SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 "compat.h"
  26 #include "hash.h"
  27 #include "ctree.h"
  28 #include "disk-io.h"
  29 #include "print-tree.h"
  30 #include "transaction.h"
  31 #include "volumes.h"
  32 #include "locking.h"
  33 #include "free-space-cache.h"
  34
  35 static int update_block_group(struct btrfs_trans_handle *trans,
  36                               struct btrfs_root *root,
  37                               u64 bytenr, u64 num_bytes, int alloc,
  38                               int mark_free);
  39 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
  40                                    u64 num_bytes, int reserve);
  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 pin_down_bytes(struct btrfs_trans_handle *trans,
  64                           struct btrfs_root *root,
  65                           struct btrfs_path *path,
  66                           u64 bytenr, u64 num_bytes,
  67                           int is_data, int reserved,
  68                           struct extent_buffer **must_clean);
  69 static int find_next_key(struct btrfs_path *path, int level,
  70                          struct btrfs_key *key);
  71 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
  72                             int dump_block_groups);
  73
  74 static noinline int
  75 block_group_cache_done(struct btrfs_block_group_cache *cache)
  76 {
  77         smp_mb();
  78         return cache->cached == BTRFS_CACHE_FINISHED;
  79 }
  80
  81 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  82 {
  83         return (cache->flags & bits) == bits;
  84 }
  85
  86 /*
  87  * this adds the block group to the fs_info rb tree for the block group
  88  * cache
  89  */
  90 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  91                                 struct btrfs_block_group_cache *block_group)
  92 {
  93         struct rb_node **p;
  94         struct rb_node *parent = NULL;
  95         struct btrfs_block_group_cache *cache;
  96
  97         spin_lock(&info->block_group_cache_lock);
  98         p = &info->block_group_cache_tree.rb_node;
  99
 100         while (*p) {
 101                 parent = *p;
 102                 cache = rb_entry(parent, struct btrfs_block_group_cache,
 103                                  cache_node);
 104                 if (block_group->key.objectid < cache->key.objectid) {
 105                         p = &(*p)->rb_left;
 106                 } else if (block_group->key.objectid > cache->key.objectid) {
 107                         p = &(*p)->rb_right;
 108                 } else {
 109                         spin_unlock(&info->block_group_cache_lock);
 110                         return -EEXIST;
 111                 }
 112         }
 113
 114         rb_link_node(&block_group->cache_node, parent, p);
 115         rb_insert_color(&block_group->cache_node,
 116                         &info->block_group_cache_tree);
 117         spin_unlock(&info->block_group_cache_lock);
 118
 119         return 0;
 120 }
 121
 122 /*
 123  * This will return the block group at or after bytenr if contains is 0, else
 124  * it will return the block group that contains the bytenr
 125  */
 126 static struct btrfs_block_group_cache *
 127 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 128                               int contains)
 129 {
 130         struct btrfs_block_group_cache *cache, *ret = NULL;
 131         struct rb_node *n;
 132         u64 end, start;
 133
 134         spin_lock(&info->block_group_cache_lock);
 135         n = info->block_group_cache_tree.rb_node;
 136
 137         while (n) {
 138                 cache = rb_entry(n, struct btrfs_block_group_cache,
 139                                  cache_node);
 140                 end = cache->key.objectid + cache->key.offset - 1;
 141                 start = cache->key.objectid;
 142
 143                 if (bytenr < start) {
 144                         if (!contains && (!ret || start < ret->key.objectid))
 145                                 ret = cache;
 146                         n = n->rb_left;
 147                 } else if (bytenr > start) {
 148                         if (contains && bytenr <= end) {
 149                                 ret = cache;
 150                                 break;
 151                         }
 152                         n = n->rb_right;
 153                 } else {
 154                         ret = cache;
 155                         break;
 156                 }
 157         }
 158         if (ret)
 159                 atomic_inc(&ret->count);
 160         spin_unlock(&info->block_group_cache_lock);
 161
 162         return ret;
 163 }
 164
 165 static int add_excluded_extent(struct btrfs_root *root,
 166                                u64 start, u64 num_bytes)
 167 {
 168         u64 end = start + num_bytes - 1;
 169         set_extent_bits(&root->fs_info->freed_extents[0],
 170                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 171         set_extent_bits(&root->fs_info->freed_extents[1],
 172                         start, end, EXTENT_UPTODATE, GFP_NOFS);
 173         return 0;
 174 }
 175
 176 static void free_excluded_extents(struct btrfs_root *root,
 177                                   struct btrfs_block_group_cache *cache)
 178 {
 179         u64 start, end;
 180
 181         start = cache->key.objectid;
 182         end = start + cache->key.offset - 1;
 183
 184         clear_extent_bits(&root->fs_info->freed_extents[0],
 185                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 186         clear_extent_bits(&root->fs_info->freed_extents[1],
 187                           start, end, EXTENT_UPTODATE, GFP_NOFS);
 188 }
 189
 190 static int exclude_super_stripes(struct btrfs_root *root,
 191                                  struct btrfs_block_group_cache *cache)
 192 {
 193         u64 bytenr;
 194         u64 *logical;
 195         int stripe_len;
 196         int i, nr, ret;
 197
 198         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 199                 bytenr = btrfs_sb_offset(i);
 200                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 201                                        cache->key.objectid, bytenr,
 202                                        0, &logical, &nr, &stripe_len);
 203                 BUG_ON(ret);
 204
 205                 while (nr--) {
 206                         cache->bytes_super += stripe_len;
 207                         ret = add_excluded_extent(root, logical[nr],
 208                                                   stripe_len);
 209                         BUG_ON(ret);
 210                 }
 211
 212                 kfree(logical);
 213         }
 214         return 0;
 215 }
 216
 217 static struct btrfs_caching_control *
 218 get_caching_control(struct btrfs_block_group_cache *cache)
 219 {
 220         struct btrfs_caching_control *ctl;
 221
 222         spin_lock(&cache->lock);
 223         if (cache->cached != BTRFS_CACHE_STARTED) {
 224                 spin_unlock(&cache->lock);
 225                 return NULL;
 226         }
 227
 228         ctl = cache->caching_ctl;
 229         atomic_inc(&ctl->count);
 230         spin_unlock(&cache->lock);
 231         return ctl;
 232 }
 233
 234 static void put_caching_control(struct btrfs_caching_control *ctl)
 235 {
 236         if (atomic_dec_and_test(&ctl->count))
 237                 kfree(ctl);
 238 }
 239
 240 /*
 241  * this is only called by cache_block_group, since we could have freed extents
 242  * we need to check the pinned_extents for any extents that can't be used yet
 243  * since their free space will be released as soon as the transaction commits.
 244  */
 245 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 246                               struct btrfs_fs_info *info, u64 start, u64 end)
 247 {
 248         u64 extent_start, extent_end, size, total_added = 0;
 249         int ret;
 250
 251         while (start < end) {
 252                 ret = find_first_extent_bit(info->pinned_extents, start,
 253                                             &extent_start, &extent_end,
 254                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 255                 if (ret)
 256                         break;
 257
 258                 if (extent_start == start) {
 259                         start = extent_end + 1;
 260                 } else if (extent_start > start && extent_start < end) {
 261                         size = extent_start - start;
 262                         total_added += size;
 263                         ret = btrfs_add_free_space(block_group, start,
 264                                                    size);
 265                         BUG_ON(ret);
 266                         start = extent_end + 1;
 267                 } else {
 268                         break;
 269                 }
 270         }
 271
 272         if (start < end) {
 273                 size = end - start;
 274                 total_added += size;
 275                 ret = btrfs_add_free_space(block_group, start, size);
 276                 BUG_ON(ret);
 277         }
 278
 279         return total_added;
 280 }
 281
 282 static int caching_kthread(void *data)
 283 {
 284         struct btrfs_block_group_cache *block_group = data;
 285         struct btrfs_fs_info *fs_info = block_group->fs_info;
 286         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 287         struct btrfs_root *extent_root = fs_info->extent_root;
 288         struct btrfs_path *path;
 289         struct extent_buffer *leaf;
 290         struct btrfs_key key;
 291         u64 total_found = 0;
 292         u64 last = 0;
 293         u32 nritems;
 294         int ret = 0;
 295
 296         path = btrfs_alloc_path();
 297         if (!path)
 298                 return -ENOMEM;
 299
 300         exclude_super_stripes(extent_root, block_group);
 301         spin_lock(&block_group->space_info->lock);
 302         block_group->space_info->bytes_super += block_group->bytes_super;
 303         spin_unlock(&block_group->space_info->lock);
 304
 305         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 306
 307         /*
 308          * We don't want to deadlock with somebody trying to allocate a new
 309          * extent for the extent root while also trying to search the extent
 310          * root to add free space.  So we skip locking and search the commit
 311          * root, since its read-only
 312          */
 313         path->skip_locking = 1;
 314         path->search_commit_root = 1;
 315         path->reada = 2;
 316
 317         key.objectid = last;
 318         key.offset = 0;
 319         key.type = BTRFS_EXTENT_ITEM_KEY;
 320 again:
 321         mutex_lock(&caching_ctl->mutex);
 322         /* need to make sure the commit_root doesn't disappear */
 323         down_read(&fs_info->extent_commit_sem);
 324
 325         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 326         if (ret < 0)
 327                 goto err;
 328
 329         leaf = path->nodes[0];
 330         nritems = btrfs_header_nritems(leaf);
 331
 332         while (1) {
 333                 smp_mb();
 334                 if (fs_info->closing > 1) {
 335                         last = (u64)-1;
 336                         break;
 337                 }
 338
 339                 if (path->slots[0] < nritems) {
 340                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 341                 } else {
 342                         ret = find_next_key(path, 0, &key);
 343                         if (ret)
 344                                 break;
 345
