root, nr_extents);
4112         } else {
4113                 nr_extents = 0;
4114                 to_reserve = 0;
4115         }
4116
4117         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4118         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4119         if (ret)
4120                 return ret;
4121
4122         atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
4123         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4124
4125         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4126
4127         if (block_rsv->size > 512 * 1024 * 1024)
4128                 shrink_delalloc(NULL, root, to_reserve, 0);
4129
4130         return 0;
4131 }
4132
4133 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4134 {
4135         struct btrfs_root *root = BTRFS_I(inode)->root;
4136         u64 to_free;
4137         int nr_extents;
4138         int reserved_extents;
4139
4140         num_bytes = ALIGN(num_bytes, root->sectorsize);
4141         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4142         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4143
4144         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4145         do {
4146                 int old, new;
4147
4148                 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4149                 if (nr_extents >= reserved_extents) {
4150                         nr_extents = 0;
4151                         break;
4152                 }
4153                 old = reserved_extents;
4154                 nr_extents = reserved_extents - nr_extents;
4155                 new = reserved_extents - nr_extents;
4156                 old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
4157                                      reserved_extents, new);
4158                 if (likely(old == reserved_extents))
4159                         break;
4160                 reserved_extents = old;
4161         } while (1);
4162
4163         to_free = calc_csum_metadata_size(inode, num_bytes);
4164         if (nr_extents > 0)
4165                 to_free += calc_trans_metadata_size(root, nr_extents);
4166
4167         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4168                                 to_free);
4169 }
4170
4171 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4172 {
4173         int ret;
4174
4175         ret = btrfs_check_data_free_space(inode, num_bytes);
4176         if (ret)
4177                 return ret;
4178
4179         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4180         if (ret) {
4181                 btrfs_free_reserved_data_space(inode, num_bytes);
4182                 return ret;
4183         }
4184
4185         return 0;
4186 }
4187
4188 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4189 {
4190         btrfs_delalloc_release_metadata(inode, num_bytes);
4191         btrfs_free_reserved_data_space(inode, num_bytes);
4192 }
4193
4194 static int update_block_group(struct btrfs_trans_handle *trans,
4195                               struct btrfs_root *root,
4196                               u64 bytenr, u64 num_bytes, int alloc)
4197 {
4198         struct btrfs_block_group_cache *cache = NULL;
4199         struct btrfs_fs_info *info = root->fs_info;
4200         u64 total = num_bytes;
4201         u64 old_val;
4202         u64 byte_in_group;
4203         int factor;
4204
4205         /* block accounting for super block */
4206         spin_lock(&info->delalloc_lock);
4207         old_val = btrfs_super_bytes_used(&info->super_copy);
4208         if (alloc)
4209                 old_val += num_bytes;
4210         else
4211                 old_val -= num_bytes;
4212         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4213         spin_unlock(&info->delalloc_lock);
4214
4215         while (total) {
4216                 cache = btrfs_lookup_block_group(info, bytenr);
4217                 if (!cache)
4218                         return -1;
4219                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4220                                     BTRFS_BLOCK_GROUP_RAID1 |
4221                                     BTRFS_BLOCK_GROUP_RAID10))
4222                         factor = 2;
4223                 else
4224                         factor = 1;
4225                 /*
4226                  * If this block group has free space cache written out, we
4227                  * need to make sure to load it if we are removing space.  This
4228                  * is because we need the unpinning stage to actually add the
4229                  * space back to the block group, otherwise we will leak space.
4230                  */
4231                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4232                         cache_block_group(cache, trans, NULL, 1);
4233
4234                 byte_in_group = bytenr - cache->key.objectid;
4235                 WARN_ON(byte_in_group > cache->key.offset);
4236
4237                 spin_lock(&cache->space_info->lock);
4238                 spin_lock(&cache->lock);
4239
4240                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4241                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4242                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4243
4244                 cache->dirty = 1;
4245                 old_val = btrfs_block_group_used(&cache->item);
4246                 num_bytes = min(total, cache->key.offset - byte_in_group);
4247                 if (alloc) {
4248                         old_val += num_bytes;
4249                         btrfs_set_block_group_used(&cache->item, old_val);
4250                         cache->reserved -= num_bytes;
4251                         cache->space_info->bytes_reserved -= num_bytes;
4252                         cache->space_info->reservation_progress++;
4253                         cache->space_info->bytes_used += num_bytes;
4254                         cache->space_info->disk_used += num_bytes * factor;
4255                         spin_unlock(&cache->lock);
4256                         spin_unlock(&cache->space_info->lock);
4257                 } else {
4258                         old_val -= num_bytes;
4259                         btrfs_set_block_group_used(&cache->item, old_val);
4260                         cache->pinned += num_bytes;
4261                         cache->space_info->bytes_pinned += num_bytes;
4262                         cache->space_info->bytes_used -= num_bytes;
4263                         cache->space_info->disk_used -= num_bytes * factor;
4264                         spin_unlock(&cache->lock);
4265                         spin_unlock(&cache->space_info->lock);
4266
4267                         set_extent_dirty(info->pinned_extents,
4268                                          bytenr, bytenr + num_bytes - 1,
4269                                          GFP_NOFS | __GFP_NOFAIL);
4270                 }
4271                 btrfs_put_block_group(cache);
4272                 total -= num_bytes;
4273                 bytenr += num_bytes;
4274         }
4275         return 0;
4276 }
4277
4278 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4279 {
4280         struct btrfs_block_group_cache *cache;
4281         u64 bytenr;
4282
4283         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4284         if (!cache)
4285                 return 0;
4286
4287         bytenr = cache->key.objectid;
4288         btrfs_put_block_group(cache);
4289
4290         return bytenr;
4291 }
4292
4293 static int pin_down_extent(struct btrfs_root *root,
4294                            struct btrfs_block_group_cache *cache,
4295                            u64 bytenr, u64 num_bytes, int reserved)
4296 {
4297         spin_lock(&cache->space_info->lock);
4298         spin_lock(&cache->lock);
4299         cache->pinned += num_bytes;
4300         cache->space_info->bytes_pinned += num_bytes;
4301         if (reserved) {
4302                 cache->reserved -= num_bytes;
4303                 cache->space_info->bytes_reserved -= num_bytes;
4304                 cache->space_info->reservation_progress++;
4305         }
4306         spin_unlock(&cache->lock);
4307         spin_unlock(&cache->space_info->lock);
4308
4309         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4310                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4311         return 0;
4312 }
4313
4314 /*
4315  * this function must be called within transaction
4316  */
4317 int btrfs_pin_extent(struct btrfs_root *root,
4318                      u64 bytenr, u64 num_bytes, int reserved)
4319 {
4320         struct btrfs_block_group_cache *cache;
4321
4322         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4323         BUG_ON(!cache);
4324
4325         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4326
4327         btrfs_put_block_group(cache);
4328         return 0;
4329 }
4330
4331 /*
4332  * update size of reserved extents. this function may return -EAGAIN
4333  * if 'reserve' is true or 'sinfo' is false.
4334  */
4335 int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
4336                                 u64 num_bytes, int reserve, int sinfo)
4337 {
4338         int ret = 0;
4339         if (sinfo) {
4340                 struct btrfs_space_info *space_info = cache->space_info;
4341                 spin_lock(&space_info->lock);
4342                 spin_lock(&cache->lock);
4343                 if (reserve) {
4344                         if (cache->ro) {
4345                                 ret = -EAGAIN;
4346                         } else {
4347                                 cache->reserved += num_bytes;
4348                                 space_info->bytes_reserved += num_bytes;
4349                         }
4350                 } else {
4351                         if (cache->ro)
4352                                 space_info->bytes_readonly += num_bytes;
4353                         cache->reserved -= num_bytes;
4354                         space_info->bytes_reserved -= num_bytes;
4355                         space_info->reservation_progress++;
4356                 }
4357                 spin_unlock(&cache->lock);
4358                 spin_unlock(&space_info->lock);
4359         } else {
4360                 spin_lock(&cache->lock);
4361                 if (cache->ro) {
4362                         ret = -EAGAIN;
4363                 } else {
4364                         if (reserve)
4365                                 cache->reserved += num_bytes;
4366                         else
4367                                 cache->reserved -= num_bytes;
4368                 }
4369                 spin_unlock(&cache->lock);
4370         }
4371         return ret;
4372 }
4373
4374 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4375                                 struct btrfs_root *root)
4376 {
4377         struct btrfs_fs_info *fs_info = root->fs_info;
4378         struct btrfs_caching_control *next;
4379         struct btrfs_caching_control *caching_ctl;
4380         struct btrfs_block_group_cache *cache;
4381
4382         down_write(&fs_info->extent_commit_sem);
4383
4384         list_for_each_entry_safe(caching_ctl, next,
4385                                  &fs_info->caching_block_groups, list) {
4386                 cache = caching_ctl->block_group;
4387                 if (block_group_cache_done(cache)) {
4388                         cache->last_byte_to_unpin = (u64)-1;
4389                         list_del_init(&caching_ctl->list);
4390                         put_caching_control(caching_ctl);
4391                 } else {
4392                         cache->last_byte_to_unpin = caching_ctl->progress;
4393                 }
4394         }
4395
4396         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4397                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4398         else
4399                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4400
4401         up_write(&fs_info->extent_commit_sem);
4402
4403         update_global_block_rsv(fs_info);
4404         return 0;
4405 }
4406
4407 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4408 {
4409         struct btrfs_fs_info *fs_info = root->fs_info;
4410         struct btrfs_block_group_cache *cache = NULL;
4411         u64 len;
4412
4413         while (start <= end) {
4414                 if (!cache ||
4415                     start >= cache->key.objectid + cache->key.offset) {
4416                         if (cache)
4417                                 btrfs_put_block_group(cache);
4418                         cache = btrfs_lookup_block_group(fs_info, start);
4419                         BUG_ON(!cache);
4420                 }
4421
4422                 len = cache->key.objectid + cache->key.offset - start;
4423                 len = min(len, end + 1 - start);
4424
4425                 if (start < cache->last_byte_to_unpin) {
4426                         len = min(len, cache->last_byte_to_unpin - start);
4427                         btrfs_add_free_space(cache, start, len);
4428                 }
4429
4430                 start += len;
4431
4432                 spin_lock(&cache->space_info->lock);
4433                 spin_lock(&cache->lock);
4434                 cache->pinned -= len;
4435                 cache->space_info->bytes_pinned -= len;
4436                 if (cache->ro) {
4437                         cache->space_info->bytes_readonly += len;
4438                 } else if (cache->reserved_pinned > 0) {
4439                         len = min(len, cache->reserved_pinned);
4440                         cache->reserved_pinned -= len;
4441                         cache->space_info->bytes_reserved += len;
4442                 }
4443                 spin_unlock(&cache->lock);
4444                 spin_unlock(&cache->space_info->lock);
4445         }
4446
4447         if (cache)
4448                 btrfs_put_block_group(cache);
4449         return 0;
4450 }
4451
4452 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4453                                struct btrfs_root *root)
4454 {
4455         struct btrfs_fs_info *fs_info = root->fs_info;
4456         struct extent_io_tree *unpin;
4457         struct btrfs_block_rsv *block_rsv;
4458         struct btrfs_block_rsv *next_rsv;
4459         u64 start;
4460         u64 end;
4461         int idx;
4462         int ret;
4463
4464         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4465                 unpin = &fs_info->freed_extents[1];
4466         else
4467                 unpin = &fs_info->freed_extents[0];
4468
4469         while (1) {
4470                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4471                                             EXTENT_DIRTY);
4472                 if (ret)
4473                         break;
4474
4475                 if (btrfs_test_opt(root, DISCARD))
4476                         ret = btrfs_discard_extent(root, start,
4477                                                    end + 1 - start, NULL);
4478
4479                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4480                 unpin_extent_range(root, start, end);
4481                 cond_resched();
4482         }
4483
4484         mutex_lock(&fs_info->durable_block_rsv_mutex);
4485         list_for_each_entry_safe(block_rsv, next_rsv,
4486                                  &fs_info->durable_block_rsv_list, list) {
4487
4488                 idx = trans->transid & 0x1;
4489                 if (block_rsv->freed[idx] > 0) {
4490                         block_rsv_add_bytes(block_rsv,
4491                                             block_rsv->freed[idx], 0);
4492                         block_rsv->freed[idx] = 0;
4493                 }
4494                 if (atomic_read(&block_rsv->usage) == 0) {
4495                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4496
4497                         if (block_rsv->freed[0] == 0 &&
4498                             block_rsv->freed[1] == 0) {
4499                                 list_del_init(&block_rsv->list);
4500                                 kfree(block_rsv);
4501                         }
4502                 } else {
4503                         btrfs_block_rsv_release(root, block_rsv, 0);
4504                 }
4505         }
4506         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4507
4508         return 0;
4509 }
4510
4511 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4512                                 struct btrfs_root *root,
4513                                 u64 bytenr, u64 num_bytes, u64 parent,
4514                                 u64 root_objectid, u64 owner_objectid,
4515                                 u64 owner_offset, int refs_to_drop,
4516                                 struct btrfs_delayed_extent_op *extent_op)
4517 {
4518         struct btrfs_key key;
4519         struct btrfs_path *path;
4520         struct btrfs_fs_info *info = root->fs_info;
4521         struct btrfs_root *extent_root = info->extent_root;
4522         struct extent_buffer *leaf;
4523         struct btrfs_extent_item *ei;
4524         struct btrfs_extent_inline_ref *iref;
4525         int ret;
4526         int is_data;
4527         int extent_slot = 0;
4528         int found_extent = 0;
4529         int num_to_del = 1;
4530         u32 item_size;
4531         u64 refs;
4532
4533         path = btrfs_alloc_path();
4534         if (!path)
4535                 return -ENOMEM;
4536
4537         path->reada = 1;
4538         path->leave_spinning = 1;
4539
4540         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4541         BUG_ON(!is_data && refs_to_drop != 1);
4542
4543         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4544                                     bytenr, num_bytes, parent,
4545                                     root_objectid, owner_objectid,
4546                                     owner_offset);
4547         if (ret == 0) {
4548                 extent_slot = path->slots[0];
4549                 while (extent_slot >= 0) {
4550                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4551                                               extent_slot);
4552                         if (key.objectid != bytenr)
4553                                 break;
4554                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4555                             key.offset == num_bytes) {
4556                                 found_extent = 1;
4557                                 break;
4558                         }
4559                         if (path->slots[0] - extent_slot > 5)
4560                                 break;
4561                         extent_slot--;
4562                 }
4563 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4564                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4565                 if (found_extent && item_size < sizeof(*ei))
4566                         found_extent = 0;
4567 #endif
4568                 if (!found_extent) {
4569                         BUG_ON(iref);
4570                         ret = remove_extent_backref(trans, extent_root, path,
4571                                                     NULL, refs_to_drop,
4572                                                     is_data);
4573                         BUG_ON(ret);
4574                         btrfs_release_path(extent_root, path);
4575                         path->leave_spinning = 1;
4576
4577                         key.objectid = bytenr;
4578                         key.type = BTRFS_EXTENT_ITEM_KEY;
4579                         key.offset = num_bytes;
4580
4581                         ret = btrfs_search_slot(trans, extent_root,
4582                                                 &key, path, -1, 1);
4583                         if (ret) {
4584                                 printk(KERN_ERR "umm, got %d back from search"
4585                                        ", was looking for %llu\n", ret,
4586                                        (unsigned long long)bytenr);
4587                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4588                         }
4589                         BUG_ON(ret);
4590                         extent_slot = path->slots[0];
4591                 }
4592         } else {
4593                 btrfs_print_leaf(extent_root, path->nodes[0]);
4594                 WARN_ON(1);
4595                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4596                        "parent %llu root %llu  owner %llu offset %llu\n",
4597                        (unsigned long long)bytenr,
4598                        (unsigned long long)parent,
4599                        (unsigned long long)root_objectid,
4600                        (unsigned long long)owner_objectid,
4601                        (unsigned long long)owner_offset);
4602         }
4603
4604         leaf = path->nodes[0];
4605         item_size = btrfs_item_size_nr(leaf, extent_slot);
4606 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4607         if (item_size < sizeof(*ei)) {
4608                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4609                 ret = convert_extent_item_v0(trans, extent_root, path,
4610                                              owner_objectid, 0);
4611                 BUG_ON(ret < 0);
4612
4613                 btrfs_release_path(extent_root, path);
4614                 path->leave_spinning = 1;
4615
4616                 key.objectid = bytenr;
4617                 key.type = BTRFS_EXTENT_ITEM_KEY;
4618                 key.offset = num_bytes;
4619
4620                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4621                                         -1, 1);
4622                 if (ret) {
4623                         printk(KERN_ERR "umm, got %d back from search"
4624                                ", was looking for %llu\n", ret,
4625                                (unsigned long long)bytenr);
4626                         btrfs_print_leaf(extent_root, path->nodes[0]);
4627                 }
4628                 BUG_ON(ret);
4629                 extent_slot = path->slots[0];
4630                 leaf = path->nodes[0];
4631                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4632         }
4633 #endif
4634         BUG_ON(item_size < sizeof(*ei));
4635         ei = btrfs_item_ptr(leaf, extent_slot,
4636                             struct btrfs_extent_item);
4637         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4638                 struct btrfs_tree_block_info *bi;
4639                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4640                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4641                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4642         }
4643
4644         refs = btrfs_extent_refs(leaf, ei);
4645         BUG_ON(refs < refs_to_drop);
4646         refs -= refs_to_drop;
4647
4648         if (refs > 0) {
4649                 if (extent_op)
4650                         __run_delayed_extent_op(extent_op, leaf, ei);
4651                 /*
4652                  * In the case of inline back ref, reference count will
4653                  * be updated by remove_extent_backref
4654                  */
4655                 if (iref) {
4656                         BUG_ON(!found_extent);
4657                 } else {
4658                         btrfs_set_extent_refs(leaf, ei, refs);
4659                         btrfs_mark_buffer_dirty(leaf);
4660                 }
4661                 if (found_extent) {
4662                         ret = remove_extent_backref(trans, extent_root, path,
4663                                                     iref, refs_to_drop,
4664                                                     is_data);
4665                         BUG_ON(ret);
4666                 }
4667         } else {
4668                 if (found_extent) {
4669                         BUG_ON(is_data && refs_to_drop !=
4670                                extent_data_ref_count(root, path, iref));
4671                         if (iref) {
4672                                 BUG_ON(path->slots[0] != extent_slot);
4673                         } else {
4674                                 BUG_ON(path->slots[0] != extent_slot + 1);
4675                                 path->slots[0] = extent_slot;
4676                                 num_to_del = 2;
4677                         }
4678                 }
4679
4680                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4681                                       num_to_del);
4682                 BUG_ON(ret);
4683                 btrfs_release_path(extent_root, path);
4684
4685                 if (is_data) {
4686                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4687                         BUG_ON(ret);
4688                 } else {
4689                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4690                              bytenr >> PAGE_CACHE_SHIFT,
4691                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4692                 }
4693
4694                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4695                 BUG_ON(ret);
4696         }
4697         btrfs_free_path(path);
4698         return ret;
4699 }
4700
4701 /*
4702  * when we free an block, it is possible (and likely) that we free the last
4703  * delayed ref for that extent as well.  This searches the delayed ref tree for
4704  * a given extent, and if there are no other delayed refs to be processed, it
4705  * removes it from the tree.
