fs/btrfs/extent-tree.c

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