 346                         caching_ctl->progress = last;
 347                         btrfs_release_path(extent_root, path);
 348                         up_read(&fs_info->extent_commit_sem);
 349                         mutex_unlock(&caching_ctl->mutex);
 350                         if (btrfs_transaction_in_commit(fs_info))
 351                                 schedule_timeout(1);
 352                         else
 353                                 cond_resched();
 354                         goto again;
 355                 }
 356
 357                 if (key.objectid < block_group->key.objectid) {
 358                         path->slots[0]++;
 359                         continue;
 360                 }
 361
 362                 if (key.objectid >= block_group->key.objectid +
 363                     block_group->key.offset)
 364                         break;
 365
 366                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 367                         total_found += add_new_free_space(block_group,
 368                                                           fs_info, last,
 369                                                           key.objectid);
 370                         last = key.objectid + key.offset;
 371
 372                         if (total_found > (1024 * 1024 * 2)) {
 373                                 total_found = 0;
 374                                 wake_up(&caching_ctl->wait);
 375                         }
 376                 }
 377                 path->slots[0]++;
 378         }
 379         ret = 0;
 380
 381         total_found += add_new_free_space(block_group, fs_info, last,
 382                                           block_group->key.objectid +
 383                                           block_group->key.offset);
 384         caching_ctl->progress = (u64)-1;
 385
 386         spin_lock(&block_group->lock);
 387         block_group->caching_ctl = NULL;
 388         block_group->cached = BTRFS_CACHE_FINISHED;
 389         spin_unlock(&block_group->lock);
 390
 391 err:
 392         btrfs_free_path(path);
 393         up_read(&fs_info->extent_commit_sem);
 394
 395         free_excluded_extents(extent_root, block_group);
 396
 397         mutex_unlock(&caching_ctl->mutex);
 398         wake_up(&caching_ctl->wait);
 399
 400         put_caching_control(caching_ctl);
 401         atomic_dec(&block_group->space_info->caching_threads);
 402         return 0;
 403 }
 404
 405 static int cache_block_group(struct btrfs_block_group_cache *cache)
 406 {
 407         struct btrfs_fs_info *fs_info = cache->fs_info;
 408         struct btrfs_caching_control *caching_ctl;
 409         struct task_struct *tsk;
 410         int ret = 0;
 411
 412         smp_mb();
 413         if (cache->cached != BTRFS_CACHE_NO)
 414                 return 0;
 415
 416         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 417         BUG_ON(!caching_ctl);
 418
 419         INIT_LIST_HEAD(&caching_ctl->list);
 420         mutex_init(&caching_ctl->mutex);
 421         init_waitqueue_head(&caching_ctl->wait);
 422         caching_ctl->block_group = cache;
 423         caching_ctl->progress = cache->key.objectid;
 424         /* one for caching kthread, one for caching block group list */
 425         atomic_set(&caching_ctl->count, 2);
 426
 427         spin_lock(&cache->lock);
 428         if (cache->cached != BTRFS_CACHE_NO) {
 429                 spin_unlock(&cache->lock);
 430                 kfree(caching_ctl);
 431                 return 0;
 432         }
 433         cache->caching_ctl = caching_ctl;
 434         cache->cached = BTRFS_CACHE_STARTED;
 435         spin_unlock(&cache->lock);
 436
 437         down_write(&fs_info->extent_commit_sem);
 438         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 439         up_write(&fs_info->extent_commit_sem);
 440
 441         atomic_inc(&cache->space_info->caching_threads);
 442
 443         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 444                           cache->key.objectid);
 445         if (IS_ERR(tsk)) {
 446                 ret = PTR_ERR(tsk);
 447                 printk(KERN_ERR "error running thread %d\n", ret);
 448                 BUG();
 449         }
 450
 451         return ret;
 452 }
 453
 454 /*
 455  * return the block group that starts at or after bytenr
 456  */
 457 static struct btrfs_block_group_cache *
 458 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 459 {
 460         struct btrfs_block_group_cache *cache;
 461
 462         cache = block_group_cache_tree_search(info, bytenr, 0);
 463
 464         return cache;
 465 }
 466
 467 /*
 468  * return the block group that contains the given bytenr
 469  */
 470 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 471                                                  struct btrfs_fs_info *info,
 472                                                  u64 bytenr)
 473 {
 474         struct btrfs_block_group_cache *cache;
 475
 476         cache = block_group_cache_tree_search(info, bytenr, 1);
 477
 478         return cache;
 479 }
 480
 481 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
 482 {
 483         if (atomic_dec_and_test(&cache->count))
 484                 kfree(cache);
 485 }
 486
 487 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 488                                                   u64 flags)
 489 {
 490         struct list_head *head = &info->space_info;
 491         struct btrfs_space_info *found;
 492
 493         rcu_read_lock();
 494         list_for_each_entry_rcu(found, head, list) {
 495                 if (found->flags == flags) {
 496                         rcu_read_unlock();
 497                         return found;
 498                 }
 499         }
 500         rcu_read_unlock();
 501         return NULL;
 502 }
 503
 504 /*
 505  * after adding space to the filesystem, we need to clear the full flags
 506  * on all the space infos.
 507  */
 508 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 509 {
 510         struct list_head *head = &info->space_info;
 511         struct btrfs_space_info *found;
 512
 513         rcu_read_lock();
 514         list_for_each_entry_rcu(found, head, list)
 515                 found->full = 0;
 516         rcu_read_unlock();
 517 }
 518
 519 static u64 div_factor(u64 num, int factor)
 520 {
 521         if (factor == 10)
 522                 return num;
 523         num *= factor;
 524         do_div(num, 10);
 525         return num;
 526 }
 527
 528 u64 btrfs_find_block_group(struct btrfs_root *root,
 529                            u64 search_start, u64 search_hint, int owner)
 530 {
 531         struct btrfs_block_group_cache *cache;
 532         u64 used;
 533         u64 last = max(search_hint, search_start);
 534         u64 group_start = 0;
 535         int full_search = 0;
 536         int factor = 9;
 537         int wrapped = 0;
 538 again:
 539         while (1) {
 540                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 541                 if (!cache)
 542                         break;
 543
 544                 spin_lock(&cache->lock);
 545                 last = cache->key.objectid + cache->key.offset;
 546                 used = btrfs_block_group_used(&cache->item);
 547
 548                 if ((full_search || !cache->ro) &&
 549                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 550                         if (used + cache->pinned + cache->reserved <
 551                             div_factor(cache->key.offset, factor)) {
 552                                 group_start = cache->key.objectid;
 553                                 spin_unlock(&cache->lock);
 554                                 btrfs_put_block_group(cache);
 555                                 goto found;
 556                         }
 557                 }
 558                 spin_unlock(&cache->lock);
 559                 btrfs_put_block_group(cache);
 560                 cond_resched();
 561         }
 562         if (!wrapped) {
 563                 last = search_start;
 564                 wrapped = 1;
 565                 goto again;
 566         }
 567         if (!full_search && factor < 10) {
 568                 last = search_start;
 569                 full_search = 1;
 570                 factor = 10;
 571                 goto again;
 572         }
 573 found:
 574         return group_start;
 575 }
 576
 577 /* simple helper to search for an existing extent at a given offset */
 578 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 579 {
 580         int ret;
 581         struct btrfs_key key;
 582         struct btrfs_path *path;
 583
 584         path = btrfs_alloc_path();
 585         BUG_ON(!path);
 586         key.objectid = start;
 587         key.offset = len;
 588         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 589         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 590                                 0, 0);
 591         btrfs_free_path(path);
 592         return ret;
 593 }
 594
 595 /*
 596  * Back reference rules.  Back refs have three main goals:
 597  *
 598  * 1) differentiate between all holders of references to an extent so that
 599  *    when a reference is dropped we can make sure it was a valid reference
 600  *    before freeing the extent.
 601  *
 602  * 2) Provide enough information to quickly find the holders of an extent
 603  *    if we notice a given block is corrupted or bad.
 604  *
 605  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 606  *    maintenance.  This is actually the same as #2, but with a slightly
 607  *    different use case.
 608  *
 609  * There are two kinds of back refs. The implicit back refs is optimized
 610  * for pointers in non-shared tree blocks. For a given pointer in a block,
 611  * back refs of this kind provide information about the block's owner tree
 612  * and the pointer's key. These information allow us to find the block by
 613  * b-tree searching. The full back refs is for pointers in tree blocks not
 614  * referenced by their owner trees. The location of tree block is recorded
 615  * in the back refs. Actually the full back refs is generic, and can be
 616  * used in all cases the implicit back refs is used. The major shortcoming
 617  * of the full back refs is its overhead. Every time a tree block gets
 618  * COWed, we have to update back refs entry for all pointers in it.
 619  *
 620  * For a newly allocated tree block, we use implicit back refs for
 621  * pointers in it. This means most tree related operations only involve
 622  * implicit back refs. For a tree block created in old transaction, the
 623  * only way to drop a reference to it is COW it. So we can detect the
 624  * event that tree block loses its owner tree's reference and do the
 625  * back refs conversion.
 626  *
 627  * When a tree block is COW'd through a tree, there are four cases:
 628  *
 629  * The reference count of the block is one and the tree is the block's
 630  * owner tree. Nothing to do in this case.
 631  *
 632  * The reference count of the block is one and the tree is not the
 633  * block's owner tree. In this case, full back refs is used for pointers
 634  * in the block. Remove these full back refs, add implicit back refs for
 635  * every pointers in the new block.
 636  *
 637  * The reference count of the block is greater than one and the tree is
 638  * the block's owner tree. In this case, implicit back refs is used for
 639  * pointers in the block. Add full back refs for every pointers in the
 640  * block, increase lower level extents' reference counts. The original
 641  * implicit back refs are entailed to the new block.