4706  */
4707 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4708                                       struct btrfs_root *root, u64 bytenr)
4709 {
4710         struct btrfs_delayed_ref_head *head;
4711         struct btrfs_delayed_ref_root *delayed_refs;
4712         struct btrfs_delayed_ref_node *ref;
4713         struct rb_node *node;
4714         int ret = 0;
4715
4716         delayed_refs = &trans->transaction->delayed_refs;
4717         spin_lock(&delayed_refs->lock);
4718         head = btrfs_find_delayed_ref_head(trans, bytenr);
4719         if (!head)
4720                 goto out;
4721
4722         node = rb_prev(&head->node.rb_node);
4723         if (!node)
4724                 goto out;
4725
4726         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4727
4728         /* there are still entries for this ref, we can't drop it */
4729         if (ref->bytenr == bytenr)
4730                 goto out;
4731
4732         if (head->extent_op) {
4733                 if (!head->must_insert_reserved)
4734                         goto out;
4735                 kfree(head->extent_op);
4736                 head->extent_op = NULL;
4737         }
4738
4739         /*
4740          * waiting for the lock here would deadlock.  If someone else has it
4741          * locked they are already in the process of dropping it anyway
4742          */
4743         if (!mutex_trylock(&head->mutex))
4744                 goto out;
4745
4746         /*
4747          * at this point we have a head with no other entries.  Go
4748          * ahead and process it.
4749          */
4750         head->node.in_tree = 0;
4751         rb_erase(&head->node.rb_node, &delayed_refs->root);
4752
4753         delayed_refs->num_entries--;
4754
4755         /*
4756          * we don't take a ref on the node because we're removing it from the
4757          * tree, so we just steal the ref the tree was holding.
4758          */
4759         delayed_refs->num_heads--;
4760         if (list_empty(&head->cluster))
4761                 delayed_refs->num_heads_ready--;
4762
4763         list_del_init(&head->cluster);
4764         spin_unlock(&delayed_refs->lock);
4765
4766         BUG_ON(head->extent_op);
4767         if (head->must_insert_reserved)
4768                 ret = 1;
4769
4770         mutex_unlock(&head->mutex);
4771         btrfs_put_delayed_ref(&head->node);
4772         return ret;
4773 out:
4774         spin_unlock(&delayed_refs->lock);
4775         return 0;
4776 }
4777
4778 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4779                            struct btrfs_root *root,
4780                            struct extent_buffer *buf,
4781                            u64 parent, int last_ref)
4782 {
4783         struct btrfs_block_rsv *block_rsv;
4784         struct btrfs_block_group_cache *cache = NULL;
4785         int ret;
4786
4787         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4788                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4789                                                 parent, root->root_key.objectid,
4790                                                 btrfs_header_level(buf),
4791                                                 BTRFS_DROP_DELAYED_REF, NULL);
4792                 BUG_ON(ret);
4793         }
4794
4795         if (!last_ref)
4796                 return;
4797
4798         block_rsv = get_block_rsv(trans, root);
4799         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4800         if (block_rsv->space_info != cache->space_info)
4801                 goto out;
4802
4803         if (btrfs_header_generation(buf) == trans->transid) {
4804                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4805                         ret = check_ref_cleanup(trans, root, buf->start);
4806                         if (!ret)
4807                                 goto pin;
4808                 }
4809
4810                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4811                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4812                         goto pin;
4813                 }
4814
4815                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4816
4817                 btrfs_add_free_space(cache, buf->start, buf->len);
4818                 ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0);
4819                 if (ret == -EAGAIN) {
4820                         /* block group became read-only */
4821                         btrfs_update_reserved_bytes(cache, buf->len, 0, 1);
4822                         goto out;
4823                 }
4824
4825                 ret = 1;
4826                 spin_lock(&block_rsv->lock);
4827                 if (block_rsv->reserved < block_rsv->size) {
4828                         block_rsv->reserved += buf->len;
4829                         ret = 0;
4830                 }
4831                 spin_unlock(&block_rsv->lock);
4832
4833                 if (ret) {
4834                         spin_lock(&cache->space_info->lock);
4835                         cache->space_info->bytes_reserved -= buf->len;
4836                         cache->space_info->reservation_progress++;
4837                         spin_unlock(&cache->space_info->lock);
4838                 }
4839                 goto out;
4840         }
4841 pin:
4842         if (block_rsv->durable && !cache->ro) {
4843                 ret = 0;
4844                 spin_lock(&cache->lock);
4845                 if (!cache->ro) {
4846                         cache->reserved_pinned += buf->len;
4847                         ret = 1;
4848                 }
4849                 spin_unlock(&cache->lock);
4850
4851                 if (ret) {
4852                         spin_lock(&block_rsv->lock);
4853                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4854                         spin_unlock(&block_rsv->lock);
4855                 }
4856         }
4857 out:
4858         /*
4859          * Deleting the buffer, clear the corrupt flag since it doesn't matter
4860          * anymore.
4861          */
4862         clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
4863         btrfs_put_block_group(cache);
4864 }
4865
4866 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4867                       struct btrfs_root *root,
4868                       u64 bytenr, u64 num_bytes, u64 parent,
4869                       u64 root_objectid, u64 owner, u64 offset)
4870 {
4871         int ret;
4872
4873         /*
4874          * tree log blocks never actually go into the extent allocation
4875          * tree, just update pinning info and exit early.
4876          */
4877         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4878                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4879                 /* unlocks the pinned mutex */
4880                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4881                 ret = 0;
4882         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4883                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4884                                         parent, root_objectid, (int)owner,
4885                                         BTRFS_DROP_DELAYED_REF, NULL);
4886                 BUG_ON(ret);
4887         } else {
4888                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4889                                         parent, root_objectid, owner,
4890                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4891                 BUG_ON(ret);
4892         }
4893         return ret;
4894 }
4895
4896 static u64 stripe_align(struct btrfs_root *root, u64 val)
4897 {
4898         u64 mask = ((u64)root->stripesize - 1);
4899         u64 ret = (val + mask) & ~mask;
4900         return ret;
4901 }
4902
4903 /*
4904  * when we wait for progress in the block group caching, its because
4905  * our allocation attempt failed at least once.  So, we must sleep
4906  * and let some progress happen before we try again.
4907  *
4908  * This function will sleep at least once waiting for new free space to
4909  * show up, and then it will check the block group free space numbers
4910  * for our min num_bytes.  Another option is to have it go ahead
4911  * and look in the rbtree for a free extent of a given size, but this
4912  * is a good start.
4913  */
4914 static noinline int
4915 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4916                                 u64 num_bytes)
4917 {
4918         struct btrfs_caching_control *caching_ctl;
4919         DEFINE_WAIT(wait);
4920
4921         caching_ctl = get_caching_control(cache);
4922         if (!caching_ctl)
4923                 return 0;
4924
4925         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4926                    (cache->free_space >= num_bytes));
4927
4928         put_caching_control(caching_ctl);
4929         return 0;
4930 }
4931
4932 static noinline int
4933 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4934 {
4935         struct btrfs_caching_control *caching_ctl;
4936         DEFINE_WAIT(wait);
4937
4938         caching_ctl = get_caching_control(cache);
4939         if (!caching_ctl)
4940                 return 0;
4941
4942         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4943
4944         put_caching_control(caching_ctl);
4945         return 0;
4946 }
4947
4948 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4949 {
4950         int index;
4951         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4952                 index = 0;
4953         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4954                 index = 1;
4955         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4956                 index = 2;
4957         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4958                 index = 3;
4959         else
4960                 index = 4;
4961         return index;
4962 }
4963
4964 enum btrfs_loop_type {
4965         LOOP_FIND_IDEAL = 0,
4966         LOOP_CACHING_NOWAIT = 1,
4967         LOOP_CACHING_WAIT = 2,
4968         LOOP_ALLOC_CHUNK = 3,
4969         LOOP_NO_EMPTY_SIZE = 4,
4970 };
4971
4972 /*
4973  * walks the btree of allocated extents and find a hole of a given size.
4974  * The key ins is changed to record the hole:
4975  * ins->objectid == block start
4976  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4977  * ins->offset == number of blocks
4978  * Any available blocks before search_start are skipped.
4979  */
4980 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4981                                      struct btrfs_root *orig_root,
4982                                      u64 num_bytes, u64 empty_size,
4983                                      u64 search_start, u64 search_end,
4984                                      u64 hint_byte, struct btrfs_key *ins,
4985                                      int data)
4986 {
4987         int ret = 0;
4988         struct btrfs_root *root = orig_root->fs_info->extent_root;
4989         struct btrfs_free_cluster *last_ptr = NULL;
4990         struct btrfs_block_group_cache *block_group = NULL;
4991         int empty_cluster = 2 * 1024 * 1024;
4992         int allowed_chunk_alloc = 0;
4993         int done_chunk_alloc = 0;
4994         struct btrfs_space_info *space_info;
4995         int last_ptr_loop = 0;
4996         int loop = 0;
4997         int index = 0;
4998         bool found_uncached_bg = false;
4999         bool failed_cluster_refill = false;
5000         bool failed_alloc = false;
5001         bool use_cluster = true;
5002         u64 ideal_cache_percent = 0;
5003         u64 ideal_cache_offset = 0;
5004
5005         WARN_ON(num_bytes < root->sectorsize);
5006         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
5007         ins->objectid = 0;
5008         ins->offset = 0;
5009
5010         space_info = __find_space_info(root->fs_info, data);
5011         if (!space_info) {
5012                 printk(KERN_ERR "No space info for %d\n", data);
5013                 return -ENOSPC;
5014         }
5015
5016         /*
5017          * If the space info is for both data and metadata it means we have a
5018          * small filesystem and we can't use the clustering stuff.
5019          */
5020         if (btrfs_mixed_space_info(space_info))
5021                 use_cluster = false;
5022
5023         if (orig_root->ref_cows || empty_size)
5024                 allowed_chunk_alloc = 1;
5025
5026         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
5027                 last_ptr = &root->fs_info->meta_alloc_cluster;
5028                 if (!btrfs_test_opt(root, SSD))
5029                         empty_cluster = 64 * 1024;
5030         }
5031
5032         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
5033             btrfs_test_opt(root, SSD)) {
5034                 last_ptr = &root->fs_info->data_alloc_cluster;
5035         }
5036
5037         if (last_ptr) {
5038                 spin_lock(&last_ptr->lock);
5039                 if (last_ptr->block_group)
5040                         hint_byte = last_ptr->window_start;
5041                 spin_unlock(&last_ptr->lock);
5042         }
5043
5044         search_start = max(search_start, first_logical_byte(root, 0));
5045         search_start = max(search_start, hint_byte);
5046
5047         if (!last_ptr)
5048                 empty_cluster = 0;
5049
5050         if (search_start == hint_byte) {
5051 ideal_cache:
5052                 block_group = btrfs_lookup_block_group(root->fs_info,
5053                                                        search_start);
5054                 /*
5055                  * we don't want to use the block group if it doesn't match our
5056                  * allocation bits, or if its not cached.
5057                  *
5058                  * However if we are re-searching with an ideal block group
5059                  * picked out then we don't care that the block group is cached.
5060                  */
5061                 if (block_group && block_group_bits(block_group, data) &&
5062                     (block_group->cached != BTRFS_CACHE_NO ||
5063                      search_start == ideal_cache_offset)) {
5064                         down_read(&space_info->groups_sem);
5065                         if (list_empty(&block_group->list) ||
5066                             block_group->ro) {
5067                                 /*
5068                                  * someone is removing this block group,
5069                                  * we can't jump into the have_block_group
5070                                  * target because our list pointers are not
5071                                  * valid
5072                                  */
5073                                 btrfs_put_block_group(block_group);
5074                                 up_read(&space_info->groups_sem);
5075                         } else {
5076                                 index = get_block_group_index(block_group);
5077                                 goto have_block_group;
5078                         }
5079                 } else if (block_group) {
5080                         btrfs_put_block_group(block_group);
5081                 }
5082         }
5083 search:
5084         down_read(&space_info->groups_sem);
5085         list_for_each_entry(block_group, &space_info->block_groups[index],
5086                             list) {
5087                 u64 offset;
5088                 int cached;
5089
5090                 btrfs_get_block_group(block_group);
5091                 search_start = block_group->key.objectid;
5092
5093                 /*
5094                  * this can happen if we end up cycling through all the
5095                  * raid types, but we want to make sure we only allocate
5096                  * for the proper type.
5097                  */
5098                 if (!block_group_bits(block_group, data)) {
5099                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
5100                                 BTRFS_BLOCK_GROUP_RAID1 |
5101                                 BTRFS_BLOCK_GROUP_RAID10;
5102
5103                         /*
5104                          * if they asked for extra copies and this block group
5105                          * doesn't provide them, bail.  This does allow us to
5106                          * fill raid0 from raid1.
5107                          */
5108                         if ((data & extra) && !(block_group->flags & extra))
5109                                 goto loop;
5110                 }
5111
5112 have_block_group:
5113                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5114                         u64 free_percent;
5115
5116                         ret = cache_block_group(block_group, trans,
5117                                                 orig_root, 1);
5118                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5119                                 goto have_block_group;
5120
5121                         free_percent = btrfs_block_group_used(&block_group->item);
5122                         free_percent *= 100;
5123                         free_percent = div64_u64(free_percent,
5124                                                  block_group->key.offset);
5125                         free_percent = 100 - free_percent;
5126                         if (free_percent > ideal_cache_percent &&
5127                             likely(!block_group->ro)) {
5128                                 ideal_cache_offset = block_group->key.objectid;
5129                                 ideal_cache_percent = free_percent;
5130                         }
5131
5132                         /*
5133                          * We only want to start kthread caching if we are at
5134                          * the point where we will wait for caching to make
5135                          * progress, or if our ideal search is over and we've
5136                          * found somebody to start caching.
5137                          */
5138                         if (loop > LOOP_CACHING_NOWAIT ||
5139                             (loop > LOOP_FIND_IDEAL &&
5140                              atomic_read(&space_info->caching_threads) < 2)) {
5141                                 ret = cache_block_group(block_group, trans,
5142                                                         orig_root, 0);
5143                                 BUG_ON(ret);
5144                         }
5145                         found_uncached_bg = true;
5146
5147                         /*
5148                          * If loop is set for cached only, try the next block
5149                          * group.
5150                          */
5151                         if (loop == LOOP_FIND_IDEAL)
5152                                 goto loop;
5153                 }
5154
5155                 cached = block_group_cache_done(block_group);
5156                 if (unlikely(!cached))
5157                         found_uncached_bg = true;
5158
5159                 if (unlikely(block_group->ro))
5160                         goto loop;
5161
5162                 /*
5163                  * Ok we want to try and use the cluster allocator, so lets look
5164                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5165                  * have tried the cluster allocator plenty of times at this
5166                  * point and not have found anything, so we are likely way too
5167                  * fragmented for the clustering stuff to find anything, so lets
5168                  * just skip it and let the allocator find whatever block it can
5169                  * find
5170                  */
5171                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5172                         /*
5173                          * the refill lock keeps out other
5174                          * people trying to start a new cluster
5175                          */
5176                         spin_lock(&last_ptr->refill_lock);
5177                         if (last_ptr->block_group &&
5178                             (last_ptr->block_group->ro ||
5179                             !block_group_bits(last_ptr->block_group, data))) {
5180                                 offset = 0;
5181                                 goto refill_cluster;
5182                         }
5183
5184                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5185                                                  num_bytes, search_start);
5186                         if (offset) {
5187                                 /* we have a block, we're done */
5188                                 spin_unlock(&last_ptr->refill_lock);
5189                                 goto checks;
5190                         }
5191
5192                         spin_lock(&last_ptr->lock);
5193                         /*
5194                          * whoops, this cluster doesn't actually point to
5195                          * this block group.  Get a ref on the block
5196                          * group is does point to and try again
5197                          */
5198                         if (!last_ptr_loop && last_ptr->block_group &&
5199                             last_ptr->block_group != block_group) {
5200
5201                                 btrfs_put_block_group(block_group);
5202                                 block_group = last_ptr->block_group;
5203                                 btrfs_get_block_group(block_group);
5204                                 spin_unlock(&last_ptr->lock);
5205                                 spin_unlock(&last_ptr->refill_lock);
5206
5207                                 last_ptr_loop = 1;
5208                                 search_start = block_group->key.objectid;
5209                                 /*
5210                                  * we know this block group is properly
5211                                  * in the list because
5212                                  * btrfs_remove_block_group, drops the
5213                                  * cluster before it removes the block
5214                                  * group from the list
5215                                  */
5216                                 goto have_block_group;
5217                         }
5218                         spin_unlock(&last_ptr->lock);
5219 refill_cluster:
5220                         /*
5221                          * this cluster didn't work out, free it and
5222                          * start over
5223                          */
5224                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5225
5226                         last_ptr_loop = 0;
5227
5228                         /* allocate a cluster in this block group */
5229                         ret = btrfs_find_space_cluster(trans, root,
5230                                                block_group, last_ptr,
5231                                                offset, num_bytes,
5232                                                empty_cluster + empty_size);
5233                         if (ret == 0) {
5234                                 /*
5235                                  * now pull our allocation out of this
5236                                  * cluster
5237                                  */
5238                                 offset = btrfs_alloc_from_cluster(block_group,
5239                                                   last_ptr, num_bytes,
5240                                                   search_start);
5241                                 if (offset) {
5242                                         /* we found one, proceed */
5243                                         spin_unlock(&last_ptr->refill_lock);
5244                                         goto checks;
5245                                 }
5246                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5247                                    && !failed_cluster_refill) {
5248                                 spin_unlock(&last_ptr->refill_lock);
5249
5250                                 failed_cluster_refill = true;
5251                                 wait_block_group_cache_progress(block_group,
5252                                        num_bytes + empty_cluster + empty_size);
5253                                 goto have_block_group;
5254                         }
5255
5256                         /*
5257                          * at this point we either didn't find a cluster
5258                          * or we weren't able to allocate a block from our
5259                          * cluster.  Free the cluster we've been trying
5260                          * to use, and go to the next block group
5261                          */
5262                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5263                         spin_unlock(&last_ptr->refill_lock);
5264                         goto loop;
5265                 }
5266
5267                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5268                                                     num_bytes, empty_size);
5269                 /*
5270                  * If we didn't find a chunk, and we haven't failed on this
5271                  * block group before, and this block group is in the middle of
5272                  * caching and we are ok with waiting, then go ahead and wait
5273                  * for progress to be made, and set failed_alloc to true.
5274                  *
5275                  * If failed_alloc is true then we've already waited on this
5276                  * block group once and should move on to the next block group.