 642  *
 643  * The reference count of the block is greater than one and the tree is
 644  * not the block's owner tree. Add implicit back refs for every pointer in
 645  * the new block, increase lower level extents' reference count.
 646  *
 647  * Back Reference Key composing:
 648  *
 649  * The key objectid corresponds to the first byte in the extent,
 650  * The key type is used to differentiate between types of back refs.
 651  * There are different meanings of the key offset for different types
 652  * of back refs.
 653  *
 654  * File extents can be referenced by:
 655  *
 656  * - multiple snapshots, subvolumes, or different generations in one subvol
 657  * - different files inside a single subvolume
 658  * - different offsets inside a file (bookend extents in file.c)
 659  *
 660  * The extent ref structure for the implicit back refs has fields for:
 661  *
 662  * - Objectid of the subvolume root
 663  * - objectid of the file holding the reference
 664  * - original offset in the file
 665  * - how many bookend extents
 666  *
 667  * The key offset for the implicit back refs is hash of the first
 668  * three fields.
 669  *
 670  * The extent ref structure for the full back refs has field for:
 671  *
 672  * - number of pointers in the tree leaf
 673  *
 674  * The key offset for the implicit back refs is the first byte of
 675  * the tree leaf
 676  *
 677  * When a file extent is allocated, The implicit back refs is used.
 678  * the fields are filled in:
 679  *
 680  *     (root_key.objectid, inode objectid, offset in file, 1)
 681  *
 682  * When a file extent is removed file truncation, we find the
 683  * corresponding implicit back refs and check the following fields:
 684  *
 685  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 686  *
 687  * Btree extents can be referenced by:
 688  *
 689  * - Different subvolumes
 690  *
 691  * Both the implicit back refs and the full back refs for tree blocks
 692  * only consist of key. The key offset for the implicit back refs is
 693  * objectid of block's owner tree. The key offset for the full back refs
 694  * is the first byte of parent block.
 695  *
 696  * When implicit back refs is used, information about the lowest key and
 697  * level of the tree block are required. These information are stored in
 698  * tree block info structure.
 699  */
 700
 701 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 702 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 703                                   struct btrfs_root *root,
 704                                   struct btrfs_path *path,
 705                                   u64 owner, u32 extra_size)
 706 {
 707         struct btrfs_extent_item *item;
 708         struct btrfs_extent_item_v0 *ei0;
 709         struct btrfs_extent_ref_v0 *ref0;
 710         struct btrfs_tree_block_info *bi;
 711         struct extent_buffer *leaf;
 712         struct btrfs_key key;
 713         struct btrfs_key found_key;
 714         u32 new_size = sizeof(*item);
 715         u64 refs;
 716         int ret;
 717
 718         leaf = path->nodes[0];
 719         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 720
 721         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 722         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 723                              struct btrfs_extent_item_v0);
 724         refs = btrfs_extent_refs_v0(leaf, ei0);
 725
 726         if (owner == (u64)-1) {
 727                 while (1) {
 728                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 729                                 ret = btrfs_next_leaf(root, path);
 730                                 if (ret < 0)
 731                                         return ret;
 732                                 BUG_ON(ret > 0);
 733                                 leaf = path->nodes[0];
 734                         }
 735                         btrfs_item_key_to_cpu(leaf, &found_key,
 736                                               path->slots[0]);
 737                         BUG_ON(key.objectid != found_key.objectid);
 738                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 739                                 path->slots[0]++;
 740                                 continue;
 741                         }
 742                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 743                                               struct btrfs_extent_ref_v0);
 744                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 745                         break;
 746                 }
 747         }
 748         btrfs_release_path(root, path);
 749
 750         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 751                 new_size += sizeof(*bi);
 752
 753         new_size -= sizeof(*ei0);
 754         ret = btrfs_search_slot(trans, root, &key, path,
 755                                 new_size + extra_size, 1);
 756         if (ret < 0)
 757                 return ret;
 758         BUG_ON(ret);
 759
 760         ret = btrfs_extend_item(trans, root, path, new_size);
 761         BUG_ON(ret);
 762
 763         leaf = path->nodes[0];
 764         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 765         btrfs_set_extent_refs(leaf, item, refs);
 766         /* FIXME: get real generation */
 767         btrfs_set_extent_generation(leaf, item, 0);
 768         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 769                 btrfs_set_extent_flags(leaf, item,
 770                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 771                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 772                 bi = (struct btrfs_tree_block_info *)(item + 1);
 773                 /* FIXME: get first key of the block */
 774                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 775                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 776         } else {
 777                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 778         }
 779         btrfs_mark_buffer_dirty(leaf);
 780         return 0;
 781 }
 782 #endif
 783
 784 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 785 {
 786         u32 high_crc = ~(u32)0;
 787         u32 low_crc = ~(u32)0;
 788         __le64 lenum;
 789
 790         lenum = cpu_to_le64(root_objectid);
 791         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 792         lenum = cpu_to_le64(owner);
 793         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 794         lenum = cpu_to_le64(offset);
 795         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 796
 797         return ((u64)high_crc << 31) ^ (u64)low_crc;
 798 }
 799
 800 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 801                                      struct btrfs_extent_data_ref *ref)
 802 {
 803         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 804                                     btrfs_extent_data_ref_objectid(leaf, ref),
 805                                     btrfs_extent_data_ref_offset(leaf, ref));
 806 }
 807
 808 static int match_extent_data_ref(struct extent_buffer *leaf,
 809                                  struct btrfs_extent_data_ref *ref,
 810                                  u64 root_objectid, u64 owner, u64 offset)
 811 {
 812         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 813             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 814             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 815                 return 0;
 816         return 1;
 817 }
 818
 819 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 820                                            struct btrfs_root *root,
 821                                            struct btrfs_path *path,
 822                                            u64 bytenr, u64 parent,
 823                                            u64 root_objectid,
 824                                            u64 owner, u64 offset)
 825 {
 826         struct btrfs_key key;
 827         struct btrfs_extent_data_ref *ref;
 828         struct extent_buffer *leaf;
 829         u32 nritems;
 830         int ret;
 831         int recow;
 832         int err = -ENOENT;
 833
 834         key.objectid = bytenr;
 835         if (parent) {
 836                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 837                 key.offset = parent;
 838         } else {
 839                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 840                 key.offset = hash_extent_data_ref(root_objectid,
 841                                                   owner, offset);
 842         }
 843 again:
 844         recow = 0;
 845         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 846         if (ret < 0) {
 847                 err = ret;
 848                 goto fail;
 849         }
 850
 851         if (parent) {
 852                 if (!ret)
 853                         return 0;
 854 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 855                 key.type = BTRFS_EXTENT_REF_V0_KEY;
 856                 btrfs_release_path(root, path);
 857                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 858                 if (ret < 0) {
 859                         err = ret;
 860                         goto fail;
 861                 }
 862                 if (!ret)
 863                         return 0;
 864 #endif
 865                 goto fail;
 866         }
 867
 868         leaf = path->nodes[0];
 869         nritems = btrfs_header_nritems(leaf);
 870         while (1) {
 871                 if (path->slots[0] >= nritems) {
 872                         ret = btrfs_next_leaf(root, path);
 873                         if (ret < 0)
 874                                 err = ret;
 875                         if (ret)
 876                                 goto fail;
 877
 878                         leaf = path->nodes[0];
 879                         nritems = btrfs_header_nritems(leaf);
 880                         recow = 1;
 881                 }
 882
 883                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 884                 if (key.objectid != bytenr ||
 885                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
 886                         goto fail;
 887
 888                 ref = btrfs_item_ptr(leaf, path->slots[0],
 889                                      struct btrfs_extent_data_ref);
 890
 891                 if (match_extent_data_ref(leaf, ref, root_objectid,
 892                                           owner, offset)) {
 893                         if (recow) {
 894                                 btrfs_release_path(root, path);
 895                                 goto again;
 896                         }
 897                         err = 0;
 898                         break;
 899                 }
 900                 path->slots[0]++;
 901         }
 902 fail:
 903         return err;
 904 }
 905
 906 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
 907                                            struct btrfs_root *root,
 908                                            struct btrfs_path *path,
 909                                            u64 bytenr, u64 parent,
 910                                            u64 root_objectid, u64 owner,
 911                                            u64 offset, int refs_to_add)
 912 {
 913         struct btrfs_key key;
 914         struct extent_buffer *leaf;
 915         u32 size;
 916         u32 num_refs;
 917         int ret;
 918
 919         key.objectid = bytenr;
 920         if (parent) {
 921                 key.type = BTRFS_SHARED_DATA_REF_KEY;
 922                 key.offset = parent;
 923                 size = sizeof(struct btrfs_shared_data_ref);
 924         } else {
 925                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
 926                 key.offset = hash_extent_data_ref(root_objectid,
 927                                                   owner, offset);
 928                 size = sizeof(struct btrfs_extent_data_ref);
 929         }
 930
 931         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
 932         if (ret && ret != -EEXIST)
 933                 goto fail;
 934
 935         leaf = path->nodes[0];
 936         if (parent) {
 937                 struct btrfs_shared_data_ref *ref;
 938                 ref = btrfs_item_ptr(leaf, path->slots[0],
 939                                      struct btrfs_shared_data_ref);
 940                 if (ret == 0) {
 941                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
 942                 } else {
 943                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
 944                         num_refs += refs_to_add;
 945                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
 946                 }
 947         } else {
 948                 struct btrfs_extent_data_ref *ref;
 949                 while (ret == -EEXIST) {
 950                         ref = btrfs_item_ptr(leaf, path->slots[0],
 951                                              struct btrfs_extent_data_ref);
 952                         if (match_extent_data_ref(leaf, ref, root_objectid,
 953                                                   owner, offset))
 954                                 break;
 955                         btrfs_release_path(root, path);
 956                         key.offset++;
 957                         ret = btrfs_insert_empty_item(trans, root, path, &key,
 958                                                       size);
 959                         if (ret && ret != -EEXIST)
 960                                 goto fail;
 961
 962                         leaf = path->nodes[0];
 963                 }
 964                 ref = btrfs_item_ptr(leaf, path->slots[0],
 965                                      struct btrfs_extent_data_ref);
 966                 if (ret == 0) {
 967                         btrfs_set_extent_data_ref_root(leaf, ref,
 968                                                        root_objectid);
 969                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
 970                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
 971                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
 972                 } else {
 973                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
 974                         num_refs += refs_to_add;
 975                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
 976                 }
 977         }
 978         btrfs_mark_buffer_dirty(leaf);
 979         ret = 0;
 980 fail:
 981         btrfs_release_path(root, path);
 982         return ret;
 983 }
 984
 985 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
 986                                            struct btrfs_root *root,
 987                                            struct btrfs_path *path,
 988                                            int refs_to_drop)
 989 {
 990         struct btrfs_key key;
 991         struct btrfs_extent_data_ref *ref1 = NULL;
 992         struct btrfs_shared_data_ref *ref2 = NULL;
 993         struct extent_buffer *leaf;
 994         u32 num_refs = 0;
 995         int ret = 0;
 996
 997         leaf = path->nodes[0];
 998         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 999
1000         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1001                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1002                                       struct btrfs_extent_data_ref);
1003                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1004         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1005                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1006                                       struct btrfs_shared_data_ref);
1007                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1008 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1009         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1010                 struct btrfs_extent_ref_v0 *ref0;
1011                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1012                                       struct btrfs_extent_ref_v0);
1013                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1014 #endif
1015         } else {
1016                 BUG();
1017         }
1018
1019         BUG_ON(num_refs < refs_to_drop);
1020         num_refs -= refs_to_drop;
1021
1022         if (num_refs == 0) {
1023                 ret = btrfs_del_item(trans, root, path);
1024         } else {
1025                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1026                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1027                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1028                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1029 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1030                 else {
1031                         struct btrfs_extent_ref_v0 *ref0;
1032                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1033                                         struct btrfs_extent_ref_v0);
1034                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1035                 }
1036 #endif
1037                 btrfs_mark_buffer_dirty(leaf);
1038         }
1039         return ret;
1040 }
1041
1042 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1043                                           struct btrfs_path *path,
1044                                           struct btrfs_extent_inline_ref *iref)
1045 {
1046         struct btrfs_key key;
1047         struct extent_buffer *leaf;
1048         struct btrfs_extent_data_ref *ref1;
1049         struct btrfs_shared_data_ref *ref2;
1050         u32 num_refs = 0;
1051
1052         leaf = path->nodes[0];
1053         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1054         if (iref) {
1055                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1056                     BTRFS_EXTENT_DATA_REF_KEY) {
1057                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1058                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1059                 } else {
1060                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1061                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1062                 }
1063         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1064                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1065                                       struct btrfs_extent_data_ref);
1066                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1067         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1068                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1069                                       struct btrfs_shared_data_ref);
1070                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1071 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1072         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1073                 struct btrfs_extent_ref_v0 *ref0;
1074                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1075                                       struct btrfs_extent_ref_v0);
1076                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1077 #endif
1078         } else {
1079                 WARN_ON(1);
1080         }
1081         return num_refs;
1082 }
1083
1084 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1085                                           struct btrfs_root *root,
1086                                           struct btrfs_path *path,
1087                                           u64 bytenr, u64 parent,
1088                                           u64 root_objectid)
1089 {
1090         struct btrfs_key key;
1091         int ret;
1092
1093         key.objectid = bytenr;
1094         if (parent) {
1095                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1096                 key.offset = parent;
1097         } else {
1098                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1099                 key.offset = root_objectid;
1100         }
1101
1102         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1103         if (ret > 0)
1104                 ret = -ENOENT;
1105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1106         if (ret == -ENOENT && parent) {
1107                 btrfs_release_path(root, path);
1108                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1109                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1110                 if (ret > 0)
1111                         ret = -ENOENT;
1112         }
1113 #endif
1114         return ret;
1115 }
1116
1117 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1118                                           struct btrfs_root *root,
1119                                           struct btrfs_path *path,
1120                                           u64 bytenr, u64 parent,
1121                                           u64 root_objectid)
1122 {
1123         struct btrfs_key key;
1124         int ret;
1125
1126         key.objectid = bytenr;
1127         if (parent) {
1128                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1129                 key.offset = parent;
1130         } else {
1131                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1132                 key.offset = root_objectid;
1133         }
1134
1135         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1136         btrfs_release_path(root, path);
1137         return ret;
1138 }
1139
1140 static inline int extent_ref_type(u64 parent, u64 owner)
1141 {
1142         int type;
1143         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1144                 if (parent > 0)
1145                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1146                 else
1147                         type = BTRFS_TREE_BLOCK_REF_KEY;
1148         } else {
1149                 if (parent > 0)
1150                         type = BTRFS_SHARED_DATA_REF_KEY;
1151                 else
1152                         type = BTRFS_EXTENT_DATA_REF_KEY;
1153         }
1154         return type;
1155 }
1156
1157 static int find_next_key(struct btrfs_path *path, int level,
1158                          struct btrfs_key *key)
1159
1160 {
1161         for (; level < BTRFS_MAX_LEVEL; level++) {
1162                 if (!path->nodes[level])
1163                         break;
1164                 if (path->slots[level] + 1 >=
1165                     btrfs_header_nritems(path->nodes[level]))
1166                         continue;
1167                 if (level == 0)
1168                         btrfs_item_key_to_cpu(path->nodes[level], key,
1169                                               path->slots[level] + 1);
1170                 else
1171                         btrfs_node_key_to_cpu(path->nodes[level], key,
1172                                               path->slots[level] + 1);
1173                 return 0;
1174         }
1175         return 1;
1176 }
1177
1178 /*
1179  * look for inline back ref. if back ref is found, *ref_ret is set
1180  * to the address of inline back ref, and 0 is returned.
1181  *
1182  * if back ref isn't found, *ref_ret is set to the address where it
1183  * should be inserted, and -ENOENT is returned.
1184  *
1185  * if insert is true and there are too many inline back refs, the path
1186  * points to the extent item, and -EAGAIN is returned.
1187  *
1188  * NOTE: inline back refs are ordered in the same way that back ref
1189  *       items in the tree are ordered.
1190  */
1191 static noinline_for_stack
1192 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1193                                  struct btrfs_root *root,
1194                                  struct btrfs_path *path,
1195                                  struct btrfs_extent_inline_ref **ref_ret,
1196                                  u64 bytenr, u64 num_bytes,
1197                                  u64 parent, u64 root_objectid,
1198                                  u64 owner, u64 offset, int insert)
1199 {
1200         struct btrfs_key key;
1201         struct extent_buffer *leaf;
1202         struct btrfs_extent_item *ei;
1203         struct btrfs_extent_inline_ref *iref;
1204         u64 flags;
1205         u64 item_size;
1206         unsigned long ptr;
1207         unsigned long end;
1208         int extra_size;
1209         int type;
1210         int want;
1211         int ret;
1212         int err = 0;
1213
1214         key.objectid = bytenr;
1215         key.type = BTRFS_EXTENT_ITEM_KEY;
1216         key.offset = num_bytes;
1217
1218         want = extent_ref_type(parent, owner);
1219         if (insert) {
1220                 extra_size = btrfs_extent_inline_ref_size(want);
1221                 path->keep_locks = 1;
1222         } else
1223                 extra_size = -1;
1224         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1225         if (ret < 0) {
1226                 err = ret;
1227                 goto out;
1228         }
1229         BUG_ON(ret);
1230
1231         leaf = path->nodes[0];
1232         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1233 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1234         if (item_size < sizeof(*ei)) {
1235                 if (!insert) {
1236                         err = -ENOENT;
1237                         goto out;
1238                 }
1239                 ret = convert_extent_item_v0(trans, root, path, owner,
1240                                              extra_size);
1241                 if (ret < 0) {
1242                         err = ret;
1243                         goto out;
1244                 }
1245                 leaf = path->nodes[0];
1246                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1247         }
1248 #endif
1249         BUG_ON(item_size < sizeof(*ei));
1250
1251         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1252         flags = btrfs_extent_flags(leaf, ei);
1253
1254         ptr = (unsigned long)(ei + 1);
1255         end = (unsigned long)ei + item_size;
1256
1257         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1258                 ptr += sizeof(struct btrfs_tree_block_info);
1259                 BUG_ON(ptr > end);
1260         } else {
1261                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1262         }
1263
1264         err = -ENOENT;
1265         while (1) {
1266                 if (ptr >= end) {
1267                         WARN_ON(ptr > end);
1268                         break;
1269                 }
1270                 iref = (struct btrfs_extent_inline_ref *)ptr;
1271                 type = btrfs_extent_inline_ref_type(leaf, iref);
1272                 if (want < type)
1273                         break;
1274                 if (want > type) {
1275                         ptr += btrfs_extent_inline_ref_size(type);
1276                         continue;
1277                 }
1278
1279                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1280                         struct btrfs_extent_data_ref *dref;
1281                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1282                         if (match_extent_data_ref(leaf, dref, root_objectid,
1283                                                   owner, offset)) {
1284                                 err = 0;
1285                                 break;
1286                         }
1287                         if (hash_extent_data_ref_item(leaf, dref) <
1288                             hash_extent_data_ref(root_objectid, owner, offset))
1289                                 break;
1290                 } else {
1291                         u64 ref_offset;
1292                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1293                         if (parent > 0) {
1294                                 if (parent == ref_offset) {
1295                                         err = 0;
1296                                         break;
1297                                 }
1298                                 if (ref_offset < parent)
1299                                         break;
1300                         } else {
1301                                 if (root_objectid == ref_offset) {
1302                                         err = 0;
1303                                         break;
1304                                 }
1305                                 if (ref_offset < root_objectid)
1306                                         break;
1307                         }
1308                 }
1309                 ptr += btrfs_extent_inline_ref_size(type);
1310         }
1311         if (err == -ENOENT && insert) {
1312                 if (item_size + extra_size >=
1313                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1314                         err = -EAGAIN;
1315                         goto out;
1316                 }
1317                 /*
1318                  * To add new inline back ref, we have to make sure
1319                  * there is no corresponding back ref item.