5277                  */
5278                 if (!offset && !failed_alloc && !cached &&
5279                     loop > LOOP_CACHING_NOWAIT) {
5280                         wait_block_group_cache_progress(block_group,
5281                                                 num_bytes + empty_size);
5282                         failed_alloc = true;
5283                         goto have_block_group;
5284                 } else if (!offset) {
5285                         goto loop;
5286                 }
5287 checks:
5288                 search_start = stripe_align(root, offset);
5289                 /* move on to the next group */
5290                 if (search_start + num_bytes >= search_end) {
5291                         btrfs_add_free_space(block_group, offset, num_bytes);
5292                         goto loop;
5293                 }
5294
5295                 /* move on to the next group */
5296                 if (search_start + num_bytes >
5297                     block_group->key.objectid + block_group->key.offset) {
5298                         btrfs_add_free_space(block_group, offset, num_bytes);
5299                         goto loop;
5300                 }
5301
5302                 ins->objectid = search_start;
5303                 ins->offset = num_bytes;
5304
5305                 if (offset < search_start)
5306                         btrfs_add_free_space(block_group, offset,
5307                                              search_start - offset);
5308                 BUG_ON(offset > search_start);
5309
5310                 ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1,
5311                                             (data & BTRFS_BLOCK_GROUP_DATA));
5312                 if (ret == -EAGAIN) {
5313                         btrfs_add_free_space(block_group, offset, num_bytes);
5314                         goto loop;
5315                 }
5316
5317                 /* we are all good, lets return */
5318                 ins->objectid = search_start;
5319                 ins->offset = num_bytes;
5320
5321                 if (offset < search_start)
5322                         btrfs_add_free_space(block_group, offset,
5323                                              search_start - offset);
5324                 BUG_ON(offset > search_start);
5325                 btrfs_put_block_group(block_group);
5326                 break;
5327 loop:
5328                 failed_cluster_refill = false;
5329                 failed_alloc = false;
5330                 BUG_ON(index != get_block_group_index(block_group));
5331                 btrfs_put_block_group(block_group);
5332         }
5333         up_read(&space_info->groups_sem);
5334
5335         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5336                 goto search;
5337
5338         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5339          *                      for them to make caching progress.  Also
5340          *                      determine the best possible bg to cache
5341          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5342          *                      caching kthreads as we move along
5343          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5344          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5345          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5346          *                      again
5347          */
5348         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5349             (found_uncached_bg || empty_size || empty_cluster ||
5350              allowed_chunk_alloc)) {
5351                 index = 0;
5352                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5353                         found_uncached_bg = false;
5354                         loop++;
5355                         if (!ideal_cache_percent &&
5356                             atomic_read(&space_info->caching_threads))
5357                                 goto search;
5358
5359                         /*
5360                          * 1 of the following 2 things have happened so far
5361                          *
5362                          * 1) We found an ideal block group for caching that
5363                          * is mostly full and will cache quickly, so we might
5364                          * as well wait for it.
5365                          *
5366                          * 2) We searched for cached only and we didn't find
5367                          * anything, and we didn't start any caching kthreads
5368                          * either, so chances are we will loop through and
5369                          * start a couple caching kthreads, and then come back
5370                          * around and just wait for them.  This will be slower
5371                          * because we will have 2 caching kthreads reading at
5372                          * the same time when we could have just started one
5373                          * and waited for it to get far enough to give us an
5374                          * allocation, so go ahead and go to the wait caching
5375                          * loop.
5376                          */
5377                         loop = LOOP_CACHING_WAIT;
5378                         search_start = ideal_cache_offset;
5379                         ideal_cache_percent = 0;
5380                         goto ideal_cache;
5381                 } else if (loop == LOOP_FIND_IDEAL) {
5382                         /*
5383                          * Didn't find a uncached bg, wait on anything we find
5384                          * next.
5385                          */
5386                         loop = LOOP_CACHING_WAIT;
5387                         goto search;
5388                 }
5389
5390                 if (loop < LOOP_CACHING_WAIT) {
5391                         loop++;
5392                         goto search;
5393                 }
5394
5395                 if (loop == LOOP_ALLOC_CHUNK) {
5396                         empty_size = 0;
5397                         empty_cluster = 0;
5398                 }
5399
5400                 if (allowed_chunk_alloc) {
5401                         ret = do_chunk_alloc(trans, root, num_bytes +
5402                                              2 * 1024 * 1024, data,
5403                                              CHUNK_ALLOC_LIMITED);
5404                         allowed_chunk_alloc = 0;
5405                         done_chunk_alloc = 1;
5406                 } else if (!done_chunk_alloc &&
5407                            space_info->force_alloc == CHUNK_ALLOC_NO_FORCE) {
5408                         space_info->force_alloc = CHUNK_ALLOC_LIMITED;
5409                 }
5410
5411                 if (loop < LOOP_NO_EMPTY_SIZE) {
5412                         loop++;
5413                         goto search;
5414                 }
5415                 ret = -ENOSPC;
5416         } else if (!ins->objectid) {
5417                 ret = -ENOSPC;
5418         } else if (ins->objectid) {
5419                 ret = 0;
5420         }
5421
5422         return ret;
5423 }
5424
5425 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5426                             int dump_block_groups)
5427 {
5428         struct btrfs_block_group_cache *cache;
5429         int index = 0;
5430
5431         spin_lock(&info->lock);
5432         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5433                (unsigned long long)(info->total_bytes - info->bytes_used -
5434                                     info->bytes_pinned - info->bytes_reserved -
5435                                     info->bytes_readonly),
5436                (info->full) ? "" : "not ");
5437         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5438                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5439                (unsigned long long)info->total_bytes,
5440                (unsigned long long)info->bytes_used,
5441                (unsigned long long)info->bytes_pinned,
5442                (unsigned long long)info->bytes_reserved,
5443                (unsigned long long)info->bytes_may_use,
5444                (unsigned long long)info->bytes_readonly);
5445         spin_unlock(&info->lock);
5446
5447         if (!dump_block_groups)
5448                 return;
5449
5450         down_read(&info->groups_sem);
5451 again:
5452         list_for_each_entry(cache, &info->block_groups[index], list) {
5453                 spin_lock(&cache->lock);
5454                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5455                        "%llu pinned %llu reserved\n",
5456                        (unsigned long long)cache->key.objectid,
5457                        (unsigned long long)cache->key.offset,
5458                        (unsigned long long)btrfs_block_group_used(&cache->item),
5459                        (unsigned long long)cache->pinned,
5460                        (unsigned long long)cache->reserved);
5461                 btrfs_dump_free_space(cache, bytes);
5462                 spin_unlock(&cache->lock);
5463         }
5464         if (++index < BTRFS_NR_RAID_TYPES)
5465                 goto again;
5466         up_read(&info->groups_sem);
5467 }
5468
5469 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5470                          struct btrfs_root *root,
5471                          u64 num_bytes, u64 min_alloc_size,
5472                          u64 empty_size, u64 hint_byte,
5473                          u64 search_end, struct btrfs_key *ins,
5474                          u64 data)
5475 {
5476         int ret;
5477         u64 search_start = 0;
5478
5479         data = btrfs_get_alloc_profile(root, data);
5480 again:
5481         /*
5482          * the only place that sets empty_size is btrfs_realloc_node, which
5483          * is not called recursively on allocations
5484          */
5485         if (empty_size || root->ref_cows)
5486                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5487                                      num_bytes + 2 * 1024 * 1024, data,
5488                                      CHUNK_ALLOC_NO_FORCE);
5489
5490         WARN_ON(num_bytes < root->sectorsize);
5491         ret = find_free_extent(trans, root, num_bytes, empty_size,
5492                                search_start, search_end, hint_byte,
5493                                ins, data);
5494
5495         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5496                 num_bytes = num_bytes >> 1;
5497                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5498                 num_bytes = max(num_bytes, min_alloc_size);
5499                 do_chunk_alloc(trans, root->fs_info->extent_root,
5500                                num_bytes, data, CHUNK_ALLOC_FORCE);
5501                 goto again;
5502         }
5503         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5504                 struct btrfs_space_info *sinfo;
5505
5506                 sinfo = __find_space_info(root->fs_info, data);
5507                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5508                        "wanted %llu\n", (unsigned long long)data,
5509                        (unsigned long long)num_bytes);
5510                 dump_space_info(sinfo, num_bytes, 1);
5511         }
5512
5513         trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
5514
5515         return ret;
5516 }
5517
5518 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5519 {
5520         struct btrfs_block_group_cache *cache;
5521         int ret = 0;
5522
5523         cache = btrfs_lookup_block_group(root->fs_info, start);
5524         if (!cache) {
5525                 printk(KERN_ERR "Unable to find block group for %llu\n",
5526                        (unsigned long long)start);
5527                 return -ENOSPC;
5528         }
5529
5530         if (btrfs_test_opt(root, DISCARD))
5531                 ret = btrfs_discard_extent(root, start, len, NULL);
5532
5533         btrfs_add_free_space(cache, start, len);
5534         btrfs_update_reserved_bytes(cache, len, 0, 1);
5535         btrfs_put_block_group(cache);
5536
5537         trace_btrfs_reserved_extent_free(root, start, len);
5538
5539         return ret;
5540 }
5541
5542 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5543                                       struct btrfs_root *root,
5544                                       u64 parent, u64 root_objectid,
5545                                       u64 flags, u64 owner, u64 offset,
5546                                       struct btrfs_key *ins, int ref_mod)
5547 {
5548         int ret;
5549         struct btrfs_fs_info *fs_info = root->fs_info;
5550         struct btrfs_extent_item *extent_item;
5551         struct btrfs_extent_inline_ref *iref;
5552         struct btrfs_path *path;
5553         struct extent_buffer *leaf;
5554         int type;
5555         u32 size;
5556
5557         if (parent > 0)
5558                 type = BTRFS_SHARED_DATA_REF_KEY;
5559         else
5560                 type = BTRFS_EXTENT_DATA_REF_KEY;
5561
5562         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5563
5564         path = btrfs_alloc_path();
5565         if (!path)
5566                 return -ENOMEM;
5567
5568         path->leave_spinning = 1;
5569         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5570                                       ins, size);
5571         BUG_ON(ret);
5572
5573         leaf = path->nodes[0];
5574         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5575                                      struct btrfs_extent_item);
5576         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5577         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5578         btrfs_set_extent_flags(leaf, extent_item,
5579                                flags | BTRFS_EXTENT_FLAG_DATA);
5580
5581         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5582         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5583         if (parent > 0) {
5584                 struct btrfs_shared_data_ref *ref;
5585                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5586                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5587                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5588         } else {
5589                 struct btrfs_extent_data_ref *ref;
5590                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5591                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5592                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5593                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5594                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5595         }
5596
5597         btrfs_mark_buffer_dirty(path->nodes[0]);
5598         btrfs_free_path(path);
5599
5600         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5601         if (ret) {
5602                 printk(KERN_ERR "btrfs update block group failed for %llu "
5603                        "%llu\n", (unsigned long long)ins->objectid,
5604                        (unsigned long long)ins->offset);
5605                 BUG();
5606         }
5607         return ret;
5608 }
5609
5610 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5611                                      struct btrfs_root *root,
5612                                      u64 parent, u64 root_objectid,
5613                                      u64 flags, struct btrfs_disk_key *key,
5614                                      int level, struct btrfs_key *ins)
5615 {
5616         int ret;
5617         struct btrfs_fs_info *fs_info = root->fs_info;
5618         struct btrfs_extent_item *extent_item;
5619         struct btrfs_tree_block_info *block_info;
5620         struct btrfs_extent_inline_ref *iref;
5621         struct btrfs_path *path;
5622         struct extent_buffer *leaf;
5623         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5624
5625         path = btrfs_alloc_path();
5626         BUG_ON(!path);
5627
5628         path->leave_spinning = 1;
5629         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5630                                       ins, size);
5631         BUG_ON(ret);
5632
5633         leaf = path->nodes[0];
5634         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5635                                      struct btrfs_extent_item);
5636         btrfs_set_extent_refs(leaf, extent_item, 1);
5637         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5638         btrfs_set_extent_flags(leaf, extent_item,
5639                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5640         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5641
5642         btrfs_set_tree_block_key(leaf, block_info, key);
5643         btrfs_set_tree_block_level(leaf, block_info, level);
5644
5645         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5646         if (parent > 0) {
5647                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5648                 btrfs_set_extent_inline_ref_type(leaf, iref,
5649                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5650                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5651         } else {
5652                 btrfs_set_extent_inline_ref_type(leaf, iref,
5653                                                  BTRFS_TREE_BLOCK_REF_KEY);
5654                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5655         }
5656
5657         btrfs_mark_buffer_dirty(leaf);
5658         btrfs_free_path(path);
5659
5660         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5661         if (ret) {
5662                 printk(KERN_ERR "btrfs update block group failed for %llu "
5663                        "%llu\n", (unsigned long long)ins->objectid,
5664                        (unsigned long long)ins->offset);
5665                 BUG();
5666         }
5667         return ret;
5668 }
5669
5670 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5671                                      struct btrfs_root *root,
5672                                      u64 root_objectid, u64 owner,
5673                                      u64 offset, struct btrfs_key *ins)
5674 {
5675         int ret;
5676
5677         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5678
5679         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5680                                          0, root_objectid, owner, offset,
5681                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5682         return ret;
5683 }
5684
5685 /*
5686  * this is used by the tree logging recovery code.  It records that
5687  * an extent has been allocated and makes sure to clear the free
5688  * space cache bits as well
5689  */
5690 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5691                                    struct btrfs_root *root,
5692                                    u64 root_objectid, u64 owner, u64 offset,
5693                                    struct btrfs_key *ins)
5694 {
5695         int ret;
5696         struct btrfs_block_group_cache *block_group;
5697         struct btrfs_caching_control *caching_ctl;
5698         u64 start = ins->objectid;
5699         u64 num_bytes = ins->offset;
5700
5701         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5702         cache_block_group(block_group, trans, NULL, 0);
5703         caching_ctl = get_caching_control(block_group);
5704
5705         if (!caching_ctl) {
5706                 BUG_ON(!block_group_cache_done(block_group));
5707                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5708                 BUG_ON(ret);
5709         } else {
5710                 mutex_lock(&caching_ctl->mutex);
5711
5712                 if (start >= caching_ctl->progress) {
5713                         ret = add_excluded_extent(root, start, num_bytes);
5714                         BUG_ON(ret);
5715                 } else if (start + num_bytes <= caching_ctl->progress) {
5716                         ret = btrfs_remove_free_space(block_group,
5717                                                       start, num_bytes);
5718                         BUG_ON(ret);
5719                 } else {
5720                         num_bytes = caching_ctl->progress - start;
5721                         ret = btrfs_remove_free_space(block_group,
5722                                                       start, num_bytes);
5723                         BUG_ON(ret);
5724
5725                         start = caching_ctl->progress;
5726                         num_bytes = ins->objectid + ins->offset -
5727                                     caching_ctl->progress;
5728                         ret = add_excluded_extent(root, start, num_bytes);
5729                         BUG_ON(ret);
5730                 }
5731
5732                 mutex_unlock(&caching_ctl->mutex);
5733                 put_caching_control(caching_ctl);
5734         }
5735
5736         ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1);
5737         BUG_ON(ret);
5738         btrfs_put_block_group(block_group);
5739         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5740                                          0, owner, offset, ins, 1);
5741         return ret;
5742 }
5743
5744 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5745                                             struct btrfs_root *root,
5746                                             u64 bytenr, u32 blocksize,
5747                                             int level)
5748 {
5749         struct extent_buffer *buf;
5750
5751         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5752         if (!buf)
5753                 return ERR_PTR(-ENOMEM);
5754         btrfs_set_header_generation(buf, trans->transid);
5755         btrfs_set_buffer_lockdep_class(buf, level);
5756         btrfs_tree_lock(buf);
5757         clean_tree_block(trans, root, buf);
5758
5759         btrfs_set_lock_blocking(buf);
5760         btrfs_set_buffer_uptodate(buf);
5761
5762         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5763                 /*
5764                  * we allow two log transactions at a time, use different
5765                  * EXENT bit to differentiate dirty pages.
5766                  */
5767                 if (root->log_transid % 2 == 0)
5768                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5769                                         buf->start + buf->len - 1, GFP_NOFS);
5770                 else
5771                         set_extent_new(&root->dirty_log_pages, buf->start,
5772                                         buf->start + buf->len - 1, GFP_NOFS);
5773         } else {
5774                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5775                          buf->start + buf->len - 1, GFP_NOFS);
5776         }
5777         trans->blocks_used++;
5778         /* this returns a buffer locked for blocking */
5779         return buf;
5780 }
5781
5782 static struct btrfs_block_rsv *
5783 use_block_rsv(struct btrfs_trans_handle *trans,
5784               struct btrfs_root *root, u32 blocksize)
5785 {
5786         struct btrfs_block_rsv *block_rsv;
5787         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5788         int ret;
5789
5790         block_rsv = get_block_rsv(trans, root);
5791
5792         if (block_rsv->size == 0) {
5793                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5794                                              blocksize, 0);
5795                 /*
5796                  * If we couldn't reserve metadata bytes try and use some from
5797                  * the global reserve.
5798                  */
5799                 if (ret && block_rsv != global_rsv) {
5800                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5801                         if (!ret)
5802                                 return global_rsv;
5803                         return ERR_PTR(ret);
5804                 } else if (ret) {
5805                         return ERR_PTR(ret);
5806                 }
5807                 return block_rsv;
5808         }
5809
5810         ret = block_rsv_use_bytes(block_rsv, blocksize);
5811         if (!ret)
5812                 return block_rsv;
5813         if (ret) {
5814                 WARN_ON(1);
5815                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5816                                              0);
5817                 if (!ret) {
5818                         spin_lock(&block_rsv->lock);
5819                         block_rsv->size += blocksize;
5820                         spin_unlock(&block_rsv->lock);
5821                         return block_rsv;
5822                 } else if (ret && block_rsv != global_rsv) {
5823                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5824                         if (!ret)
5825                                 return global_rsv;
5826                 }
5827         }
5828
5829         return ERR_PTR(-ENOSPC);
5830 }
5831
5832 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5833 {
5834         block_rsv_add_bytes(block_rsv, blocksize, 0);
5835         block_rsv_release_bytes(block_rsv, NULL, 0);
5836 }
5837
5838 /*
5839  * finds a free extent and does all the dirty work required for allocation
5840  * returns the key for the extent through ins, and a tree buffer for
5841  * the first block of the extent through buf.
5842  *
5843  * returns the tree buffer or NULL.