1320                  * For simplicity, we just do not add new inline back
1321                  * ref if there is any kind of item for this block
1322                  */
1323                 if (find_next_key(path, 0, &key) == 0 &&
1324                     key.objectid == bytenr &&
1325                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1326                         err = -EAGAIN;
1327                         goto out;
1328                 }
1329         }
1330         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1331 out:
1332         if (insert) {
1333                 path->keep_locks = 0;
1334                 btrfs_unlock_up_safe(path, 1);
1335         }
1336         return err;
1337 }
1338
1339 /*
1340  * helper to add new inline back ref
1341  */
1342 static noinline_for_stack
1343 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1344                                 struct btrfs_root *root,
1345                                 struct btrfs_path *path,
1346                                 struct btrfs_extent_inline_ref *iref,
1347                                 u64 parent, u64 root_objectid,
1348                                 u64 owner, u64 offset, int refs_to_add,
1349                                 struct btrfs_delayed_extent_op *extent_op)
1350 {
1351         struct extent_buffer *leaf;
1352         struct btrfs_extent_item *ei;
1353         unsigned long ptr;
1354         unsigned long end;
1355         unsigned long item_offset;
1356         u64 refs;
1357         int size;
1358         int type;
1359         int ret;
1360
1361         leaf = path->nodes[0];
1362         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1363         item_offset = (unsigned long)iref - (unsigned long)ei;
1364
1365         type = extent_ref_type(parent, owner);
1366         size = btrfs_extent_inline_ref_size(type);
1367
1368         ret = btrfs_extend_item(trans, root, path, size);
1369         BUG_ON(ret);
1370
1371         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1372         refs = btrfs_extent_refs(leaf, ei);
1373         refs += refs_to_add;
1374         btrfs_set_extent_refs(leaf, ei, refs);
1375         if (extent_op)
1376                 __run_delayed_extent_op(extent_op, leaf, ei);
1377
1378         ptr = (unsigned long)ei + item_offset;
1379         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1380         if (ptr < end - size)
1381                 memmove_extent_buffer(leaf, ptr + size, ptr,
1382                                       end - size - ptr);
1383
1384         iref = (struct btrfs_extent_inline_ref *)ptr;
1385         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1386         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1387                 struct btrfs_extent_data_ref *dref;
1388                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1389                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1390                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1391                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1392                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1393         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1394                 struct btrfs_shared_data_ref *sref;
1395                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1396                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1397                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1398         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1399                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1400         } else {
1401                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1402         }
1403         btrfs_mark_buffer_dirty(leaf);
1404         return 0;
1405 }
1406
1407 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1408                                  struct btrfs_root *root,
1409                                  struct btrfs_path *path,
1410                                  struct btrfs_extent_inline_ref **ref_ret,
1411                                  u64 bytenr, u64 num_bytes, u64 parent,
1412                                  u64 root_objectid, u64 owner, u64 offset)
1413 {
1414         int ret;
1415
1416         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1417                                            bytenr, num_bytes, parent,
1418                                            root_objectid, owner, offset, 0);
1419         if (ret != -ENOENT)
1420                 return ret;
1421
1422         btrfs_release_path(root, path);
1423         *ref_ret = NULL;
1424
1425         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1426                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1427                                             root_objectid);
1428         } else {
1429                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1430                                              root_objectid, owner, offset);
1431         }
1432         return ret;
1433 }
1434
1435 /*
1436  * helper to update/remove inline back ref
1437  */
1438 static noinline_for_stack
1439 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1440                                  struct btrfs_root *root,
1441                                  struct btrfs_path *path,
1442                                  struct btrfs_extent_inline_ref *iref,
1443                                  int refs_to_mod,
1444                                  struct btrfs_delayed_extent_op *extent_op)
1445 {
1446         struct extent_buffer *leaf;
1447         struct btrfs_extent_item *ei;
1448         struct btrfs_extent_data_ref *dref = NULL;
1449         struct btrfs_shared_data_ref *sref = NULL;
1450         unsigned long ptr;
1451         unsigned long end;
1452         u32 item_size;
1453         int size;
1454         int type;
1455         int ret;
1456         u64 refs;
1457
1458         leaf = path->nodes[0];
1459         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1460         refs = btrfs_extent_refs(leaf, ei);
1461         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1462         refs += refs_to_mod;
1463         btrfs_set_extent_refs(leaf, ei, refs);
1464         if (extent_op)
1465                 __run_delayed_extent_op(extent_op, leaf, ei);
1466
1467         type = btrfs_extent_inline_ref_type(leaf, iref);
1468
1469         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1470                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1471                 refs = btrfs_extent_data_ref_count(leaf, dref);
1472         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1473                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1474                 refs = btrfs_shared_data_ref_count(leaf, sref);
1475         } else {
1476                 refs = 1;
1477                 BUG_ON(refs_to_mod != -1);
1478         }
1479
1480         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1481         refs += refs_to_mod;
1482
1483         if (refs > 0) {
1484                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1485                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1486                 else
1487                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1488         } else {
1489                 size =  btrfs_extent_inline_ref_size(type);
1490                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1491                 ptr = (unsigned long)iref;
1492                 end = (unsigned long)ei + item_size;
1493                 if (ptr + size < end)
1494                         memmove_extent_buffer(leaf, ptr, ptr + size,
1495                                               end - ptr - size);
1496                 item_size -= size;
1497                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1498                 BUG_ON(ret);
1499         }
1500         btrfs_mark_buffer_dirty(leaf);
1501         return 0;
1502 }
1503
1504 static noinline_for_stack
1505 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1506                                  struct btrfs_root *root,
1507                                  struct btrfs_path *path,
1508                                  u64 bytenr, u64 num_bytes, u64 parent,
1509                                  u64 root_objectid, u64 owner,
1510                                  u64 offset, int refs_to_add,
1511                                  struct btrfs_delayed_extent_op *extent_op)
1512 {
1513         struct btrfs_extent_inline_ref *iref;
1514         int ret;
1515
1516         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1517                                            bytenr, num_bytes, parent,
1518                                            root_objectid, owner, offset, 1);
1519         if (ret == 0) {
1520                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1521                 ret = update_inline_extent_backref(trans, root, path, iref,
1522                                                    refs_to_add, extent_op);
1523         } else if (ret == -ENOENT) {
1524                 ret = setup_inline_extent_backref(trans, root, path, iref,
1525                                                   parent, root_objectid,
1526                                                   owner, offset, refs_to_add,
1527                                                   extent_op);
1528         }
1529         return ret;
1530 }
1531
1532 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1533                                  struct btrfs_root *root,
1534                                  struct btrfs_path *path,
1535                                  u64 bytenr, u64 parent, u64 root_objectid,
1536                                  u64 owner, u64 offset, int refs_to_add)
1537 {
1538         int ret;
1539         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1540                 BUG_ON(refs_to_add != 1);
1541                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1542                                             parent, root_objectid);
1543         } else {
1544                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1545                                              parent, root_objectid,
1546                                              owner, offset, refs_to_add);
1547         }
1548         return ret;
1549 }
1550
1551 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1552                                  struct btrfs_root *root,
1553                                  struct btrfs_path *path,
1554                                  struct btrfs_extent_inline_ref *iref,
1555                                  int refs_to_drop, int is_data)
1556 {
1557         int ret;
1558
1559         BUG_ON(!is_data && refs_to_drop != 1);
1560         if (iref) {
1561                 ret = update_inline_extent_backref(trans, root, path, iref,
1562                                                    -refs_to_drop, NULL);
1563         } else if (is_data) {
1564                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1565         } else {
1566                 ret = btrfs_del_item(trans, root, path);
1567         }
1568         return ret;
1569 }
1570
1571 #ifdef BIO_RW_DISCARD
1572 static void btrfs_issue_discard(struct block_device *bdev,
1573                                 u64 start, u64 len)
1574 {
1575         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
1576 }
1577 #endif
1578
1579 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1580                                 u64 num_bytes)
1581 {
1582 #ifdef BIO_RW_DISCARD
1583         int ret;
1584         u64 map_length = num_bytes;
1585         struct btrfs_multi_bio *multi = NULL;
1586
1587         if (!btrfs_test_opt(root, DISCARD))
1588                 return 0;
1589
1590         /* Tell the block device(s) that the sectors can be discarded */
1591         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1592                               bytenr, &map_length, &multi, 0);
1593         if (!ret) {
1594                 struct btrfs_bio_stripe *stripe = multi->stripes;
1595                 int i;
1596
1597                 if (map_length > num_bytes)
1598                         map_length = num_bytes;
1599
1600                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1601                         btrfs_issue_discard(stripe->dev->bdev,
1602                                             stripe->physical,
1603                                             map_length);
1604                 }
1605                 kfree(multi);
1606         }
1607
1608         return ret;
1609 #else
1610         return 0;
1611 #endif
1612 }
1613
1614 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1615                          struct btrfs_root *root,
1616                          u64 bytenr, u64 num_bytes, u64 parent,
1617                          u64 root_objectid, u64 owner, u64 offset)
1618 {
1619         int ret;
1620         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1621                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1622
1623         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1624                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1625                                         parent, root_objectid, (int)owner,
1626                                         BTRFS_ADD_DELAYED_REF, NULL);
1627         } else {
1628                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1629                                         parent, root_objectid, owner, offset,
1630                                         BTRFS_ADD_DELAYED_REF, NULL);
1631         }
1632         return ret;
1633 }
1634
1635 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1636                                   struct btrfs_root *root,
1637                                   u64 bytenr, u64 num_bytes,
1638                                   u64 parent, u64 root_objectid,
1639                                   u64 owner, u64 offset, int refs_to_add,
1640                                   struct btrfs_delayed_extent_op *extent_op)
1641 {
1642         struct btrfs_path *path;
1643         struct extent_buffer *leaf;
1644         struct btrfs_extent_item *item;
1645         u64 refs;
1646         int ret;
1647         int err = 0;
1648
1649         path = btrfs_alloc_path();
1650         if (!path)
1651                 return -ENOMEM;
1652
1653         path->reada = 1;
1654         path->leave_spinning = 1;
1655         /* this will setup the path even if it fails to insert the back ref */
1656         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1657                                            path, bytenr, num_bytes, parent,
1658                                            root_objectid, owner, offset,
1659                                            refs_to_add, extent_op);
1660         if (ret == 0)
1661                 goto out;
1662
1663         if (ret != -EAGAIN) {
1664                 err = ret;
1665                 goto out;
1666         }
1667
1668         leaf = path->nodes[0];
1669         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1670         refs = btrfs_extent_refs(leaf, item);
1671         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1672         if (extent_op)
1673                 __run_delayed_extent_op(extent_op, leaf, item);
1674
1675         btrfs_mark_buffer_dirty(leaf);
1676         btrfs_release_path(root->fs_info->extent_root, path);
1677
1678         path->reada = 1;
1679         path->leave_spinning = 1;
1680
1681         /* now insert the actual backref */
1682         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1683                                     path, bytenr, parent, root_objectid,
1684                                     owner, offset, refs_to_add);
1685         BUG_ON(ret);
1686 out:
1687         btrfs_free_path(path);
1688         return err;
1689 }
1690
1691 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1692                                 struct btrfs_root *root,
1693                                 struct btrfs_delayed_ref_node *node,
1694                                 struct btrfs_delayed_extent_op *extent_op,
1695                                 int insert_reserved)
1696 {
1697         int ret = 0;
1698         struct btrfs_delayed_data_ref *ref;
1699         struct btrfs_key ins;
1700         u64 parent = 0;
1701         u64 ref_root = 0;
1702         u64 flags = 0;
1703
1704         ins.objectid = node->bytenr;
1705         ins.offset = node->num_bytes;
1706         ins.type = BTRFS_EXTENT_ITEM_KEY;
1707
1708         ref = btrfs_delayed_node_to_data_ref(node);
1709         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1710                 parent = ref->parent;
1711         else
1712                 ref_root = ref->root;
1713
1714         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1715                 if (extent_op) {
1716                         BUG_ON(extent_op->update_key);
1717                         flags |= extent_op->flags_to_set;
1718                 }
1719                 ret = alloc_reserved_file_extent(trans, root,
1720                                                  parent, ref_root, flags,
1721                                                  ref->objectid, ref->offset,
1722                                                  &ins, node->ref_mod);
1723         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1724                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1725                                              node->num_bytes, parent,
1726                                              ref_root, ref->objectid,
1727                                              ref->offset, node->ref_mod,
1728                                              extent_op);
1729         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1730                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1731                                           node->num_bytes, parent,
1732                                           ref_root, ref->objectid,
1733                                           ref->offset, node->ref_mod,
1734                                           extent_op);
1735         } else {
1736                 BUG();
1737         }
1738         return ret;
1739 }
1740
1741 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1742                                     struct extent_buffer *leaf,
1743                                     struct btrfs_extent_item *ei)
1744 {
1745         u64 flags = btrfs_extent_flags(leaf, ei);
1746         if (extent_op->update_flags) {
1747                 flags |= extent_op->flags_to_set;
1748                 btrfs_set_extent_flags(leaf, ei, flags);
1749         }
1750
1751         if (extent_op->update_key) {
1752                 struct btrfs_tree_block_info *bi;
1753                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1754                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1755                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1756         }
1757 }
1758
1759 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1760                                  struct btrfs_root *root,
1761                                  struct btrfs_delayed_ref_node *node,
1762                                  struct btrfs_delayed_extent_op *extent_op)
1763 {
1764         struct btrfs_key key;
1765         struct btrfs_path *path;
1766         struct btrfs_extent_item *ei;
1767         struct extent_buffer *leaf;
1768         u32 item_size;
1769         int ret;
1770         int err = 0;
1771
1772         path = btrfs_alloc_path();
1773         if (!path)
1774                 return -ENOMEM;
1775
1776         key.objectid = node->bytenr;
1777         key.type = BTRFS_EXTENT_ITEM_KEY;
1778         key.offset = node->num_bytes;
1779
1780         path->reada = 1;
1781         path->leave_spinning = 1;
1782         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1783                                 path, 0, 1);
1784         if (ret < 0) {
1785                 err = ret;
1786                 goto out;
1787         }
1788         if (ret > 0) {
1789                 err = -EIO;
1790                 goto out;
1791         }
1792
1793         leaf = path->nodes[0];
1794         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1795 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1796         if (item_size < sizeof(*ei)) {
1797                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1798                                              path, (u64)-1, 0);
1799                 if (ret < 0) {
1800                         err = ret;
1801                         goto out;
1802                 }
1803                 leaf = path->nodes[0];
1804                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1805         }
1806 #endif
1807         BUG_ON(item_size < sizeof(*ei));
1808         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1809         __run_delayed_extent_op(extent_op, leaf, ei);
1810
1811         btrfs_mark_buffer_dirty(leaf);
1812 out:
1813         btrfs_free_path(path);
1814         return err;
1815 }
1816
1817 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1818                                 struct btrfs_root *root,
1819                                 struct btrfs_delayed_ref_node *node,
1820                                 struct btrfs_delayed_extent_op *extent_op,
1821                                 int insert_reserved)
1822 {
1823         int ret = 0;
1824         struct btrfs_delayed_tree_ref *ref;
1825         struct btrfs_key ins;
1826         u64 parent = 0;
1827         u64 ref_root = 0;
1828
1829         ins.objectid = node->bytenr;
1830         ins.offset = node->num_bytes;
1831         ins.type = BTRFS_EXTENT_ITEM_KEY;
1832
1833         ref = btrfs_delayed_node_to_tree_ref(node);
1834         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1835                 parent = ref->parent;
1836         else
1837                 ref_root = ref->root;
1838
1839         BUG_ON(node->ref_mod != 1);
1840         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1841                 BUG_ON(!extent_op || !extent_op->update_flags ||
1842                        !extent_op->update_key);
1843                 ret = alloc_reserved_tree_block(trans, root,
1844                                                 parent, ref_root,
1845                                                 extent_op->flags_to_set,
1846                                                 &extent_op->key,
1847                                                 ref->level, &ins);
1848         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1849                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1850                                              node->num_bytes, parent, ref_root,
1851                                              ref->level, 0, 1, extent_op);
1852         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1853                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1854                                           node->num_bytes, parent, ref_root,
1855                                           ref->level, 0, 1, extent_op);
1856         } else {
1857                 BUG();
1858         }
1859         return ret;
1860 }
1861
1862
1863 /* helper function to actually process a single delayed ref entry */
1864 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1865                                struct btrfs_root *root,
1866                                struct btrfs_delayed_ref_node *node,
1867                                struct btrfs_delayed_extent_op *extent_op,
1868                                int insert_reserved)
1869 {
1870         int ret;
1871         if (btrfs_delayed_ref_is_head(node)) {
1872                 struct btrfs_delayed_ref_head *head;
1873                 /*
1874                  * we've hit the end of the chain and we were supposed
1875                  * to insert this extent into the tree.  But, it got
1876                  * deleted before we ever needed to insert it, so all
1877                  * we have to do is clean up the accounting
1878                  */
1879                 BUG_ON(extent_op);
1880                 head = btrfs_delayed_node_to_head(node);
1881                 if (insert_reserved) {
1882                         int mark_free = 0;
1883                         struct extent_buffer *must_clean = NULL;
1884
1885                         ret = pin_down_bytes(trans, root, NULL,
1886                                              node->bytenr, node->num_bytes,
1887                                              head->is_data, 1, &must_clean);
1888                         if (ret > 0)
1889                                 mark_free = 1;
1890
1891                         if (must_clean) {
1892                                 clean_tree_block(NULL, root, must_clean);
1893                                 btrfs_tree_unlock(must_clean);
1894                                 free_extent_buffer(must_clean);
1895                         }
1896                         if (head->is_data) {
1897                                 ret = btrfs_del_csums(trans, root,
1898                                                       node->bytenr,
1899                                                       node->num_bytes);
1900                                 BUG_ON(ret);
1901                         }
1902                         if (mark_free) {
1903                                 ret = btrfs_free_reserved_extent(root,
1904                                                         node->bytenr,
1905                                                         node->num_bytes);
1906                                 BUG_ON(ret);
1907                         }
1908                 }
1909                 mutex_unlock(&head->mutex);
1910                 return 0;
1911         }
1912
1913         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1914             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1915                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1916                                            insert_reserved);
1917         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1918                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1919                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1920                                            insert_reserved);
1921         else
1922                 BUG();
1923         return ret;
1924 }
1925
1926 static noinline struct btrfs_delayed_ref_node *
1927 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1928 {
1929         struct rb_node *node;
1930         struct btrfs_delayed_ref_node *ref;
1931         int action = BTRFS_ADD_DELAYED_REF;
1932 again:
1933         /*
1934          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1935          * this prevents ref count from going down to zero when
1936          * there still are pending delayed ref.