5844  */
5845 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5846                                         struct btrfs_root *root, u32 blocksize,
5847                                         u64 parent, u64 root_objectid,
5848                                         struct btrfs_disk_key *key, int level,
5849                                         u64 hint, u64 empty_size)
5850 {
5851         struct btrfs_key ins;
5852         struct btrfs_block_rsv *block_rsv;
5853         struct extent_buffer *buf;
5854         u64 flags = 0;
5855         int ret;
5856
5857
5858         block_rsv = use_block_rsv(trans, root, blocksize);
5859         if (IS_ERR(block_rsv))
5860                 return ERR_CAST(block_rsv);
5861
5862         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5863                                    empty_size, hint, (u64)-1, &ins, 0);
5864         if (ret) {
5865                 unuse_block_rsv(block_rsv, blocksize);
5866                 return ERR_PTR(ret);
5867         }
5868
5869         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5870                                     blocksize, level);
5871         BUG_ON(IS_ERR(buf));
5872
5873         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5874                 if (parent == 0)
5875                         parent = ins.objectid;
5876                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5877         } else
5878                 BUG_ON(parent > 0);
5879
5880         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5881                 struct btrfs_delayed_extent_op *extent_op;
5882                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5883                 BUG_ON(!extent_op);
5884                 if (key)
5885                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5886                 else
5887                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5888                 extent_op->flags_to_set = flags;
5889                 extent_op->update_key = 1;
5890                 extent_op->update_flags = 1;
5891                 extent_op->is_data = 0;
5892
5893                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5894                                         ins.offset, parent, root_objectid,
5895                                         level, BTRFS_ADD_DELAYED_EXTENT,
5896                                         extent_op);
5897                 BUG_ON(ret);
5898         }
5899         return buf;
5900 }
5901
5902 struct walk_control {
5903         u64 refs[BTRFS_MAX_LEVEL];
5904         u64 flags[BTRFS_MAX_LEVEL];
5905         struct btrfs_key update_progress;
5906         int stage;
5907         int level;
5908         int shared_level;
5909         int update_ref;
5910         int keep_locks;
5911         int reada_slot;
5912         int reada_count;
5913 };
5914
5915 #define DROP_REFERENCE  1
5916 #define UPDATE_BACKREF  2
5917
5918 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5919                                      struct btrfs_root *root,
5920                                      struct walk_control *wc,
5921                                      struct btrfs_path *path)
5922 {
5923         u64 bytenr;
5924         u64 generation;
5925         u64 refs;
5926         u64 flags;
5927         u32 nritems;
5928         u32 blocksize;
5929         struct btrfs_key key;
5930         struct extent_buffer *eb;
5931         int ret;
5932         int slot;
5933         int nread = 0;
5934
5935         if (path->slots[wc->level] < wc->reada_slot) {
5936                 wc->reada_count = wc->reada_count * 2 / 3;
5937                 wc->reada_count = max(wc->reada_count, 2);
5938         } else {
5939                 wc->reada_count = wc->reada_count * 3 / 2;
5940                 wc->reada_count = min_t(int, wc->reada_count,
5941                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5942         }
5943
5944         eb = path->nodes[wc->level];
5945         nritems = btrfs_header_nritems(eb);
5946         blocksize = btrfs_level_size(root, wc->level - 1);
5947
5948         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5949                 if (nread >= wc->reada_count)
5950                         break;
5951
5952                 cond_resched();
5953                 bytenr = btrfs_node_blockptr(eb, slot);
5954                 generation = btrfs_node_ptr_generation(eb, slot);
5955
5956                 if (slot == path->slots[wc->level])
5957                         goto reada;
5958
5959                 if (wc->stage == UPDATE_BACKREF &&
5960                     generation <= root->root_key.offset)
5961                         continue;
5962
5963                 /* We don't lock the tree block, it's OK to be racy here */
5964                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5965                                                &refs, &flags);
5966                 BUG_ON(ret);
5967                 BUG_ON(refs == 0);
5968
5969                 if (wc->stage == DROP_REFERENCE) {
5970                         if (refs == 1)
5971                                 goto reada;
5972
5973                         if (wc->level == 1 &&
5974                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5975                                 continue;
5976                         if (!wc->update_ref ||
5977                             generation <= root->root_key.offset)
5978                                 continue;
5979                         btrfs_node_key_to_cpu(eb, &key, slot);
5980                         ret = btrfs_comp_cpu_keys(&key,
5981                                                   &wc->update_progress);
5982                         if (ret < 0)
5983                                 continue;
5984                 } else {
5985                         if (wc->level == 1 &&
5986                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5987                                 continue;
5988                 }
5989 reada:
5990                 ret = readahead_tree_block(root, bytenr, blocksize,
5991                                            generation);
5992                 if (ret)
5993                         break;
5994                 nread++;
5995         }
5996         wc->reada_slot = slot;
5997 }
5998
5999 /*
6000  * hepler to process tree block while walking down the tree.
6001  *
6002  * when wc->stage == UPDATE_BACKREF, this function updates
6003  * back refs for pointers in the block.
6004  *
6005  * NOTE: return value 1 means we should stop walking down.
6006  */
6007 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
6008                                    struct btrfs_root *root,
6009                                    struct btrfs_path *path,
6010                                    struct walk_control *wc, int lookup_info)
6011 {
6012         int level = wc->level;
6013         struct extent_buffer *eb = path->nodes[level];
6014         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6015         int ret;
6016
6017         if (wc->stage == UPDATE_BACKREF &&
6018             btrfs_header_owner(eb) != root->root_key.objectid)
6019                 return 1;
6020
6021         /*
6022          * when reference count of tree block is 1, it won't increase
6023          * again. once full backref flag is set, we never clear it.
6024          */
6025         if (lookup_info &&
6026             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
6027              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
6028                 BUG_ON(!path->locks[level]);
6029                 ret = btrfs_lookup_extent_info(trans, root,
6030                                                eb->start, eb->len,
6031                                                &wc->refs[level],
6032                                                &wc->flags[level]);
6033                 BUG_ON(ret);
6034                 BUG_ON(wc->refs[level] == 0);
6035         }
6036
6037         if (wc->stage == DROP_REFERENCE) {
6038                 if (wc->refs[level] > 1)
6039                         return 1;
6040
6041                 if (path->locks[level] && !wc->keep_locks) {
6042                         btrfs_tree_unlock(eb);
6043                         path->locks[level] = 0;
6044                 }
6045                 return 0;
6046         }
6047
6048         /* wc->stage == UPDATE_BACKREF */
6049         if (!(wc->flags[level] & flag)) {
6050                 BUG_ON(!path->locks[level]);
6051                 ret = btrfs_inc_ref(trans, root, eb, 1);
6052                 BUG_ON(ret);
6053                 ret = btrfs_dec_ref(trans, root, eb, 0);
6054                 BUG_ON(ret);
6055                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
6056                                                   eb->len, flag, 0);
6057                 BUG_ON(ret);
6058                 wc->flags[level] |= flag;
6059         }
6060
6061         /*
6062          * the block is shared by multiple trees, so it's not good to
6063          * keep the tree lock
6064          */
6065         if (path->locks[level] && level > 0) {
6066                 btrfs_tree_unlock(eb);
6067                 path->locks[level] = 0;
6068         }
6069         return 0;
6070 }
6071
6072 /*
6073  * hepler to process tree block pointer.
6074  *
6075  * when wc->stage == DROP_REFERENCE, this function checks
6076  * reference count of the block pointed to. if the block
6077  * is shared and we need update back refs for the subtree
6078  * rooted at the block, this function changes wc->stage to
6079  * UPDATE_BACKREF. if the block is shared and there is no
6080  * need to update back, this function drops the reference
6081  * to the block.
6082  *
6083  * NOTE: return value 1 means we should stop walking down.
6084  */
6085 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
6086                                  struct btrfs_root *root,
6087                                  struct btrfs_path *path,
6088                                  struct walk_control *wc, int *lookup_info)
6089 {
6090         u64 bytenr;
6091         u64 generation;
6092         u64 parent;
6093         u32 blocksize;
6094         struct btrfs_key key;
6095         struct extent_buffer *next;
6096         int level = wc->level;
6097         int reada = 0;
6098         int ret = 0;
6099
6100         generation = btrfs_node_ptr_generation(path->nodes[level],
6101                                                path->slots[level]);
6102         /*
6103          * if the lower level block was created before the snapshot
6104          * was created, we know there is no need to update back refs
6105          * for the subtree
6106          */
6107         if (wc->stage == UPDATE_BACKREF &&
6108             generation <= root->root_key.offset) {
6109                 *lookup_info = 1;
6110                 return 1;
6111         }
6112
6113         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6114         blocksize = btrfs_level_size(root, level - 1);
6115
6116         next = btrfs_find_tree_block(root, bytenr, blocksize);
6117         if (!next) {
6118                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6119                 if (!next)
6120                         return -ENOMEM;
6121                 reada = 1;
6122         }
6123         btrfs_tree_lock(next);
6124         btrfs_set_lock_blocking(next);
6125
6126         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6127                                        &wc->refs[level - 1],
6128                                        &wc->flags[level - 1]);
6129         BUG_ON(ret);
6130         BUG_ON(wc->refs[level - 1] == 0);
6131         *lookup_info = 0;
6132
6133         if (wc->stage == DROP_REFERENCE) {
6134                 if (wc->refs[level - 1] > 1) {
6135                         if (level == 1 &&
6136                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6137                                 goto skip;
6138
6139                         if (!wc->update_ref ||
6140                             generation <= root->root_key.offset)
6141                                 goto skip;
6142
6143                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6144                                               path->slots[level]);
6145                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6146                         if (ret < 0)
6147                                 goto skip;
6148
6149                         wc->stage = UPDATE_BACKREF;
6150                         wc->shared_level = level - 1;
6151                 }
6152         } else {
6153                 if (level == 1 &&
6154                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6155                         goto skip;
6156         }
6157
6158         if (!btrfs_buffer_uptodate(next, generation)) {
6159                 btrfs_tree_unlock(next);
6160                 free_extent_buffer(next);
6161                 next = NULL;
6162                 *lookup_info = 1;
6163         }
6164
6165         if (!next) {
6166                 if (reada && level == 1)
6167                         reada_walk_down(trans, root, wc, path);
6168                 next = read_tree_block(root, bytenr, blocksize, generation);
6169                 if (!next)
6170                         return -EIO;
6171                 btrfs_tree_lock(next);
6172                 btrfs_set_lock_blocking(next);
6173         }
6174
6175         level--;
6176         BUG_ON(level != btrfs_header_level(next));
6177         path->nodes[level] = next;
6178         path->slots[level] = 0;
6179         path->locks[level] = 1;
6180         wc->level = level;
6181         if (wc->level == 1)
6182                 wc->reada_slot = 0;
6183         return 0;
6184 skip:
6185         wc->refs[level - 1] = 0;
6186         wc->flags[level - 1] = 0;
6187         if (wc->stage == DROP_REFERENCE) {
6188                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6189                         parent = path->nodes[level]->start;
6190                 } else {
6191                         BUG_ON(root->root_key.objectid !=
6192                                btrfs_header_owner(path->nodes[level]));
6193                         parent = 0;
6194                 }
6195
6196                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6197                                         root->root_key.objectid, level - 1, 0);
6198                 BUG_ON(ret);
6199         }
6200         btrfs_tree_unlock(next);
6201         free_extent_buffer(next);
6202         *lookup_info = 1;
6203         return 1;
6204 }
6205
6206 /*
6207  * hepler to process tree block while walking up the tree.
6208  *
6209  * when wc->stage == DROP_REFERENCE, this function drops
6210  * reference count on the block.
6211  *
6212  * when wc->stage == UPDATE_BACKREF, this function changes
6213  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6214  * to UPDATE_BACKREF previously while processing the block.
6215  *
6216  * NOTE: return value 1 means we should stop walking up.
6217  */
6218 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6219                                  struct btrfs_root *root,
6220                                  struct btrfs_path *path,
6221                                  struct walk_control *wc)
6222 {
6223         int ret;
6224         int level = wc->level;
6225         struct extent_buffer *eb = path->nodes[level];
6226         u64 parent = 0;
6227
6228         if (wc->stage == UPDATE_BACKREF) {
6229                 BUG_ON(wc->shared_level < level);
6230                 if (level < wc->shared_level)
6231                         goto out;
6232
6233                 ret = find_next_key(path, level + 1, &wc->update_progress);
6234                 if (ret > 0)
6235                         wc->update_ref = 0;
6236
6237                 wc->stage = DROP_REFERENCE;
6238                 wc->shared_level = -1;
6239                 path->slots[level] = 0;
6240
6241                 /*
6242                  * check reference count again if the block isn't locked.
6243                  * we should start walking down the tree again if reference
6244                  * count is one.
6245                  */
6246                 if (!path->locks[level]) {
6247                         BUG_ON(level == 0);
6248                         btrfs_tree_lock(eb);
6249                         btrfs_set_lock_blocking(eb);
6250                         path->locks[level] = 1;
6251
6252                         ret = btrfs_lookup_extent_info(trans, root,
6253                                                        eb->start, eb->len,
6254                                                        &wc->refs[level],
6255                                                        &wc->flags[level]);
6256                         BUG_ON(ret);
6257                         BUG_ON(wc->refs[level] == 0);
6258                         if (wc->refs[level] == 1) {
6259                                 btrfs_tree_unlock(eb);
6260                                 path->locks[level] = 0;
6261                                 return 1;
6262                         }
6263                 }
6264         }
6265
6266         /* wc->stage == DROP_REFERENCE */
6267         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6268
6269         if (wc->refs[level] == 1) {
6270                 if (level == 0) {
6271                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6272                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6273                         else
6274                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6275                         BUG_ON(ret);
6276                 }
6277                 /* make block locked assertion in clean_tree_block happy */
6278                 if (!path->locks[level] &&
6279                     btrfs_header_generation(eb) == trans->transid) {
6280                         btrfs_tree_lock(eb);
6281                         btrfs_set_lock_blocking(eb);
6282                         path->locks[level] = 1;
6283                 }
6284                 clean_tree_block(trans, root, eb);
6285         }
6286
6287         if (eb == root->node) {
6288                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6289                         parent = eb->start;
6290                 else
6291                         BUG_ON(root->root_key.objectid !=
6292                                btrfs_header_owner(eb));
6293         } else {
6294                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6295                         parent = path->nodes[level + 1]->start;
6296                 else
6297                         BUG_ON(root->root_key.objectid !=
6298                                btrfs_header_owner(path->nodes[level + 1]));
6299         }
6300
6301         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6302 out:
6303         wc->refs[level] = 0;
6304         wc->flags[level] = 0;
6305         return 0;
6306 }
6307
6308 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6309                                    struct btrfs_root *root,
6310                                    struct btrfs_path *path,
6311                                    struct walk_control *wc)
6312 {
6313         int level = wc->level;
6314         int lookup_info = 1;
6315         int ret;
6316
6317         while (level >= 0) {
6318                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6319                 if (ret > 0)
6320                         break;
6321
6322                 if (level == 0)
6323                         break;
6324
6325                 if (path->slots[level] >=
6326                     btrfs_header_nritems(path->nodes[level]))
6327                         break;
6328
6329                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6330                 if (ret > 0) {
6331                         path->slots[level]++;
6332                         continue;
6333                 } else if (ret < 0)
6334                         return ret;
6335                 level = wc->level;
6336         }
6337         return 0;
6338 }
6339
6340 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6341                                  struct btrfs_root *root,
6342                                  struct btrfs_path *path,
6343                                  struct walk_control *wc, int max_level)
6344 {
6345         int level = wc->level;
6346         int ret;
6347
6348         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6349         while (level < max_level && path->nodes[level]) {
6350                 wc->level = level;
6351                 if (path->slots[level] + 1 <
6352                     btrfs_header_nritems(path->nodes[level])) {
6353                         path->slots[level]++;
6354                         return 0;
6355                 } else {
6356                         ret = walk_up_proc(trans, root, path, wc);
6357                         if (ret > 0)
6358                                 return 0;
6359
6360                         if (path->locks[level]) {
6361                                 btrfs_tree_unlock(path->nodes[level]);
6362                                 path->locks[level] = 0;
6363                         }
6364                         free_extent_buffer(path->nodes[level]);
6365                         path->nodes[level] = NULL;
6366                         level++;
6367                 }
6368         }
6369         return 1;
6370 }
6371
6372 /*
6373  * drop a subvolume tree.
6374  *
6375  * this function traverses the tree freeing any blocks that only
6376  * referenced by the tree.
6377  *
6378  * when a shared tree block is found. this function decreases its
6379  * reference count by one. if update_ref is true, this function
6380  * also make sure backrefs for the shared block and all lower level
6381  * blocks are properly updated.
6382  */
6383 int btrfs_drop_snapshot(struct btrfs_root *root,
6384                         struct btrfs_block_rsv *block_rsv, int update_ref)
6385 {
6386         struct btrfs_path *path;
6387         struct btrfs_trans_handle *trans;
6388         struct btrfs_root *tree_root = root->fs_info->tree_root;
6389         struct btrfs_root_item *root_item = &root->root_item;
6390         struct walk_control *wc;
6391         struct btrfs_key key;
6392         int err = 0;
6393         int ret;
6394         int level;
6395
6396         path = btrfs_alloc_path();
6397         BUG_ON(!path);
6398
6399         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6400         BUG_ON(!wc);
6401
6402         trans = btrfs_start_transaction(tree_root, 0);
6403         BUG_ON(IS_ERR(trans));
6404
6405         if (block_rsv)
6406                 trans->block_rsv = block_rsv;
6407
6408         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6409                 level = btrfs_header_level(root->node);
6410                 path->nodes[level] = btrfs_lock_root_node(root);
6411                 btrfs_set_lock_blocking(path->nodes[level]);
6412                 path->slots[level] = 0;
6413                 path->locks[level] = 1;
6414                 memset(&wc->update_progress, 0,
6415                        sizeof(wc->update_progress));
6416         } else {
6417                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6418                 memcpy(&wc->update_progress, &key,
6419                        sizeof(wc->update_progress));
6420
6421                 level = root_item->drop_level;
6422                 BUG_ON(level == 0);
6423                 path->lowest_level = level;
6424                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6425                 path->lowest_level = 0;
6426                 if (ret < 0) {
6427                         err = ret;
6428                         goto out;
6429                 }
6430                 WARN_ON(ret > 0);
6431
6432                 /*
6433                  * unlock our path, this is safe because only this
6434                  * function is allowed to delete this snapshot
6435                  */
6436                 btrfs_unlock_up_safe(path, 0);
6437
6438                 level = btrfs_header_level(root->node);
6439                 while (1) {
6440                         btrfs_tree_lock(path->nodes[level]);
6441                         btrfs_set_lock_blocking(path->nodes[level]);
6442
6443                         ret = btrfs_lookup_extent_info(trans, root,
6444                                                 path->nodes[level]->start,
6445                                                 path->nodes[level]->len,
6446                                                 &wc->refs[level],
6447                                                 &wc->flags[level]);
6448                         BUG_ON(ret);
6449                         BUG_ON(wc->refs[level] == 0);
6450
6451                         if (level == root_item->drop_level)
6452                                 break;
6453
6454                         btrfs_tree_unlock(path->nodes[level]);
6455                         WARN_ON(wc->refs[level] != 1);
6456                         level--;
6457                 }
6458         }
6459
6460         wc->level = level;
6461         wc->shared_level = -1;
6462         wc->stage = DROP_REFERENCE;
6463         wc->update_ref = update_ref;
6464         wc->keep_locks = 0;
6465         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6466
6467         while (1) {
6468                 ret = walk_down_tree(trans, root, path, wc);
6469                 if (ret < 0) {
6470                         err = ret;
6471                         break;
6472                 }
6473
6474                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6475                 if (ret < 0) {
6476                         err = ret;
6477                         break;
6478                 }
6479
6480                 if (ret > 0) {
6481                         BUG_ON(wc->stage != DROP_REFERENCE);
6482                         break;
6483                 }
6484
6485                 if (wc->stage == DROP_REFERENCE) {
6486                         level = wc->level;
6487                         btrfs_node_key(path->nodes[level],
6488                                        &root_item->drop_progress,
6489                                        path->slots[level]);
6490                         root_item->drop_level = level;
6491                 }
6492
6493                 BUG_ON(wc->level == 0);
6494                 if (btrfs_should_end_transaction(trans, tree_root)) {
6495                         ret = btrfs_update_root(trans, tree_root,
6496                                                 &root->root_key,
6497                                                 root_item);
6498                         BUG_ON(ret);
6499
6500                         btrfs_end_transaction_throttle(trans, tree_root);
6501                         trans = btrfs_start_transaction(tree_root, 0);
6502                         BUG_ON(IS_ERR(trans));
6503                         if (block_rsv)
6504                                 trans->block_rsv = block_rsv;
6505                 }
6506         }
6507         btrfs_release_path(root, path);
6508         BUG_ON(err);
6509
6510         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6511         BUG_ON(ret);
6512
6513         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6514                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6515                                            NULL, NULL);
6516                 BUG_ON(ret < 0);
6517                 if (ret > 0) {
6518                         /* if we fail to delete the orphan item this time
6519                          * around, it'll get picked up the next time.