1937          */
1938         node = rb_prev(&head->node.rb_node);
1939         while (1) {
1940                 if (!node)
1941                         break;
1942                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1943                                 rb_node);
1944                 if (ref->bytenr != head->node.bytenr)
1945                         break;
1946                 if (ref->action == action)
1947                         return ref;
1948                 node = rb_prev(node);
1949         }
1950         if (action == BTRFS_ADD_DELAYED_REF) {
1951                 action = BTRFS_DROP_DELAYED_REF;
1952                 goto again;
1953         }
1954         return NULL;
1955 }
1956
1957 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1958                                        struct btrfs_root *root,
1959                                        struct list_head *cluster)
1960 {
1961         struct btrfs_delayed_ref_root *delayed_refs;
1962         struct btrfs_delayed_ref_node *ref;
1963         struct btrfs_delayed_ref_head *locked_ref = NULL;
1964         struct btrfs_delayed_extent_op *extent_op;
1965         int ret;
1966         int count = 0;
1967         int must_insert_reserved = 0;
1968
1969         delayed_refs = &trans->transaction->delayed_refs;
1970         while (1) {
1971                 if (!locked_ref) {
1972                         /* pick a new head ref from the cluster list */
1973                         if (list_empty(cluster))
1974                                 break;
1975
1976                         locked_ref = list_entry(cluster->next,
1977                                      struct btrfs_delayed_ref_head, cluster);
1978
1979                         /* grab the lock that says we are going to process
1980                          * all the refs for this head */
1981                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
1982
1983                         /*
1984                          * we may have dropped the spin lock to get the head
1985                          * mutex lock, and that might have given someone else
1986                          * time to free the head.  If that's true, it has been
1987                          * removed from our list and we can move on.
1988                          */
1989                         if (ret == -EAGAIN) {
1990                                 locked_ref = NULL;
1991                                 count++;
1992                                 continue;
1993                         }
1994                 }
1995
1996                 /*
1997                  * record the must insert reserved flag before we
1998                  * drop the spin lock.
1999                  */
2000                 must_insert_reserved = locked_ref->must_insert_reserved;
2001                 locked_ref->must_insert_reserved = 0;
2002
2003                 extent_op = locked_ref->extent_op;
2004                 locked_ref->extent_op = NULL;
2005
2006                 /*
2007                  * locked_ref is the head node, so we have to go one
2008                  * node back for any delayed ref updates
2009                  */
2010                 ref = select_delayed_ref(locked_ref);
2011                 if (!ref) {
2012                         /* All delayed refs have been processed, Go ahead
2013                          * and send the head node to run_one_delayed_ref,
2014                          * so that any accounting fixes can happen
2015                          */
2016                         ref = &locked_ref->node;
2017
2018                         if (extent_op && must_insert_reserved) {
2019                                 kfree(extent_op);
2020                                 extent_op = NULL;
2021                         }
2022
2023                         if (extent_op) {
2024                                 spin_unlock(&delayed_refs->lock);
2025
2026                                 ret = run_delayed_extent_op(trans, root,
2027                                                             ref, extent_op);
2028                                 BUG_ON(ret);
2029                                 kfree(extent_op);
2030
2031                                 cond_resched();
2032                                 spin_lock(&delayed_refs->lock);
2033                                 continue;
2034                         }
2035
2036                         list_del_init(&locked_ref->cluster);
2037                         locked_ref = NULL;
2038                 }
2039
2040                 ref->in_tree = 0;
2041                 rb_erase(&ref->rb_node, &delayed_refs->root);
2042                 delayed_refs->num_entries--;
2043
2044                 spin_unlock(&delayed_refs->lock);
2045
2046                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2047                                           must_insert_reserved);
2048                 BUG_ON(ret);
2049
2050                 btrfs_put_delayed_ref(ref);
2051                 kfree(extent_op);
2052                 count++;
2053
2054                 cond_resched();
2055                 spin_lock(&delayed_refs->lock);
2056         }
2057         return count;
2058 }
2059
2060 /*
2061  * this starts processing the delayed reference count updates and
2062  * extent insertions we have queued up so far.  count can be
2063  * 0, which means to process everything in the tree at the start
2064  * of the run (but not newly added entries), or it can be some target
2065  * number you'd like to process.
2066  */
2067 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2068                            struct btrfs_root *root, unsigned long count)
2069 {
2070         struct rb_node *node;
2071         struct btrfs_delayed_ref_root *delayed_refs;
2072         struct btrfs_delayed_ref_node *ref;
2073         struct list_head cluster;
2074         int ret;
2075         int run_all = count == (unsigned long)-1;
2076         int run_most = 0;
2077
2078         if (root == root->fs_info->extent_root)
2079                 root = root->fs_info->tree_root;
2080
2081         delayed_refs = &trans->transaction->delayed_refs;
2082         INIT_LIST_HEAD(&cluster);
2083 again:
2084         spin_lock(&delayed_refs->lock);
2085         if (count == 0) {
2086                 count = delayed_refs->num_entries * 2;
2087                 run_most = 1;
2088         }
2089         while (1) {
2090                 if (!(run_all || run_most) &&
2091                     delayed_refs->num_heads_ready < 64)
2092                         break;
2093
2094                 /*
2095                  * go find something we can process in the rbtree.  We start at
2096                  * the beginning of the tree, and then build a cluster
2097                  * of refs to process starting at the first one we are able to
2098                  * lock
2099                  */
2100                 ret = btrfs_find_ref_cluster(trans, &cluster,
2101                                              delayed_refs->run_delayed_start);
2102                 if (ret)
2103                         break;
2104
2105                 ret = run_clustered_refs(trans, root, &cluster);
2106                 BUG_ON(ret < 0);
2107
2108                 count -= min_t(unsigned long, ret, count);
2109
2110                 if (count == 0)
2111                         break;
2112         }
2113
2114         if (run_all) {
2115                 node = rb_first(&delayed_refs->root);
2116                 if (!node)
2117                         goto out;
2118                 count = (unsigned long)-1;
2119
2120                 while (node) {
2121                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2122                                        rb_node);
2123                         if (btrfs_delayed_ref_is_head(ref)) {
2124                                 struct btrfs_delayed_ref_head *head;
2125
2126                                 head = btrfs_delayed_node_to_head(ref);
2127                                 atomic_inc(&ref->refs);
2128
2129                                 spin_unlock(&delayed_refs->lock);
2130                                 mutex_lock(&head->mutex);
2131                                 mutex_unlock(&head->mutex);
2132
2133                                 btrfs_put_delayed_ref(ref);
2134                                 cond_resched();
2135                                 goto again;
2136                         }
2137                         node = rb_next(node);
2138                 }
2139                 spin_unlock(&delayed_refs->lock);
2140                 schedule_timeout(1);
2141                 goto again;
2142         }
2143 out:
2144         spin_unlock(&delayed_refs->lock);
2145         return 0;
2146 }
2147
2148 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2149                                 struct btrfs_root *root,
2150                                 u64 bytenr, u64 num_bytes, u64 flags,
2151                                 int is_data)
2152 {
2153         struct btrfs_delayed_extent_op *extent_op;
2154         int ret;
2155
2156         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2157         if (!extent_op)
2158                 return -ENOMEM;
2159
2160         extent_op->flags_to_set = flags;
2161         extent_op->update_flags = 1;
2162         extent_op->update_key = 0;
2163         extent_op->is_data = is_data ? 1 : 0;
2164
2165         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2166         if (ret)
2167                 kfree(extent_op);
2168         return ret;
2169 }
2170
2171 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2172                                       struct btrfs_root *root,
2173                                       struct btrfs_path *path,
2174                                       u64 objectid, u64 offset, u64 bytenr)
2175 {
2176         struct btrfs_delayed_ref_head *head;
2177         struct btrfs_delayed_ref_node *ref;
2178         struct btrfs_delayed_data_ref *data_ref;
2179         struct btrfs_delayed_ref_root *delayed_refs;
2180         struct rb_node *node;
2181         int ret = 0;
2182
2183         ret = -ENOENT;
2184         delayed_refs = &trans->transaction->delayed_refs;
2185         spin_lock(&delayed_refs->lock);
2186         head = btrfs_find_delayed_ref_head(trans, bytenr);
2187         if (!head)
2188                 goto out;
2189
2190         if (!mutex_trylock(&head->mutex)) {
2191                 atomic_inc(&head->node.refs);
2192                 spin_unlock(&delayed_refs->lock);
2193
2194                 btrfs_release_path(root->fs_info->extent_root, path);
2195
2196                 mutex_lock(&head->mutex);
2197                 mutex_unlock(&head->mutex);
2198                 btrfs_put_delayed_ref(&head->node);
2199                 return -EAGAIN;
2200         }
2201
2202         node = rb_prev(&head->node.rb_node);
2203         if (!node)
2204                 goto out_unlock;
2205
2206         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2207
2208         if (ref->bytenr != bytenr)
2209                 goto out_unlock;
2210
2211         ret = 1;
2212         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2213                 goto out_unlock;
2214
2215         data_ref = btrfs_delayed_node_to_data_ref(ref);
2216
2217         node = rb_prev(node);
2218         if (node) {
2219                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2220                 if (ref->bytenr == bytenr)
2221                         goto out_unlock;
2222         }
2223
2224         if (data_ref->root != root->root_key.objectid ||
2225             data_ref->objectid != objectid || data_ref->offset != offset)
2226                 goto out_unlock;
2227
2228         ret = 0;
2229 out_unlock:
2230         mutex_unlock(&head->mutex);
2231 out:
2232         spin_unlock(&delayed_refs->lock);
2233         return ret;
2234 }
2235
2236 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2237                                         struct btrfs_root *root,