6520                          *
6521                          * The most common failure here is just -ENOENT.
6522                          */
6523                         btrfs_del_orphan_item(trans, tree_root,
6524                                               root->root_key.objectid);
6525                 }
6526         }
6527
6528         if (root->in_radix) {
6529                 btrfs_free_fs_root(tree_root->fs_info, root);
6530         } else {
6531                 free_extent_buffer(root->node);
6532                 free_extent_buffer(root->commit_root);
6533                 kfree(root);
6534         }
6535 out:
6536         btrfs_end_transaction_throttle(trans, tree_root);
6537         kfree(wc);
6538         btrfs_free_path(path);
6539         return err;
6540 }
6541
6542 /*
6543  * drop subtree rooted at tree block 'node'.
6544  *
6545  * NOTE: this function will unlock and release tree block 'node'
6546  */
6547 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6548                         struct btrfs_root *root,
6549                         struct extent_buffer *node,
6550                         struct extent_buffer *parent)
6551 {
6552         struct btrfs_path *path;
6553         struct walk_control *wc;
6554         int level;
6555         int parent_level;
6556         int ret = 0;
6557         int wret;
6558
6559         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6560
6561         path = btrfs_alloc_path();
6562         if (!path)
6563                 return -ENOMEM;
6564
6565         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6566         if (!wc) {
6567                 btrfs_free_path(path);
6568                 return -ENOMEM;
6569         }
6570
6571         btrfs_assert_tree_locked(parent);
6572         parent_level = btrfs_header_level(parent);
6573         extent_buffer_get(parent);
6574         path->nodes[parent_level] = parent;
6575         path->slots[parent_level] = btrfs_header_nritems(parent);
6576
6577         btrfs_assert_tree_locked(node);
6578         level = btrfs_header_level(node);
6579         path->nodes[level] = node;
6580         path->slots[level] = 0;
6581         path->locks[level] = 1;
6582
6583         wc->refs[parent_level] = 1;
6584         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6585         wc->level = level;
6586         wc->shared_level = -1;
6587         wc->stage = DROP_REFERENCE;
6588         wc->update_ref = 0;
6589         wc->keep_locks = 1;
6590         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6591
6592         while (1) {
6593                 wret = walk_down_tree(trans, root, path, wc);
6594                 if (wret < 0) {
6595                         ret = wret;
6596                         break;
6597                 }
6598
6599                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6600                 if (wret < 0)
6601                         ret = wret;
6602                 if (wret != 0)
6603                         break;
6604         }
6605
6606         kfree(wc);
6607         btrfs_free_path(path);
6608         return ret;
6609 }
6610
6611 #if 0
6612 static unsigned long calc_ra(unsigned long start, unsigned long last,
6613                              unsigned long nr)
6614 {
6615         return min(last, start + nr - 1);
6616 }
6617
6618 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6619                                          u64 len)
6620 {
6621         u64 page_start;
6622         u64 page_end;
6623         unsigned long first_index;
6624         unsigned long last_index;
6625         unsigned long i;
6626         struct page *page;
6627         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6628         struct file_ra_state *ra;
6629         struct btrfs_ordered_extent *ordered;
6630         unsigned int total_read = 0;
6631         unsigned int total_dirty = 0;
6632         int ret = 0;
6633
6634         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6635         if (!ra)
6636                 return -ENOMEM;
6637
6638         mutex_lock(&inode->i_mutex);
6639         first_index = start >> PAGE_CACHE_SHIFT;
6640         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6641
6642         /* make sure the dirty trick played by the caller work */
6643         ret = invalidate_inode_pages2_range(inode->i_mapping,
6644                                             first_index, last_index);
6645         if (ret)
6646                 goto out_unlock;
6647
6648         file_ra_state_init(ra, inode->i_mapping);
6649
6650         for (i = first_index ; i <= last_index; i++) {
6651                 if (total_read % ra->ra_pages == 0) {
6652                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6653                                        calc_ra(i, last_index, ra->ra_pages));
6654                 }
6655                 total_read++;
6656 again:
6657                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6658                         BUG_ON(1);
6659                 page = grab_cache_page(inode->i_mapping, i);
6660                 if (!page) {
6661                         ret = -ENOMEM;
6662                         goto out_unlock;
6663                 }
6664                 if (!PageUptodate(page)) {
6665                         btrfs_readpage(NULL, page);
6666                         lock_page(page);
6667                         if (!PageUptodate(page)) {
6668                                 unlock_page(page);
6669                                 page_cache_release(page);
6670                                 ret = -EIO;
6671                                 goto out_unlock;
6672                         }
6673                 }
6674                 wait_on_page_writeback(page);
6675
6676                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6677                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6678                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6679
6680                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6681                 if (ordered) {
6682                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6683                         unlock_page(page);
6684                         page_cache_release(page);
6685                         btrfs_start_ordered_extent(inode, ordered, 1);
6686                         btrfs_put_ordered_extent(ordered);
6687                         goto again;
6688                 }
6689                 set_page_extent_mapped(page);
6690
6691                 if (i == first_index)
6692                         set_extent_bits(io_tree, page_start, page_end,
6693                                         EXTENT_BOUNDARY, GFP_NOFS);
6694                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6695
6696                 set_page_dirty(page);
6697                 total_dirty++;
6698
6699                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6700                 unlock_page(page);
6701                 page_cache_release(page);
6702         }
6703
6704 out_unlock:
6705         kfree(ra);
6706         mutex_unlock(&inode->i_mutex);
6707         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6708         return ret;
6709 }
6710
6711 static noinline int relocate_data_extent(struct inode *reloc_inode,
6712                                          struct btrfs_key *extent_key,
6713                                          u64 offset)
6714 {
6715         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6716         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6717         struct extent_map *em;
6718         u64 start = extent_key->objectid - offset;
6719         u64 end = start + extent_key->offset - 1;
6720
6721         em = alloc_extent_map(GFP_NOFS);
6722         BUG_ON(!em);
6723
6724         em->start = start;
6725         em->len = extent_key->offset;
6726         em->block_len = extent_key->offset;
6727         em->block_start = extent_key->objectid;
6728         em->bdev = root->fs_info->fs_devices->latest_bdev;
6729         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6730
6731         /* setup extent map to cheat btrfs_readpage */
6732         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6733         while (1) {
6734                 int ret;
6735                 write_lock(&em_tree->lock);
6736                 ret = add_extent_mapping(em_tree, em);
6737                 write_unlock(&em_tree->lock);
6738                 if (ret != -EEXIST) {
6739                         free_extent_map(em);
6740                         break;
6741                 }
6742                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6743         }
6744         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6745
6746         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6747 }
6748
6749 struct btrfs_ref_path {
6750         u64 extent_start;
6751         u64 nodes[BTRFS_MAX_LEVEL];
6752         u64 root_objectid;
6753         u64 root_generation;
6754         u64 owner_objectid;
6755         u32 num_refs;
6756         int lowest_level;
6757         int current_level;
6758         int shared_level;
6759
6760         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6761         u64 new_nodes[BTRFS_MAX_LEVEL];
6762 };
6763
6764 struct disk_extent {
6765         u64 ram_bytes;
6766         u64 disk_bytenr;
6767         u64 disk_num_bytes;
6768         u64 offset;
6769         u64 num_bytes;
6770         u8 compression;
6771         u8 encryption;
6772         u16 other_encoding;
6773 };
6774
6775 static int is_cowonly_root(u64 root_objectid)
6776 {
6777         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6778             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6779             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6780             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6781             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6782             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6783                 return 1;
6784         return 0;
6785 }
6786
6787 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6788                                     struct btrfs_root *extent_root,
6789                                     struct btrfs_ref_path *ref_path,
6790                                     int first_time)
6791 {
6792         struct extent_buffer *leaf;
6793         struct btrfs_path *path;
6794         struct btrfs_extent_ref *ref;
6795         struct btrfs_key key;
6796         struct btrfs_key found_key;
6797         u64 bytenr;
6798         u32 nritems;
6799         int level;
6800         int ret = 1;
6801
6802         path = btrfs_alloc_path();
6803         if (!path)
6804                 return -ENOMEM;
6805
6806         if (first_time) {
6807                 ref_path->lowest_level = -1;
6808                 ref_path->current_level = -1;
6809                 ref_path->shared_level = -1;
6810                 goto walk_up;
6811         }
6812 walk_down:
6813         level = ref_path->current_level - 1;
6814         while (level >= -1) {
6815                 u64 parent;
6816                 if (level < ref_path->lowest_level)
6817                         break;
6818
6819                 if (level >= 0)
6820                         bytenr = ref_path->nodes[level];
6821                 else
6822                         bytenr = ref_path->extent_start;
6823                 BUG_ON(bytenr == 0);
6824
6825                 parent = ref_path->nodes[level + 1];
6826                 ref_path->nodes[level + 1] = 0;
6827                 ref_path->current_level = level;
6828                 BUG_ON(parent == 0);
6829
6830                 key.objectid = bytenr;
6831                 key.offset = parent + 1;
6832                 key.type = BTRFS_EXTENT_REF_KEY;
6833
6834                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6835                 if (ret < 0)
6836                         goto out;
6837                 BUG_ON(ret == 0);
6838
6839                 leaf = path->nodes[0];
6840                 nritems = btrfs_header_nritems(leaf);
6841                 if (path->slots[0] >= nritems) {
6842                         ret = btrfs_next_leaf(extent_root, path);
6843                         if (ret < 0)
6844                                 goto out;
6845                         if (ret > 0)
6846                                 goto next;
6847                         leaf = path->nodes[0];
6848                 }
6849
6850                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6851                 if (found_key.objectid == bytenr &&
6852                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6853                         if (level < ref_path->shared_level)
6854                                 ref_path->shared_level = level;
6855                         goto found;
6856                 }
6857 next:
6858                 level--;
6859                 btrfs_release_path(extent_root, path);
6860                 cond_resched();
6861         }
6862         /* reached lowest level */
6863         ret = 1;
6864         goto out;
6865 walk_up:
6866         level = ref_path->current_level;
6867         while (level < BTRFS_MAX_LEVEL - 1) {
6868                 u64 ref_objectid;
6869
6870                 if (level >= 0)
6871                         bytenr = ref_path->nodes[level];
6872                 else
6873                         bytenr = ref_path->extent_start;
6874
6875                 BUG_ON(bytenr == 0);
6876
6877                 key.objectid = bytenr;
6878                 key.offset = 0;
6879                 key.type = BTRFS_EXTENT_REF_KEY;
6880
6881                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6882                 if (ret < 0)
6883                         goto out;
6884
6885                 leaf = path->nodes[0];
6886                 nritems = btrfs_header_nritems(leaf);
6887                 if (path->slots[0] >= nritems) {
6888                         ret = btrfs_next_leaf(extent_root, path);
6889                         if (ret < 0)
6890                                 goto out;
6891                         if (ret > 0) {
6892                                 /* the extent was freed by someone */
6893                                 if (ref_path->lowest_level == level)
6894                                         goto out;
6895                                 btrfs_release_path(extent_root, path);
6896                                 goto walk_down;
6897                         }
6898                         leaf = path->nodes[0];
6899                 }
6900
6901                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6902                 if (found_key.objectid != bytenr ||
6903                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6904                         /* the extent was freed by someone */
6905                         if (ref_path->lowest_level == level) {
6906                                 ret = 1;
6907                                 goto out;
6908                         }
6909                         btrfs_release_path(extent_root, path);
6910                         goto walk_down;
6911                 }
6912 found:
6913                 ref = btrfs_item_ptr(leaf, path->slots[0],
6914                                 struct btrfs_extent_ref);
6915                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6916                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6917                         if (first_time) {
6918                                 level = (int)ref_objectid;
6919                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6920                                 ref_path->lowest_level = level;
6921                                 ref_path->current_level = level;
6922                                 ref_path->nodes[level] = bytenr;
6923                         } else {
6924                                 WARN_ON(ref_objectid != level);
6925                         }
6926                 } else {
6927                         WARN_ON(level != -1);
6928                 }
6929                 first_time = 0;
6930
6931                 if (ref_path->lowest_level == level) {
6932                         ref_path->owner_objectid = ref_objectid;
6933                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6934                 }
6935
6936                 /*
6937                  * the block is tree root or the block isn't in reference
6938                  * counted tree.
6939                  */
6940                 if (found_key.objectid == found_key.offset ||
6941                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6942                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6943                         ref_path->root_generation =
6944                                 btrfs_ref_generation(leaf, ref);
6945                         if (level < 0) {
6946                                 /* special reference from the tree log */
6947                                 ref_path->nodes[0] = found_key.offset;
6948                                 ref_path->current_level = 0;
6949                         }
6950                         ret = 0;
6951                         goto out;
6952                 }
6953
6954                 level++;
6955                 BUG_ON(ref_path->nodes[level] != 0);
6956                 ref_path->nodes[level] = found_key.offset;
6957                 ref_path->current_level = level;
6958
6959                 /*
6960                  * the reference was created in the running transaction,
6961                  * no need to continue walking up.
6962                  */
6963                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6964                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6965                         ref_path->root_generation =
6966                                 btrfs_ref_generation(leaf, ref);
6967                         ret = 0;
6968                         goto out;
6969                 }
6970
6971                 btrfs_release_path(extent_root, path);
6972                 cond_resched();
6973         }
6974         /* reached max tree level, but no tree root found. */
6975         BUG();
6976 out:
6977         btrfs_free_path(path);
6978         return ret;
6979 }
6980
6981 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6982                                 struct btrfs_root *extent_root,
6983                                 struct btrfs_ref_path *ref_path,
6984                                 u64 extent_start)
6985 {
6986         memset(ref_path, 0, sizeof(*ref_path));
6987         ref_path->extent_start = extent_start;
6988
6989         return __next_ref_path(trans, extent_root, ref_path, 1);
6990 }
6991
6992 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6993                                struct btrfs_root *extent_root,
6994                                struct btrfs_ref_path *ref_path)
6995 {
6996         return __next_ref_path(trans, extent_root, ref_path, 0);
6997 }
6998
6999 static noinline int get_new_locations(struct inode *reloc_inode,
7000                                       struct btrfs_key *extent_key,
7001                                       u64 offset, int no_fragment,
7002                                       struct disk_extent **extents,
7003                                       int *nr_extents)
7004 {
7005         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
7006         struct btrfs_path *path;
7007         struct btrfs_file_extent_item *fi;
7008         struct extent_buffer *leaf;
7009         struct disk_extent *exts = *extents;
7010         struct btrfs_key found_key;
7011         u64 cur_pos;
7012         u64 last_byte;
7013         u32 nritems;
7014         int nr = 0;
7015         int max = *nr_extents;
7016         int ret;
7017
7018         WARN_ON(!no_fragment && *extents);
7019         if (!exts) {
7020                 max = 1;
7021                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
7022                 if (!exts)
7023                         return -ENOMEM;
7024         }
7025
7026         path = btrfs_alloc_path();
7027         if (!path) {
7028                 if (exts != *extents)
7029                         kfree(exts);
7030                 return -ENOMEM;
7031         }
7032
7033         cur_pos = extent_key->objectid - offset;
7034         last_byte = extent_key->objectid + extent_key->offset;
7035         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
7036                                        cur_pos, 0);
7037         if (ret < 0)
7038                 goto out;
7039         if (ret > 0) {
7040                 ret = -ENOENT;
7041                 goto out;
7042         }
7043
7044         while (1) {
7045                 leaf = path->nodes[0];
7046                 nritems = btrfs_header_nritems(leaf);
7047                 if (path->slots[0] >= nritems) {
7048                         ret = btrfs_next_leaf(root, path);
7049                         if (ret < 0)
7050                                 goto out;
7051                         if (ret > 0)
7052                                 break;
7053                         leaf = path->nodes[0];
7054                 }
7055
7056                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7057                 if (found_key.offset != cur_pos ||
7058                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
7059                     found_key.objectid != reloc_inode->i_ino)
7060                         break;
7061
7062                 fi = btrfs_item_ptr(leaf, path->slots[0],
7063                                     struct btrfs_file_extent_item);
7064                 if (btrfs_file_extent_type(leaf, fi) !=
7065                     BTRFS_FILE_EXTENT_REG ||
7066                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7067                         break;
7068
7069                 if (nr == max) {
7070                         struct disk_extent *old = exts;
7071                         max *= 2;
7072                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
7073                         if (!exts) {
7074                                 ret = -ENOMEM;
7075                                 goto out;
7076                         }
7077                         memcpy(exts, old, sizeof(*exts) * nr);
7078                         if (old != *extents)
7079                                 kfree(old);
7080                 }
7081
7082                 exts[nr].disk_bytenr =
7083                         btrfs_file_extent_disk_bytenr(leaf, fi);
7084                 exts[nr].disk_num_bytes =
7085                         btrfs_file_extent_disk_num_bytes(leaf, fi);
7086                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
7087                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7088                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7089                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
7090                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
7091                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
7092                                                                            fi);
7093                 BUG_ON(exts[nr].offset > 0);
7094                 BUG_ON(exts[nr].compression || exts[nr].encryption);
7095                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
7096
7097                 cur_pos += exts[nr].num_bytes;
7098                 nr++;
7099
7100                 if (cur_pos + offset >= last_byte)
7101                         break;
7102
7103                 if (no_fragment) {
7104                         ret = 1;
7105                         goto out;
7106                 }
7107                 path->slots[0]++;
7108         }
7109
7110         BUG_ON(cur_pos + offset > last_byte);
7111         if (cur_pos + offset < last_byte) {
7112                 ret = -ENOENT;
7113                 goto out;
7114         }
7115         ret = 0;
7116 out:
7117         btrfs_free_path(path);
7118         if (ret) {
7119                 if (exts != *extents)
7120                         kfree(exts);
7121         } else {
7122                 *extents = exts;
7123                 *nr_extents = nr;
7124         }
7125         return ret;
7126 }
7127
7128 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
7129                                         struct btrfs_root *root,
7130                                         struct btrfs_path *path,
7131                                         struct btrfs_key *extent_key,
7132                                         struct btrfs_key *leaf_key,
7133                                         struct btrfs_ref_path *ref_path,
7134                                         struct disk_extent *new_extents,
7135                                         int nr_extents)
7136 {
7137         struct extent_buffer *leaf;
7138         struct btrfs_file_extent_item *fi;
7139         struct inode *inode = NULL;
7140         struct btrfs_key key;
7141         u64 lock_start = 0;
7142         u64 lock_end = 0;
7143         u64 num_bytes;
7144         u64 ext_offset;
7145         u64 search_end = (u64)-1;
7146         u32 nritems;
7147         int nr_scaned = 0;
7148         int extent_locked = 0;
7149         int extent_type;
7150         int ret;
7151
7152         memcpy(&key, leaf_key, sizeof(key));
7153         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7154                 if (key.objectid < ref_path->owner_objectid ||
7155                     (key.objectid == ref_path->owner_objectid &&
7156                      key.type < BTRFS_EXTENT_DATA_KEY)) {
7157                         key.objectid = ref_path->owner_objectid;
7158                         key.type = BTRFS_EXTENT_DATA_KEY;
7159                         key.offset = 0;
7160                 }
7161         }
7162
7163         while (1) {
7164                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7165                 if (ret < 0)
7166                         goto out;
7167
7168                 leaf = path->nodes[0];
7169                 nritems = btrfs_header_nritems(leaf);
7170 next:
7171                 if (extent_locked && ret > 0) {
7172                         /*
7173                          * the file extent item was modified by someone
7174                          * before the extent got locked.
7175                          */
7176                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7177                                       lock_end, GFP_NOFS);
7178                         extent_locked = 0;
7179                 }
7180
7181                 if (path->slots[0] >= nritems) {
7182                         if (++nr_scaned > 2)
7183                                 break;
7184
7185                         BUG_ON(extent_locked);
7186                         ret = btrfs_next_leaf(root, path);
7187                         if (ret < 0)
7188                                 goto out;
7189                         if (ret > 0)
7190                                 break;
7191                         leaf = path->nodes[0];
7192                         nritems = btrfs_header_nritems(leaf);
7193                 }
7194
7195                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7196
7197                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7198                         if ((key.objectid > ref_path->owner_objectid) ||
7199                             (key.objectid == ref_path->owner_objectid &&
7200                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7201                             key.offset >= search_end)
7202                                 break;
7203                 }
7204
7205                 if (inode && key.objectid != inode->i_ino) {
7206                         BUG_ON(extent_locked);
7207                         btrfs_release_path(root, path);
7208                         mutex_unlock(&inode->i_mutex);
7209                         iput(inode);
7210                         inode = NULL;
7211                         continue;
7212                 }
7213
7214                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7215                         path->slots[0]++;
7216                         ret = 1;
7217                         goto next;
7218                 }
7219                 fi = btrfs_item_ptr(leaf, path->slots[0],
7220                                     struct btrfs_file_extent_item);
7221                 extent_type = btrfs_file_extent_type(leaf, fi);
7222                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7223                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7224                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7225                      extent_key->objectid)) {
7226                         path->slots[0]++;
7227                         ret = 1;
7228                         goto next;
7229                 }
7230
7231                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7232                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7233
7234                 if (search_end == (u64)-1) {
7235                         search_end = key.offset - ext_offset +
7236                                 btrfs_file_extent_ram_bytes(leaf, fi);
7237                 }
7238
7239                 if (!extent_locked) {
7240                         lock_start = key.offset;
7241                         lock_end = lock_start + num_bytes - 1;
7242                 } else {
7243                         if (lock_start > key.offset ||
7244                             lock_end + 1 < key.offset + num_bytes) {
7245                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7246                                               lock_start, lock_end, GFP_NOFS);
7247                                 extent_locked = 0;
7248                         }
7249                 }
7250
7251                 if (!inode) {
7252                         btrfs_release_path(root, path);
7253
7254                         inode = btrfs_iget_locked(root->fs_info->sb,
7255                                                   key.objectid, root);
7256                         if (inode->i_state & I_NEW) {
7257                                 BTRFS_I(inode)->root = root;
7258                                 BTRFS_I(inode)->location.objectid =
7259                                         key.objectid;
7260                                 BTRFS_I(inode)->location.type =
7261                                         BTRFS_INODE_ITEM_KEY;
7262                                 BTRFS_I(inode)->location.offset = 0;
7263                                 btrfs_read_locked_inode(inode);
7264                                 unlock_new_inode(inode);
7265                         }
7266                         /*
7267                          * some code call btrfs_commit_transaction while
7268                          * holding the i_mutex, so we can't use mutex_lock
7269                          * here.
7270                          */
7271                         if (is_bad_inode(inode) ||
7272                             !mutex_trylock(&inode->i_mutex)) {
7273                                 iput(inode);
7274                                 inode = NULL;
7275                                 key.offset = (u64)-1;
7276                                 goto skip;
7277                         }
7278                 }
7279
7280                 if (!extent_locked) {
7281                         struct btrfs_ordered_extent *ordered;
7282
7283                         btrfs_release_path(root, path);
7284
7285                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7286                                     lock_end, GFP_NOFS);
7287                         ordered = btrfs_lookup_first_ordered_extent(inode,
7288                                                                     lock_end);
7289                         if (ordered &&
7290                             ordered->file_offset <= lock_end &&
7291                             ordered->file_offset + ordered->len > lock_start) {
7292                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7293                                               lock_start, lock_end, GFP_NOFS);
7294                                 btrfs_start_ordered_extent(inode, ordered, 1);
7295                                 btrfs_put_ordered_extent(ordered);
7296                                 key.offset += num_bytes;
7297                                 goto skip;
7298                         }
7299                         if (ordered)
7300                                 btrfs_put_ordered_extent(ordered);
7301
7302                         extent_locked = 1;
7303                         continue;
7304                 }
7305
7306                 if (nr_extents == 1) {
7307                         /* update extent pointer in place */
7308                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7309                                                 new_extents[0].disk_bytenr);
7310                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7311                                                 new_extents[0].disk_num_bytes);
7312                         btrfs_mark_buffer_dirty(leaf);
7313
7314                         btrfs_drop_extent_cache(inode, key.offset,
7315                                                 key.offset + num_bytes - 1, 0);
7316
7317                         ret = btrfs_inc_extent_ref(trans, root,
7318                                                 new_extents[0].disk_bytenr,
7319                                                 new_extents[0].disk_num_bytes,
7320                                                 leaf->start,
7321                                                 root->root_key.objectid,
7322                                                 trans->transid,
7323                                                 key.objectid);
7324                         BUG_ON(ret);
7325
7326                         ret = btrfs_free_extent(trans, root,
7327                                                 extent_key->objectid,
7328                                                 extent_key->offset,
7329                                                 leaf->start,
7330                                                 btrfs_header_owner(leaf),
7331                                                 btrfs_header_generation(leaf),
7332                                                 key.objectid, 0);
7333                         BUG_ON(ret);
7334
7335                         btrfs_release_path(root, path);
7336                         key.offset += num_bytes;
7337                 } else {
7338                         BUG_ON(1);
7339 #if 0
7340                         u64 alloc_hint;
7341                         u64 extent_len;
7342                         int i;
7343                         /*
7344                          * drop old extent pointer at first, then insert the
7345                          * new pointers one bye one
7346                          */
7347                         btrfs_release_path(root, path);
7348                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7349                                                  key.offset + num_bytes,
7350                                                  key.offset, &alloc_hint);
7351                         BUG_ON(ret);
7352
7353                         for (i = 0; i < nr_extents; i++) {
7354                                 if (ext_offset >= new_extents[i].num_bytes) {
7355                                         ext_offset -= new_extents[i].num_bytes;
7356                                         continue;
7357                                 }
7358                                 extent_len = min(new_extents[i].num_bytes -
7359                                                  ext_offset, num_bytes);
7360
7361                                 ret = btrfs_insert_empty_item(trans, root,
7362                                                               path, &key,
7363                                                               sizeof(*fi));
7364                                 BUG_ON(ret);
7365
7366                                 leaf = path->nodes[0];
7367                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7368                                                 struct btrfs_file_extent_item);
7369                                 btrfs_set_file_extent_generation(leaf, fi,
7370                                                         trans->transid);
7371                                 btrfs_set_file_extent_type(leaf, fi,
7372                                                         BTRFS_FILE_EXTENT_REG);
7373                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7374                                                 new_extents[i].disk_bytenr);
7375                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7376                                                 new_extents[i].disk_num_bytes);
7377                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7378                                                 new_extents[i].ram_bytes);
7379
7380                                 btrfs_set_file_extent_compression(leaf, fi,
7381                                                 new_extents[i].compression);
7382                                 btrfs_set_file_extent_encryption(leaf, fi,
7383                                                 new_extents[i].encryption);
7384                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7385                                                 new_extents[i].other_encoding);
7386
7387                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7388                                                         extent_len);
7389                                 ext_offset += new_extents[i].offset;
7390                                 btrfs_set_file_extent_offset(leaf, fi,
7391                                                         ext_offset);
7392                                 btrfs_mark_buffer_dirty(leaf);
7393
7394                                 btrfs_drop_extent_cache(inode, key.offset,
7395                                                 key.offset + extent_len - 1, 0);
7396
7397                                 ret = btrfs_inc_extent_ref(trans, root,
7398                                                 new_extents[i].disk_bytenr,
7399                                                 new_extents[i].disk_num_bytes,
7400                                                 leaf->start,
7401                                                 root->root_key.objectid,
7402                                                 trans->transid, key.objectid);
7403                                 BUG_ON(ret);
7404                                 btrfs_release_path(root, path);
7405
7406                                 inode_add_bytes(inode, extent_len);
7407
7408                                 ext_offset = 0;
7409                                 num_bytes -= extent_len;
7410                                 key.offset += extent_len;
7411
7412                                 if (num_bytes == 0)
7413                                         break;
7414                         }
7415                         BUG_ON(i >= nr_extents);
7416 #endif
7417                 }
7418
7419                 if (extent_locked) {
7420                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7421                                       lock_end, GFP_NOFS);
7422                         extent_locked = 0;
7423                 }
7424 skip:
7425                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7426                     key.offset >= search_end)
7427                         break;
7428
7429                 cond_resched();
7430         }
7431         ret = 0;
7432 out:
7433         btrfs_release_path(root, path);
7434         if (inode) {
7435                 mutex_unlock(&inode->i_mutex);
7436                 if (extent_locked) {
7437                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7438                                       lock_end, GFP_NOFS);
7439                 }
7440                 iput(inode);
7441         }
7442         return ret;
7443 }
7444
7445 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7446                                struct btrfs_root *root,
7447                                struct extent_buffer *buf, u64 orig_start)
7448 {
7449         int level;
7450         int ret;
7451
7452         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7453         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7454
7455         level = btrfs_header_level(buf);
7456         if (level == 0) {
7457                 struct btrfs_leaf_ref *ref;
7458                 struct btrfs_leaf_ref *orig_ref;
7459
7460                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7461                 if (!orig_ref)
7462                         return -ENOENT;
7463
7464                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7465                 if (!ref) {
7466                         btrfs_free_leaf_ref(root, orig_ref);
7467                         return -ENOMEM;
7468                 }
7469
7470                 ref->nritems = orig_ref->nritems;
7471                 memcpy(ref->extents, orig_ref->extents,
7472                         sizeof(ref->extents[0]) * ref->nritems);
7473
7474                 btrfs_free_leaf_ref(root, orig_ref);
7475
7476                 ref->root_gen = trans->transid;
7477                 ref->bytenr = buf->start;
7478                 ref->owner = btrfs_header_owner(buf);
7479                 ref->generation = btrfs_header_generation(buf);
7480
7481                 ret = btrfs_add_leaf_ref(root, ref, 0);
7482                 WARN_ON(ret);
7483                 btrfs_free_leaf_ref(root, ref);
7484         }
7485         return 0;
7486 }
7487
7488 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7489                                         struct extent_buffer *leaf,
7490                                         struct btrfs_block_group_cache *group,
7491                                         struct btrfs_root *target_root)
7492 {
7493         struct btrfs_key key;
7494         struct inode *inode = NULL;
7495         struct btrfs_file_extent_item *fi;
7496         struct extent_state *cached_state = NULL;
7497         u64 num_bytes;
7498         u64 skip_objectid = 0;
7499         u32 nritems;
7500         u32 i;
7501
7502         nritems = btrfs_header_nritems(leaf);
7503         for (i = 0; i < nritems; i++) {
7504                 btrfs_item_key_to_cpu(leaf, &key, i);
7505                 if (key.objectid == skip_objectid ||
7506                     key.type != BTRFS_EXTENT_DATA_KEY)
7507                         continue;
7508                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7509                 if (btrfs_file_extent_type(leaf, fi) ==
7510                     BTRFS_FILE_EXTENT_INLINE)
7511                         continue;
7512                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7513                         continue;
7514                 if (!inode || inode->i_ino != key.objectid) {
7515                         iput(inode);
7516                         inode = btrfs_ilookup(target_root->fs_info->sb,
7517                                               key.objectid, target_root, 1);
7518                 }
7519                 if (!inode) {
7520                         skip_objectid = key.objectid;
7521                         continue;
7522                 }
7523                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7524
7525                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7526                                  key.offset + num_bytes - 1, 0, &cached_state,
7527                                  GFP_NOFS);
7528                 btrfs_drop_extent_cache(inode, key.offset,
7529                                         key.offset + num_bytes - 1, 1);
7530                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7531                                      key.offset + num_bytes - 1, &cached_state,
7532                                      GFP_NOFS);
7533                 cond_resched();
7534         }
7535         iput(inode);
7536         return 0;
7537 }
7538
7539 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7540                                         struct btrfs_root *root,
7541                                         struct extent_buffer *leaf,
7542                                         struct btrfs_block_group_cache *group,
7543                                         struct inode *reloc_inode)
7544 {
7545         struct btrfs_key key;
7546         struct btrfs_key extent_key;
7547         struct btrfs_file_extent_item *fi;
7548         struct btrfs_leaf_ref *ref;
7549         struct disk_extent *new_extent;
7550         u64 bytenr;
7551         u64 num_bytes;
7552         u32 nritems;
7553         u32 i;
7554         int ext_index;
7555         int nr_extent;
7556         int ret;
7557
7558         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7559         if (!new_extent)
7560                 return -ENOMEM;
7561
7562         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7563         BUG_ON(!ref);
7564
7565         ext_index = -1;
7566         nritems = btrfs_header_nritems(leaf);
7567         for (i = 0; i < nritems; i++) {
7568                 btrfs_item_key_to_cpu(leaf, &key, i);
7569                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7570                         continue;
7571                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7572                 if (btrfs_file_extent_type(leaf, fi) ==
7573                     BTRFS_FILE_EXTENT_INLINE)
7574                         continue;
7575                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7576                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7577                 if (bytenr == 0)
7578                         continue;
7579
7580                 ext_index++;
7581                 if (bytenr >= group->key.objectid + group->key.offset ||
7582                     bytenr + num_bytes <= group->key.objectid)
7583                         continue;
7584
7585                 extent_key.objectid = bytenr;
7586                 extent_key.offset = num_bytes;
7587                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7588                 nr_extent = 1;
7589                 ret = get_new_locations(reloc_inode, &extent_key,
7590                                         group->key.objectid, 1,
7591                                         &new_extent, &nr_extent);
7592                 if (ret > 0)
7593                         continue;
7594                 BUG_ON(ret < 0);
7595
7596                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7597                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7598                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7599                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7600
7601                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7602                                                 new_extent->disk_bytenr);
7603                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7604                                                 new_extent->disk_num_bytes);
7605                 btrfs_mark_buffer_dirty(leaf);
7606
7607                 ret = btrfs_inc_extent_ref(trans, root,
7608                                         new_extent->disk_bytenr,
7609                                         new_extent->disk_num_bytes,
7610                                         leaf->start,
7611                                         root->root_key.objectid,
7612                                         trans->transid, key.objectid);
7613                 BUG_ON(ret);
7614
7615                 ret = btrfs_free_extent(trans, root,
7616                                         bytenr, num_bytes, leaf->start,
7617                                         btrfs_header_owner(leaf),
7618                                         btrfs_header_generation(leaf),
7619                                         key.objectid, 0);
7620                 BUG_ON(ret);
7621                 cond_resched();
7622         }
7623         kfree(new_extent);
7624         BUG_ON(ext_index + 1 != ref->nritems);
7625         btrfs_free_leaf_ref(root, ref);
7626         return 0;
7627 }
7628
7629 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7630                           struct btrfs_root *root)
7631 {
7632         struct btrfs_root *reloc_root;
7633         int ret;
7634
7635         if (root->reloc_root) {
7636                 reloc_root = root->reloc_root;
7637                 root->reloc_root = NULL;
7638                 list_add(&reloc_root->dead_list,
7639                          &root->fs_info->dead_reloc_roots);
7640
7641                 btrfs_set_root_bytenr(&reloc_root->root_item,
7642                                       reloc_root->node->start);
7643                 btrfs_set_root_level(&root->root_item,
7644                                      btrfs_header_level(reloc_root->node));
7645                 memset(&reloc_root->root_item.drop_progress, 0,
7646                         sizeof(struct btrfs_disk_key));
7647                 reloc_root->root_item.drop_level = 0;
7648
7649                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7650                                         &reloc_root->root_key,
7651                                         &reloc_root->root_item);
7652                 BUG_ON(ret);
7653         }
7654         return 0;
7655 }
7656
7657 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7658 {
7659         struct btrfs_trans_handle *trans;
7660         struct btrfs_root *reloc_root;
7661         struct btrfs_root *prev_root = NULL;
7662         struct list_head dead_roots;
7663         int ret;
7664         unsigned long nr;
7665
7666         INIT_LIST_HEAD(&dead_roots);
7667         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7668
7669         while (!list_empty(&dead_roots)) {
7670                 reloc_root = list_entry(dead_roots.prev,
7671                                         struct btrfs_root, dead_list);
7672                 list_del_init(&reloc_root->dead_list);
7673
7674                 BUG_ON(reloc_root->commit_root != NULL);
7675                 while (1) {
7676                         trans = btrfs_join_transaction(root);
7677                         BUG_ON(IS_ERR(trans));
7678
7679                         mutex_lock(&root->fs_info->drop_mutex);
7680                         ret = btrfs_drop_snapshot(trans, reloc_root);
7681                         if (ret != -EAGAIN)
7682                                 break;
7683                         mutex_unlock(&root->fs_info->drop_mutex);
7684
7685                         nr = trans->blocks_used;
7686                         ret = btrfs_end_transaction(trans, root);
7687                         BUG_ON(ret);
7688                         btrfs_btree_balance_dirty(root, nr);
7689                 }
7690
7691                 free_extent_buffer(reloc_root->node);
7692
7693                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7694                                      &reloc_root->root_key);
7695                 BUG_ON(ret);
7696                 mutex_unlock(&root->fs_info->drop_mutex);
7697
7698                 nr = trans->blocks_used;
7699                 ret = btrfs_end_transaction(trans, root);
7700                 BUG_ON(ret);
7701                 btrfs_btree_balance_dirty(root, nr);
7702
7703                 kfree(prev_root);
7704                 prev_root = reloc_root;
7705         }
7706         if (prev_root) {
7707                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7708                 kfree(prev_root);
7709         }
7710         return 0;
7711 }
7712
7713 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7714 {
7715         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7716         return 0;
7717 }
7718
7719 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7720 {
7721         struct btrfs_root *reloc_root;
7722         struct btrfs_trans_handle *trans;
7723         struct btrfs_key location;
7724         int found;
7725         int ret;
7726
7727         mutex_lock(&root->fs_info->tree_reloc_mutex);
7728         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7729         BUG_ON(ret);
7730         found = !list_empty(&root->fs_info->dead_reloc_roots);
7731         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7732
7733         if (found) {
7734                 trans = btrfs_start_transaction(root, 1);
7735                 BUG_ON(IS_ERR(trans));
7736                 ret = btrfs_commit_transaction(trans, root);
7737                 BUG_ON(ret);
7738         }
7739
7740         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7741         location.offset = (u64)-1;
7742         location.type = BTRFS_ROOT_ITEM_KEY;
7743
7744         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7745         BUG_ON(!reloc_root);
7746         ret = btrfs_orphan_cleanup(reloc_root);
7747         BUG_ON(ret);
7748         return 0;
7749 }
7750
7751 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7752                                     struct btrfs_root *root)
7753 {
7754         struct btrfs_root *reloc_root;
7755         struct extent_buffer *eb;
7756         struct btrfs_root_item *root_item;
7757         struct btrfs_key root_key;
7758         int ret;
7759
7760         BUG_ON(!root->ref_cows);
7761         if (root->reloc_root)
7762                 return 0;
7763
7764         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7765         if (!root_item)
7766                 return -ENOMEM;
7767
7768         ret = btrfs_copy_root(trans, root, root->commit_root,
7769                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7770         BUG_ON(ret);
7771
7772         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7773         root_key.offset = root->root_key.objectid;
7774         root_key.type = BTRFS_ROOT_ITEM_KEY;
7775
7776         memcpy(root_item, &root->root_item, sizeof(root_item));
7777         btrfs_set_root_refs(root_item, 0);
7778         btrfs_set_root_bytenr(root_item, eb->start);
7779         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7780         btrfs_set_root_generation(root_item, trans->transid);
7781
7782         btrfs_tree_unlock(eb);
7783         free_extent_buffer(eb);
7784
7785         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7786                                 &root_key, root_item);
7787         BUG_ON(ret);
7788         kfree(root_item);
7789
7790         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7791                                                  &root_key);
7792         BUG_ON(IS_ERR(reloc_root));
7793         reloc_root->last_trans = trans->transid;
7794         reloc_root->commit_root = NULL;
7795         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7796
7797         root->reloc_root = reloc_root;
7798         return 0;
7799 }
7800
7801 /*
7802  * Core function of space balance.
7803  *
7804  * The idea is using reloc trees to relocate tree blocks in reference
7805  * counted roots. There is one reloc tree for each subvol, and all
7806  * reloc trees share same root key objectid. Reloc trees are snapshots
7807  * of the latest committed roots of subvols (root->commit_root).
7808  *
7809  * To relocate a tree block referenced by a subvol, there are two steps.
7810  * COW the block through subvol's reloc tree, then update block pointer
7811  * in the subvol to point to the new block. Since all reloc trees share
7812  * same root key objectid, doing special handing for tree blocks owned
7813  * by them is easy. Once a tree block has been COWed in one reloc tree,
7814  * we can use the resulting new block directly when the same block is
7815  * required to COW again through other reloc trees. By this way, relocated
7816  * tree blocks are shared between reloc trees, so they are also shared
7817  * between subvols.
7818  */
7819 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7820                                       struct btrfs_root *root,
7821                                       struct btrfs_path *path,
7822                                       struct btrfs_key *first_key,
7823                                       struct btrfs_ref_path *ref_path,
7824                                       struct btrfs_block_group_cache *group,
7825                                       struct inode *reloc_inode)
7826 {
7827         struct btrfs_root *reloc_root;
7828         struct extent_buffer *eb = NULL;
7829         struct btrfs_key *keys;
7830         u64 *nodes;
7831         int level;
7832         int shared_level;
7833         int lowest_level = 0;
7834         int ret;
7835
7836         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7837                 lowest_level = ref_path->owner_objectid;
7838
7839         if (!root->ref_cows) {
7840                 path->lowest_level = lowest_level;
7841                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7842                 BUG_ON(ret < 0);
7843                 path->lowest_level = 0;
7844                 btrfs_release_path(root, path);
7845                 return 0;
7846         }
7847
7848         mutex_lock(&root->fs_info->tree_reloc_mutex);
7849         ret = init_reloc_tree(trans, root);
7850         BUG_ON(ret);
7851         reloc_root = root->reloc_root;
7852
7853         shared_level = ref_path->shared_level;
7854         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7855
7856         keys = ref_path->node_keys;
7857         nodes = ref_path->new_nodes;
7858         memset(&keys[shared_level + 1], 0,
7859                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7860         memset(&nodes[shared_level + 1], 0,
7861                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7862
7863         if (nodes[lowest_level] == 0) {
7864                 path->lowest_level = lowest_level;
7865                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7866                                         0, 1);
7867                 BUG_ON(ret);
7868                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7869                         eb = path->nodes[level];
7870                         if (!eb || eb == reloc_root->node)
7871                                 break;
7872                         nodes[level] = eb->start;
7873                         if (level == 0)
7874                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7875                         else
7876                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7877                 }
7878                 if (nodes[0] &&
7879                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7880                         eb = path->nodes[0];
7881                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7882                                                       group, reloc_inode);
7883                         BUG_ON(ret);
7884                 }
7885                 btrfs_release_path(reloc_root, path);
7886         } else {
7887                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7888                                        lowest_level);
7889                 BUG_ON(ret);
7890         }
7891
7892         /*
7893          * replace tree blocks in the fs tree with tree blocks in
7894          * the reloc tree.
7895          */
7896         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7897         BUG_ON(ret < 0);
7898
7899         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7900                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7901                                         0, 0);
7902                 BUG_ON(ret);
7903                 extent_buffer_get(path->nodes[0]);
7904                 eb = path->nodes[0];
7905                 btrfs_release_path(reloc_root, path);
7906                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7907                 BUG_ON(ret);
7908                 free_extent_buffer(eb);
7909         }
7910
7911         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7912         path->lowest_level = 0;
7913         return 0;
7914 }
7915
7916 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7917                                         struct btrfs_root *root,
7918                                         struct btrfs_path *path,
7919                                         struct btrfs_key *first_key,
7920                                         struct btrfs_ref_path *ref_path)
7921 {
7922         int ret;
7923
7924         ret = relocate_one_path(trans, root, path, first_key,
7925                                 ref_path, NULL, NULL);
7926         BUG_ON(ret);
7927
7928         return 0;
7929 }
7930
7931 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7932                                     struct btrfs_root *extent_root,
7933                                     struct btrfs_path *path,
7934                                     struct btrfs_key *extent_key)
7935 {
7936         int ret;
7937
7938         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7939         if (ret)
7940                 goto out;
7941         ret = btrfs_del_item(trans, extent_root, path);
7942 out:
7943         btrfs_release_path(extent_root, path);
7944         return ret;
7945 }
7946
7947 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7948                                                 struct btrfs_ref_path *ref_path)
7949 {
7950         struct btrfs_key root_key;
7951
7952         root_key.objectid = ref_path->root_objectid;
7953         root_key.type = BTRFS_ROOT_ITEM_KEY;
7954         if (is_cowonly_root(ref_path->root_objectid))
7955                 root_key.offset = 0;
7956         else
7957                 root_key.offset = (u64)-1;
7958
7959         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7960 }
7961
7962 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7963                                         struct btrfs_path *path,
7964                                         struct btrfs_key *extent_key,
7965                                         struct btrfs_block_group_cache *group,
7966                                         struct inode *reloc_inode, int pass)
7967 {
7968         struct btrfs_trans_handle *trans;
7969         struct btrfs_root *found_root;
7970         struct btrfs_ref_path *ref_path = NULL;
7971         struct disk_extent *new_extents = NULL;
7972         int nr_extents = 0;
7973         int loops;
7974         int ret;
7975         int level;
7976         struct btrfs_key first_key;
7977         u64 prev_block = 0;
7978
7979
7980         trans = btrfs_start_transaction(extent_root, 1);
7981         BUG_ON(IS_ERR(trans));
7982
7983         if (extent_key->objectid == 0) {
7984                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7985                 goto out;
7986         }
7987
7988         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7989         if (!ref_path) {
7990                 ret = -ENOMEM;
7991                 goto out;
7992         }
7993
7994         for (loops = 0; ; loops++) {
7995                 if (loops == 0) {
7996                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7997                                                    extent_key->objectid);
7998                 } else {
7999                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
8000                 }
8001                 if (ret < 0)
8002                         goto out;
8003                 if (ret > 0)
8004                         break;
8005
8006                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
8007                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
8008                         continue;
8009
8010                 found_root = read_ref_root(extent_root->fs_info, ref_path);
8011                 BUG_ON(!found_root);
8012                 /*
8013                  * for reference counted tree, only process reference paths
8014                  * rooted at the latest committed root.
8015                  */
8016                 if (found_root->ref_cows &&
8017                     ref_path->root_generation != found_root->root_key.offset)
8018                         continue;
8019
8020                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8021                         if (pass == 0) {
8022                                 /*
8023                                  * copy data extents to new locations
8024                                  */
8025                                 u64 group_start = group->key.objectid;
8026                                 ret = relocate_data_extent(reloc_inode,
8027                                                            extent_key,
8028                                                            group_start);
8029                                 if (ret < 0)
8030                                         goto out;
8031                                 break;
8032                         }
8033                         level = 0;
8034                 } else {
8035                         level = ref_path->owner_objectid;
8036                 }
8037
8038                 if (prev_block != ref_path->nodes[level]) {
8039                         struct extent_buffer *eb;
8040                         u64 block_start = ref_path->nodes[level];
8041                         u64 block_size = btrfs_level_size(found_root, level);
8042
8043                         eb = read_tree_block(found_root, block_start,
8044                                              block_size, 0);
8045                         if (!eb) {
8046                                 ret = -EIO;
8047                                 goto out;
8048                         }
8049                         btrfs_tree_lock(eb);
8050                         BUG_ON(level != btrfs_header_level(eb));
8051
8052                         if (level == 0)
8053                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
8054                         else
8055                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
8056
8057                         btrfs_tree_unlock(eb);
8058                         free_extent_buffer(eb);
8059                         prev_block = block_start;
8060                 }
8061
8062                 mutex_lock(&extent_root->fs_info->trans_mutex);
8063                 btrfs_record_root_in_trans(found_root);
8064                 mutex_unlock(&extent_root->fs_info->trans_mutex);
8065                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
8066                         /*
8067                          * try to update data extent references while
8068                          * keeping metadata shared between snapshots.
8069                          */
8070                         if (pass == 1) {
8071                                 ret = relocate_one_path(trans, found_root,
8072                                                 path, &first_key, ref_path,
8073                                                 group, reloc_inode);
8074                                 if (ret < 0)
8075                                         goto out;
8076                                 continue;
8077                         }
8078                         /*
8079                          * use fallback method to process the remaining
8080                          * references.
8081                          */
8082                         if (!new_extents) {
8083                                 u64 group_start = group->key.objectid;
8084                                 new_extents = kmalloc(sizeof(*new_extents),
8085                                                       GFP_NOFS);
8086                                 if (!new_extents) {
8087                                         ret = -ENOMEM;
8088                                         goto out;
8089                                 }
8090                                 nr_extents = 1;
8091                                 ret = get_new_locations(reloc_inode,
8092                                                         extent_key,
8093                                                         group_start, 1,
8094                                                         &new_extents,
8095                                                         &nr_extents);
8096                                 if (ret)
8097                                         goto out;
8098                         }
8099                         ret = replace_one_extent(trans, found_root,
8100                                                 path, extent_key,
8101                                                 &first_key, ref_path,
8102                                                 new_extents, nr_extents);
8103                 } else {
8104                         ret = relocate_tree_block(trans, found_root, path,
8105                                                   &first_key, ref_path);
8106                 }
8107                 if (ret < 0)
8108                         goto out;
8109         }
8110         ret = 0;
8111 out:
8112         btrfs_end_transaction(trans, extent_root);
8113         kfree(new_extents);
8114         kfree(ref_path);
8115         return ret;
8116 }
8117 #endif
8118
8119 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8120 {
8121         u64 num_devices;
8122         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8123                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8124
8125         /*
8126          * we add in the count of missing devices because we want
8127          * to make sure that any RAID levels on a degraded FS
8128          * continue to be honored.
8129          */
8130         num_devices = root->fs_info->fs_devices->rw_devices +
8131                 root->fs_info->fs_devices->missing_devices;
8132
8133         if (num_devices == 1) {
8134                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8135                 stripped = flags & ~stripped;
8136
8137                 /* turn raid0 into single device chunks */
8138                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8139                         return stripped;
8140
8141                 /* turn mirroring into duplication */
8142                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8143                              BTRFS_BLOCK_GROUP_RAID10))
8144                         return stripped | BTRFS_BLOCK_GROUP_DUP;
8145                 return flags;
8146         } else {
8147                 /* they already had raid on here, just return */
8148                 if (flags & stripped)
8149                         return flags;
8150
8151                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8152                 stripped = flags & ~stripped;
8153
8154                 /* switch duplicated blocks with raid1 */
8155                 if (flags & BTRFS_BLOCK_GROUP_DUP)
8156                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
8157
8158                 /* turn single device chunks into raid0 */
8159                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8160         }
8161         return flags;
8162 }
8163
8164 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8165 {
8166         struct btrfs_space_info *sinfo = cache->space_info;
8167         u64 num_bytes;
8168         int ret = -ENOSPC;
8169
8170         if (cache->ro)
8171                 return 0;
8172
8173         spin_lock(&sinfo->lock);
8174         spin_lock(&cache->lock);
8175         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8176                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8177
8178         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8179             sinfo->bytes_may_use + sinfo->bytes_readonly +
8180             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8181                 sinfo->bytes_readonly += num_bytes;
8182                 sinfo->bytes_reserved += cache->reserved_pinned;
8183                 cache->reserved_pinned = 0;
8184                 cache->ro = 1;
8185                 ret = 0;
8186         }
8187
8188         spin_unlock(&cache->lock);
8189         spin_unlock(&sinfo->lock);
8190         return ret;
8191 }
8192
8193 int btrfs_set_block_group_ro(struct btrfs_root *root,
8194                              struct btrfs_block_group_cache *cache)
8195
8196 {
8197         struct btrfs_trans_handle *trans;
8198         u64 alloc_flags;
8199         int ret;
8200
8201         BUG_ON(cache->ro);
8202
8203         trans = btrfs_join_transaction(root);
8204         BUG_ON(IS_ERR(trans));
8205
8206         alloc_flags = update_block_group_flags(root, cache->flags);
8207         if (alloc_flags != cache->flags)
8208                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8209                                CHUNK_ALLOC_FORCE);
8210
8211         ret = set_block_group_ro(cache);
8212         if (!ret)
8213                 goto out;
8214         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8215         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8216                              CHUNK_ALLOC_FORCE);
8217         if (ret < 0)
8218                 goto out;
8219         ret = set_block_group_ro(cache);
8220 out:
8221         btrfs_end_transaction(trans, root);
8222         return ret;
8223 }
8224
8225 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8226                             struct btrfs_root *root, u64 type)
8227 {
8228         u64 alloc_flags = get_alloc_profile(root, type);
8229         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
8230                               CHUNK_ALLOC_FORCE);
8231 }
8232
8233 /*
8234  * helper to account the unused space of all the readonly block group in the
8235  * list. takes mirrors into account.
8236  */
8237 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8238 {
8239         struct btrfs_block_group_cache *block_group;
8240         u64 free_bytes = 0;
8241         int factor;
8242
8243         list_for_each_entry(block_group, groups_list, list) {
8244                 spin_lock(&block_group->lock);
8245
8246                 if (!block_group->ro) {
8247                         spin_unlock(&block_group->lock);
8248                         continue;
8249                 }
8250
8251                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8252                                           BTRFS_BLOCK_GROUP_RAID10 |
8253                                           BTRFS_BLOCK_GROUP_DUP))
8254                         factor = 2;
8255                 else
8256                         factor = 1;
8257
8258                 free_bytes += (block_group->key.offset -
8259                                btrfs_block_group_used(&block_group->item)) *
8260                                factor;
8261
8262                 spin_unlock(&block_group->lock);
8263         }
8264
8265         return free_bytes;
8266 }
8267
8268 /*
8269  * helper to account the unused space of all the readonly block group in the
8270  * space_info. takes mirrors into account.
8271  */
8272 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8273 {
8274         int i;
8275         u64 free_bytes = 0;
8276
8277         spin_lock(&sinfo->lock);
8278
8279         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8280                 if (!list_empty(&sinfo->block_groups[i]))
8281                         free_bytes += __btrfs_get_ro_block_group_free_space(
8282                                                 &sinfo->block_groups[i]);
8283
8284         spin_unlock(&sinfo->lock);
8285
8286         return free_bytes;
8287 }
8288
8289 int btrfs_set_block_group_rw(struct btrfs_root *root,
8290                               struct btrfs_block_group_cache *cache)
8291 {
8292         struct btrfs_space_info *sinfo = cache->space_info;
8293         u64 num_bytes;
8294
8295         BUG_ON(!cache->ro);
8296
8297         spin_lock(&sinfo->lock);
8298         spin_lock(&cache->lock);
8299         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8300                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8301         sinfo->bytes_readonly -= num_bytes;
8302         cache->ro = 0;
8303         spin_unlock(&cache->lock);
8304         spin_unlock(&sinfo->lock);
8305         return 0;
8306 }
8307
8308 /*
8309  * checks to see if its even possible to relocate this block group.
8310  *
8311  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8312  * ok to go ahead and try.
8313  */
8314 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8315 {
8316         struct btrfs_block_group_cache *block_group;
8317         struct btrfs_space_info *space_info;
8318         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8319         struct btrfs_device *device;
8320         int full = 0;
8321         int ret = 0;
8322
8323         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8324
8325         /* odd, couldn't find the block group, leave it alone */
8326         if (!block_group)
8327                 return -1;
8328
8329         /* no bytes used, we're good */
8330         if (!btrfs_block_group_used(&block_group->item))
8331                 goto out;
8332
8333         space_info = block_group->space_info;
8334         spin_lock(&space_info->lock);
8335
8336         full = space_info->full;
8337
8338         /*
8339          * if this is the last block group we have in this space, we can't
8340          * relocate it unless we're able to allocate a new chunk below.
8341          *
8342          * Otherwise, we need to make sure we have room in the space to handle
8343          * all of the extents from this block group.  If we can, we're good
8344          */
8345         if ((space_info->total_bytes != block_group->key.offset) &&
8346            (space_info->bytes_used + space_info->bytes_reserved +
8347             space_info->bytes_pinned + space_info->bytes_readonly +
8348             btrfs_block_group_used(&block_group->item) <
8349             space_info->total_bytes)) {
8350                 spin_unlock(&space_info->lock);
8351                 goto out;
8352         }
8353         spin_unlock(&space_info->lock);
8354
8355         /*
8356          * ok we don't have enough space, but maybe we have free space on our
8357          * devices to allocate new chunks for relocation, so loop through our
8358          * alloc devices and guess if we have enough space.  However, if we
8359          * were marked as full, then we know there aren't enough chunks, and we
8360          * can just return.
8361          */
8362         ret = -1;
8363         if (full)
8364                 goto out;
8365
8366         mutex_lock(&root->fs_info->chunk_mutex);
8367         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8368                 u64 min_free = btrfs_block_group_used(&block_group->item);
8369                 u64 dev_offset;
8370
8371                 /*
8372                  * check to make sure we can actually find a chunk with enough
8373                  * space to fit our block group in.
8374                  */
8375                 if (device->total_bytes > device->bytes_used + min_free) {
8376                         ret = find_free_dev_extent(NULL, device, min_free,
8377                                                    &dev_offset, NULL);
8378                         if (!ret)
8379                                 break;
8380                         ret = -1;
8381                 }
8382         }
8383         mutex_unlock(&root->fs_info->chunk_mutex);
8384 out:
8385         btrfs_put_block_group(block_group);
8386         return ret;
8387 }
8388
8389 static int find_first_block_group(struct btrfs_root *root,
8390                 struct btrfs_path *path, struct btrfs_key *key)
8391 {
8392         int ret = 0;
8393         struct btrfs_key found_key;
8394         struct extent_buffer *leaf;
8395         int slot;
8396
8397         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8398         if (ret < 0)
8399                 goto out;
8400
8401         while (1) {
8402                 slot = path->slots[0];
8403                 leaf = path->nodes[0];
8404                 if (slot >= btrfs_header_nritems(leaf)) {
8405                         ret = btrfs_next_leaf(root, path);
8406                         if (ret == 0)
8407                                 continue;
8408                         if (ret < 0)
8409                                 goto out;
8410                         break;
8411                 }
8412                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8413
8414                 if (found_key.objectid >= key->objectid &&
8415                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8416                         ret = 0;
8417                         goto out;
8418                 }
8419                 path->slots[0]++;
8420         }
8421 out:
8422         return ret;
8423 }
8424
8425 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8426 {
8427         struct btrfs_block_group_cache *block_group;
8428         u64 last = 0;
8429
8430         while (1) {
8431                 struct inode *inode;
8432
8433                 block_group = btrfs_lookup_first_block_group(info, last);
8434                 while (block_group) {
8435                         spin_lock(&block_group->lock);
8436                         if (block_group->iref)
8437                                 break;
8438                         spin_unlock(&block_group->lock);
8439                         block_group = next_block_group(info->tree_root,
8440                                                        block_group);
8441                 }
8442                 if (!block_group) {
8443                         if (last == 0)
8444                                 break;
8445                         last = 0;
8446                         continue;
8447                 }
8448
8449                 inode = block_group->inode;
8450                 block_group->iref = 0;
8451                 block_group->inode = NULL;
8452                 spin_unlock(&block_group->lock);
8453                 iput(inode);
8454                 last = block_group->key.objectid + block_group->key.offset;
8455                 btrfs_put_block_group(block_group);
8456         }
8457 }
8458
8459 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8460 {
8461         struct btrfs_block_group_cache *block_group;
8462         struct btrfs_space_info *space_info;
8463         struct btrfs_caching_control *caching_ctl;
8464         struct rb_node *n;
8465
8466         down_write(&info->extent_commit_sem);
8467         while (!list_empty(&info->caching_block_groups)) {
8468                 caching_ctl = list_entry(info->caching_block_groups.next,
8469                                          struct btrfs_caching_control, list);
8470                 list_del(&caching_ctl->list);
8471                 put_caching_control(caching_ctl);
8472         }
8473         up_write(&info->extent_commit_sem);
8474
8475         spin_lock(&info->block_group_cache_lock);
8476         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8477                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8478                                        cache_node);
8479                 rb_erase(&block_group->cache_node,
8480                          &info->block_group_cache_tree);
8481                 spin_unlock(&info->block_group_cache_lock);
8482
8483                 down_write(&block_group->space_info->groups_sem);
8484                 list_del(&block_group->list);
8485                 up_write(&block_group->space_info->groups_sem);
8486
8487                 if (block_group->cached == BTRFS_CACHE_STARTED)
8488                         wait_block_group_cache_done(block_group);
8489
8490                 /*
8491                  * We haven't cached this block group, which means we could
8492                  * possibly have excluded extents on this block group.
8493                  */
8494                 if (block_group->cached == BTRFS_CACHE_NO)
8495                         free_excluded_extents(info->extent_root, block_group);
8496
8497                 btrfs_remove_free_space_cache(block_group);
8498                 btrfs_put_block_group(block_group);
8499
8500                 spin_lock(&info->block_group_cache_lock);
8501         }
8502         spin_unlock(&info->block_group_cache_lock);
8503
8504         /* now that all the block groups are freed, go through and
8505          * free all the space_info structs.  This is only called during
8506          * the final stages of unmount, and so we know nobody is
8507          * using them.  We call synchronize_rcu() once before we start,
8508          * just to be on the safe side.
8509          */
8510         synchronize_rcu();
8511
8512         release_global_block_rsv(info);
8513
8514         while(!list_empty(&info->space_info)) {
8515                 space_info = list_entry(info->space_info.next,
8516                                         struct btrfs_space_info,
8517                                         list);
8518                 if (space_info->bytes_pinned > 0 ||
8519                     space_info->bytes_reserved > 0) {
8520                         WARN_ON(1);
8521                         dump_space_info(space_info, 0, 0);
8522                 }
8523                 list_del(&space_info->list);
8524                 kfree(space_info);
8525         }
8526         return 0;
8527 }
8528
8529 static void __link_block_group(struct btrfs_space_info *space_info,
8530                                struct btrfs_block_group_cache *cache)
8531 {
8532         int index = get_block_group_index(cache);
8533
8534         down_write(&space_info->groups_sem);
8535         list_add_tail(&cache->list, &space_info->block_groups[index]);
8536         up_write(&space_info->groups_sem);
8537 }
8538
8539 int btrfs_read_block_groups(struct btrfs_root *root)
8540 {
8541         struct btrfs_path *path;
8542         int ret;
8543         struct btrfs_block_group_cache *cache;
8544         struct btrfs_fs_info *info = root->fs_info;
8545         struct btrfs_space_info *space_info;
8546         struct btrfs_key key;
8547         struct btrfs_key found_key;
8548         struct extent_buffer *leaf;
8549         int need_clear = 0;
8550         u64 cache_gen;
8551
8552         root = info->extent_root;
8553         key.objectid = 0;
8554         key.offset = 0;
8555         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8556         path = btrfs_alloc_path();
8557         if (!path)
8558                 return -ENOMEM;
8559         path->reada = 1;
8560
8561         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8562         if (cache_gen != 0 &&
8563             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8564                 need_clear = 1;
8565         if (btrfs_test_opt(root, CLEAR_CACHE))
8566                 need_clear = 1;
8567         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8568                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8569
8570         while (1) {
8571                 ret = find_first_block_group(root, path, &key);
8572                 if (ret > 0)
8573                         break;
8574                 if (ret != 0)
8575                         goto error;
8576                 leaf = path->nodes[0];
8577                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8578                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8579                 if (!cache) {
8580                         ret = -ENOMEM;
8581                         goto error;
8582                 }
8583
8584                 atomic_set(&cache->count, 1);
8585                 spin_lock_init(&cache->lock);
8586                 spin_lock_init(&cache->tree_lock);
8587                 cache->fs_info = info;
8588                 INIT_LIST_HEAD(&cache->list);
8589                 INIT_LIST_HEAD(&cache->cluster_list);
8590
8591                 if (need_clear)
8592                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8593
8594                 /*
8595                  * we only want to have 32k of ram per block group for keeping
8596                  * track of free space, and if we pass 1/2 of that we want to
8597                  * start converting things over to using bitmaps
8598                  */
8599                 cache->extents_thresh = ((1024 * 32) / 2) /
8600                         sizeof(struct btrfs_free_space);
8601
8602                 read_extent_buffer(leaf, &cache->item,
8603                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8604                                    sizeof(cache->item));
8605                 memcpy(&cache->key, &found_key, sizeof(found_key));
8606
8607                 key.objectid = found_key.objectid + found_key.offset;
8608                 btrfs_release_path(root, path);
8609                 cache->flags = btrfs_block_group_flags(&cache->item);
8610                 cache->sectorsize = root->sectorsize;
8611
8612                 /*
8613                  * We need to exclude the super stripes now so that the space
8614                  * info has super bytes accounted for, otherwise we'll think
8615                  * we have more space than we actually do.
8616                  */
8617                 exclude_super_stripes(root, cache);
8618
8619                 /*
8620                  * check for two cases, either we are full, and therefore
8621                  * don't need to bother with the caching work since we won't
8622                  * find any space, or we are empty, and we can just add all
8623                  * the space in and be done with it.  This saves us _alot_ of
8624                  * time, particularly in the full case.
8625                  */
8626                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8627                         cache->last_byte_to_unpin = (u64)-1;
8628                         cache->cached = BTRFS_CACHE_FINISHED;
8629                         free_excluded_extents(root, cache);
8630                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8631                         cache->last_byte_to_unpin = (u64)-1;
8632                         cache->cached = BTRFS_CACHE_FINISHED;
8633                         add_new_free_space(cache, root->fs_info,
8634                                            found_key.objectid,
8635                                            found_key.objectid +
8636                                            found_key.offset);
8637                         free_excluded_extents(root, cache);
8638                 }
8639
8640                 ret = update_space_info(info, cache->flags, found_key.offset,
8641                                         btrfs_block_group_used(&cache->item),
8642                                         &space_info);
8643                 BUG_ON(ret);
8644                 cache->space_info = space_info;
8645                 spin_lock(&cache->space_info->lock);
8646                 cache->space_info->bytes_readonly += cache->bytes_super;
8647                 spin_unlock(&cache->space_info->lock);
8648
8649                 __link_block_group(space_info, cache);
8650
8651                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8652                 BUG_ON(ret);
8653
8654                 set_avail_alloc_bits(root->fs_info, cache->flags);
8655                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8656                         set_block_group_ro(cache);
8657         }
8658
8659         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8660                 if (!(get_alloc_profile(root, space_info->flags) &
8661                       (BTRFS_BLOCK_GROUP_RAID10 |
8662                        BTRFS_BLOCK_GROUP_RAID1 |
8663                        BTRFS_BLOCK_GROUP_DUP)))
8664                         continue;
8665                 /*
8666                  * avoid allocating from un-mirrored block group if there are
8667                  * mirrored block groups.
8668                  */
8669                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8670                         set_block_group_ro(cache);
8671                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8672                         set_block_group_ro(cache);
8673         }
8674
8675         init_global_block_rsv(info);
8676         ret = 0;
8677 error:
8678         btrfs_free_path(path);
8679         return ret;
8680 }
8681
8682 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8683                            struct btrfs_root *root, u64 bytes_used,
8684                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8685                            u64 size)
8686 {
8687         int ret;
8688         struct btrfs_root *extent_root;
8689         struct btrfs_block_group_cache *cache;
8690
8691         extent_root = root->fs_info->extent_root;
8692
8693         root->fs_info->last_trans_log_full_commit = trans->transid;
8694
8695         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8696         if (!cache)
8697                 return -ENOMEM;
8698
8699         cache->key.objectid = chunk_offset;
8700         cache->key.offset = size;
8701         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8702         cache->sectorsize = root->sectorsize;
8703         cache->fs_info = root->fs_info;
8704
8705         /*
8706          * we only want to have 32k of ram per block group for keeping track
8707          * of free space, and if we pass 1/2 of that we want to start
8708          * converting things over to using bitmaps
8709          */
8710         cache->extents_thresh = ((1024 * 32) / 2) /
8711                 sizeof(struct btrfs_free_space);
8712         atomic_set(&cache->count, 1);
8713         spin_lock_init(&cache->lock);
8714         spin_lock_init(&cache->tree_lock);
8715         INIT_LIST_HEAD(&cache->list);
8716         INIT_LIST_HEAD(&cache->cluster_list);
8717
8718         btrfs_set_block_group_used(&cache->item, bytes_used);
8719         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8720         cache->flags = type;
8721         btrfs_set_block_group_flags(&cache->item, type);
8722
8723         cache->last_byte_to_unpin = (u64)-1;
8724         cache->cached = BTRFS_CACHE_FINISHED;
8725         exclude_super_stripes(root, cache);
8726
8727         add_new_free_space(cache, root->fs_info, chunk_offset,
8728                            chunk_offset + size);
8729
8730         free_excluded_extents(root, cache);
8731
8732         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8733                                 &cache->space_info);
8734         BUG_ON(ret);
8735
8736         spin_lock(&cache->space_info->lock);
8737         cache->space_info->bytes_readonly += cache->bytes_super;
8738         spin_unlock(&cache->space_info->lock);
8739
8740         __link_block_group(cache->space_info, cache);
8741
8742         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8743         BUG_ON(ret);
8744
8745         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8746                                 sizeof(cache->item));
8747         BUG_ON(ret);
8748
8749         set_avail_alloc_bits(extent_root->fs_info, type);
8750
8751         return 0;
8752 }
8753
8754 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8755                              struct btrfs_root *root, u64 group_start)
8756 {
8757         struct btrfs_path *path;
8758         struct btrfs_block_group_cache *block_group;
8759         struct btrfs_free_cluster *cluster;
8760         struct btrfs_root *tree_root = root->fs_info->tree_root;
8761         struct btrfs_key key;
8762         struct inode *inode;
8763         int ret;
8764         int factor;
8765
8766         root = root->fs_info->extent_root;
8767
8768         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8769         BUG_ON(!block_group);
8770         BUG_ON(!block_group->ro);
8771
8772         /*
8773          * Free the reserved super bytes from this block group before
8774          * remove it.
8775          */
8776         free_excluded_extents(root, block_group);
8777
8778         memcpy(&key, &block_group->key, sizeof(key));
8779         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8780                                   BTRFS_BLOCK_GROUP_RAID1 |
8781                                   BTRFS_BLOCK_GROUP_RAID10))
8782                 factor = 2;
8783         else
8784                 factor = 1;
8785
8786         /* make sure this block group isn't part of an allocation cluster */
8787         cluster = &root->fs_info->data_alloc_cluster;
8788         spin_lock(&cluster->refill_lock);
8789         btrfs_return_cluster_to_free_space(block_group, cluster);
8790         spin_unlock(&cluster->refill_lock);
8791
8792         /*
8793          * make sure this block group isn't part of a metadata
8794          * allocation cluster
8795          */
8796         cluster = &root->fs_info->meta_alloc_cluster;
8797         spin_lock(&cluster->refill_lock);
8798         btrfs_return_cluster_to_free_space(block_group, cluster);
8799         spin_unlock(&cluster->refill_lock);
8800
8801         path = btrfs_alloc_path();
8802         BUG_ON(!path);
8803
8804         inode = lookup_free_space_inode(root, block_group, path);
8805         if (!IS_ERR(inode)) {
8806                 btrfs_orphan_add(trans, inode);
8807                 clear_nlink(inode);
8808                 /* One for the block groups ref */
8809                 spin_lock(&block_group->lock);
8810                 if (block_group->iref) {
8811                         block_group->iref = 0;
8812                         block_group->inode = NULL;
8813                         spin_unlock(&block_group->lock);
8814                         iput(inode);
8815                 } else {
8816                         spin_unlock(&block_group->lock);
8817                 }
8818                 /* One for our lookup ref */
8819                 iput(inode);
8820         }
8821
8822         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8823         key.offset = block_group->key.objectid;
8824         key.type = 0;
8825
8826         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8827         if (ret < 0)
8828                 goto out;
8829         if (ret > 0)
8830                 btrfs_release_path(tree_root, path);
8831         if (ret == 0) {
8832                 ret = btrfs_del_item(trans, tree_root, path);
8833                 if (ret)
8834                         goto out;
8835                 btrfs_release_path(tree_root, path);
8836         }
8837
8838         spin_lock(&root->fs_info->block_group_cache_lock);
8839         rb_erase(&block_group->cache_node,
8840                  &root->fs_info->block_group_cache_tree);
8841         spin_unlock(&root->fs_info->block_group_cache_lock);
8842
8843         down_write(&block_group->space_info->groups_sem);
8844         /*
8845          * we must use list_del_init so people can check to see if they
8846          * are still on the list after taking the semaphore
8847          */
8848         list_del_init(&block_group->list);
8849         up_write(&block_group->space_info->groups_sem);
8850
8851         if (block_group->cached == BTRFS_CACHE_STARTED)
8852                 wait_block_group_cache_done(block_group);
8853
8854         btrfs_remove_free_space_cache(block_group);
8855
8856         spin_lock(&block_group->space_info->lock);
8857         block_group->space_info->total_bytes -= block_group->key.offset;
8858         block_group->space_info->bytes_readonly -= block_group->key.offset;
8859         block_group->space_info->disk_total -= block_group->key.offset * factor;
8860         spin_unlock(&block_group->space_info->lock);
8861
8862         memcpy(&key, &block_group->key, sizeof(key));
8863
8864         btrfs_clear_space_info_full(root->fs_info);
8865
8866         btrfs_put_block_group(block_group);
8867         btrfs_put_block_group(block_group);
8868
8869         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8870         if (ret > 0)
8871                 ret = -EIO;
8872         if (ret < 0)
8873                 goto out;
8874
8875         ret = btrfs_del_item(trans, root, path);
8876 out:
8877         btrfs_free_path(path);
8878         return ret;
8879 }
8880
8881 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
8882 {
8883         struct btrfs_space_info *space_info;
8884         struct btrfs_super_block *disk_super;
8885         u64 features;
8886         u64 flags;
8887         int mixed = 0;
8888         int ret;
8889
8890         disk_super = &fs_info->super_copy;
8891         if (!btrfs_super_root(disk_super))
8892                 return 1;
8893
8894         features = btrfs_super_incompat_flags(disk_super);
8895         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
8896                 mixed = 1;
8897
8898         flags = BTRFS_BLOCK_GROUP_SYSTEM;
8899         ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8900         if (ret)
8901                 goto out;
8902
8903         if (mixed) {
8904                 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
8905                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8906         } else {
8907                 flags = BTRFS_BLOCK_GROUP_METADATA;
8908                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8909                 if (ret)
8910                         goto out;
8911
8912                 flags = BTRFS_BLOCK_GROUP_DATA;
8913                 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
8914         }
8915 out:
8916         return ret;
8917 }
8918
8919 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8920 {
8921         return unpin_extent_range(root, start, end);
8922 }
8923
8924 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8925                                u64 num_bytes, u64 *actual_bytes)
8926 {
8927         return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
8928 }
8929
8930 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
8931 {
8932         struct btrfs_fs_info *fs_info = root->fs_info;
8933         struct btrfs_block_group_cache *cache = NULL;
8934         u64 group_trimmed;
8935         u64 start;
8936         u64 end;
8937         u64 trimmed = 0;
8938         int ret = 0;
8939
8940         cache = btrfs_lookup_block_group(fs_info, range->start);
8941
8942         while (cache) {
8943                 if (cache->key.objectid >= (range->start + range->len)) {
8944                         btrfs_put_block_group(cache);
8945                         break;
8946                 }
8947
8948                 start = max(range->start, cache->key.objectid);
8949                 end = min(range->start + range->len,
8950                                 cache->key.objectid + cache->key.offset);
8951
8952                 if (end - start >= range->minlen) {
8953                         if (!block_group_cache_done(cache)) {
8954                                 ret = cache_block_group(cache, NULL, root, 0);
8955                                 if (!ret)
8956                                         wait_block_group_cache_done(cache);
8957                         }
8958                         ret = btrfs_trim_block_group(cache,
8959                                                      &group_trimmed,
8960                                                      start,
8961                                                      end,
8962                                                      range->minlen);
8963
8964                         trimmed += group_trimmed;
8965                         if (ret) {
8966                                 btrfs_put_block_group(cache);
8967                                 break;
8968                         }
8969                 }
8970
8971                 cache = next_block_group(fs_info->tree_root, cache);
8972         }
8973
8974         range->len = trimmed;
8975         return ret;
8976 }