2238                                         struct btrfs_path *path,
2239                                         u64 objectid, u64 offset, u64 bytenr)
2240 {
2241         struct btrfs_root *extent_root = root->fs_info->extent_root;
2242         struct extent_buffer *leaf;
2243         struct btrfs_extent_data_ref *ref;
2244         struct btrfs_extent_inline_ref *iref;
2245         struct btrfs_extent_item *ei;
2246         struct btrfs_key key;
2247         u32 item_size;
2248         int ret;
2249
2250         key.objectid = bytenr;
2251         key.offset = (u64)-1;
2252         key.type = BTRFS_EXTENT_ITEM_KEY;
2253
2254         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2255         if (ret < 0)
2256                 goto out;
2257         BUG_ON(ret == 0);
2258
2259         ret = -ENOENT;
2260         if (path->slots[0] == 0)
2261                 goto out;
2262
2263         path->slots[0]--;
2264         leaf = path->nodes[0];
2265         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2266
2267         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2268                 goto out;
2269
2270         ret = 1;
2271         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2272 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2273         if (item_size < sizeof(*ei)) {
2274                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2275                 goto out;
2276         }
2277 #endif
2278         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2279
2280         if (item_size != sizeof(*ei) +
2281             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2282                 goto out;
2283
2284         if (btrfs_extent_generation(leaf, ei) <=
2285             btrfs_root_last_snapshot(&root->root_item))
2286                 goto out;
2287
2288         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2289         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2290             BTRFS_EXTENT_DATA_REF_KEY)
2291                 goto out;
2292
2293         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2294         if (btrfs_extent_refs(leaf, ei) !=
2295             btrfs_extent_data_ref_count(leaf, ref) ||
2296             btrfs_extent_data_ref_root(leaf, ref) !=
2297             root->root_key.objectid ||
2298             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2299             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2300                 goto out;
2301
2302         ret = 0;
2303 out:
2304         return ret;
2305 }
2306
2307 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2308                           struct btrfs_root *root,
2309                           u64 objectid, u64 offset, u64 bytenr)
2310 {
2311         struct btrfs_path *path;
2312         int ret;
2313         int ret2;
2314
2315         path = btrfs_alloc_path();
2316         if (!path)
2317                 return -ENOENT;
2318
2319         do {
2320                 ret = check_committed_ref(trans, root, path, objectid,
2321                                           offset, bytenr);
2322                 if (ret && ret != -ENOENT)
2323                         goto out;
2324
2325                 ret2 = check_delayed_ref(trans, root, path, objectid,
2326                                          offset, bytenr);
2327         } while (ret2 == -EAGAIN);
2328
2329         if (ret2 && ret2 != -ENOENT) {
2330                 ret = ret2;
2331                 goto out;
2332         }
2333
2334         if (ret != -ENOENT || ret2 != -ENOENT)
2335                 ret = 0;
2336 out:
2337         btrfs_free_path(path);
2338         return ret;
2339 }
2340
2341 #if 0
2342 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2343                     struct extent_buffer *buf, u32 nr_extents)
2344 {
2345         struct btrfs_key key;
2346         struct btrfs_file_extent_item *fi;
2347         u64 root_gen;
2348         u32 nritems;
2349         int i;
2350         int level;
2351         int ret = 0;
2352         int shared = 0;
2353
2354         if (!root->ref_cows)
2355                 return 0;
2356
2357         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2358                 shared = 0;
2359                 root_gen = root->root_key.offset;
2360         } else {
2361                 shared = 1;
2362                 root_gen = trans->transid - 1;
2363         }
2364
2365         level = btrfs_header_level(buf);
2366         nritems = btrfs_header_nritems(buf);
2367
2368         if (level == 0) {
2369                 struct btrfs_leaf_ref *ref;
2370                 struct btrfs_extent_info *info;
2371
2372                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2373                 if (!ref) {
2374                         ret = -ENOMEM;
2375                         goto out;
2376                 }
2377
2378                 ref->root_gen = root_gen;
2379                 ref->bytenr = buf->start;
2380                 ref->owner = btrfs_header_owner(buf);
2381                 ref->generation = btrfs_header_generation(buf);
2382                 ref->nritems = nr_extents;
2383                 info = ref->extents;
2384
2385                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2386                         u64 disk_bytenr;
2387                         btrfs_item_key_to_cpu(buf, &key, i);
2388                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2389                                 continue;
2390                         fi = btrfs_item_ptr(buf, i,
2391                                             struct btrfs_file_extent_item);
2392                         if (btrfs_file_extent_type(buf, fi) ==
2393                             BTRFS_FILE_EXTENT_INLINE)
2394                                 continue;
2395                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2396                         if (disk_bytenr == 0)
2397                                 continue;
2398
2399                         info->bytenr = disk_bytenr;
2400                         info->num_bytes =
2401                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2402                         info->objectid = key.objectid;
2403                         info->offset = key.offset;
2404                         info++;
2405                 }
2406
2407                 ret = btrfs_add_leaf_ref(root, ref, shared);
2408                 if (ret == -EEXIST && shared) {
2409                         struct btrfs_leaf_ref *old;
2410                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2411                         BUG_ON(!old);
2412                         btrfs_remove_leaf_ref(root, old);
2413                         btrfs_free_leaf_ref(root, old);
2414                         ret = btrfs_add_leaf_ref(root, ref, shared);
2415                 }
2416                 WARN_ON(ret);
2417                 btrfs_free_leaf_ref(root, ref);
2418         }
2419 out:
2420         return ret;
2421 }
2422
2423 /* when a block goes through cow, we update the reference counts of
2424  * everything that block points to.  The internal pointers of the block
2425  * can be in just about any order, and it is likely to have clusters of
2426  * things that are close together and clusters of things that are not.
2427  *
2428  * To help reduce the seeks that come with updating all of these reference
2429  * counts, sort them by byte number before actual updates are done.
2430  *
2431  * struct refsort is used to match byte number to slot in the btree block.
2432  * we sort based on the byte number and then use the slot to actually
2433  * find the item.
2434  *
2435  * struct refsort is smaller than strcut btrfs_item and smaller than
2436  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2437  * for a btree block, there's no way for a kmalloc of refsorts for a
2438  * single node to be bigger than a page.
2439  */
2440 struct refsort {
2441         u64 bytenr;
2442         u32 slot;
2443 };
2444
2445 /*
2446  * for passing into sort()
2447  */
2448 static int refsort_cmp(const void *a_void, const void *b_void)
2449 {
2450         const struct refsort *a = a_void;
2451         const struct refsort *b = b_void;
2452
2453         if (a->bytenr < b->bytenr)
2454                 return -1;
2455         if (a->bytenr > b->bytenr)
2456                 return 1;
2457         return 0;
2458 }
2459 #endif
2460
2461 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2462                            struct btrfs_root *root,
2463                            struct extent_buffer *buf,
2464                            int full_backref, int inc)
2465 {
2466         u64 bytenr;
2467         u64 num_bytes;
2468         u64 parent;
2469         u64 ref_root;
2470         u32 nritems;
2471         struct btrfs_key key;
2472         struct btrfs_file_extent_item *fi;
2473         int i;
2474         int level;
2475         int ret = 0;
2476         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2477                             u64, u64, u64, u64, u64, u64);
2478
2479         ref_root = btrfs_header_owner(buf);
2480         nritems = btrfs_header_nritems(buf);
2481         level = btrfs_header_level(buf);
2482
2483         if (!root->ref_cows && level == 0)
2484                 return 0;
2485
2486         if (inc)
2487                 process_func = btrfs_inc_extent_ref;
2488         else
2489                 process_func = btrfs_free_extent;
2490
2491         if (full_backref)
2492                 parent = buf->start;
2493         else
2494                 parent = 0;
2495
2496         for (i = 0; i < nritems; i++) {
2497                 if (level == 0) {
2498                         btrfs_item_key_to_cpu(buf, &key, i);
2499                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2500                                 continue;
2501                         fi = btrfs_item_ptr(buf, i,
2502                                             struct btrfs_file_extent_item);
2503                         if (btrfs_file_extent_type(buf, fi) ==
2504                             BTRFS_FILE_EXTENT_INLINE)
2505                                 continue;
2506                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2507                         if (bytenr == 0)
2508                                 continue;
2509
2510                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2511                         key.offset -= btrfs_file_extent_offset(buf, fi);
2512                         ret = process_func(trans, root, bytenr, num_bytes,
2513                                            parent, ref_root, key.objectid,
2514                                            key.offset);
2515                         if (ret)
2516                                 goto fail;
2517                 } else {
2518                         bytenr = btrfs_node_blockptr(buf, i);
2519                         num_bytes = btrfs_level_size(root, level - 1);
2520                         ret = process_func(trans, root, bytenr, num_bytes,
2521                                            parent, ref_root, level - 1, 0);
2522                         if (ret)
2523                                 goto fail;
2524                 }
2525         }
2526         return 0;
2527 fail:
2528         BUG();
2529         return ret;
2530 }
2531
2532 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2533                   struct extent_buffer *buf, int full_backref)
2534 {
2535         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2536 }
2537
2538 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2539                   struct extent_buffer *buf, int full_backref)
2540 {
2541         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2542 }
2543
2544 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2545                                  struct btrfs_root *root,
2546                                  struct btrfs_path *path,
2547                                  struct btrfs_block_group_cache *cache)
2548 {
2549         int ret;
2550         struct btrfs_root *extent_root = root->fs_info->extent_root;
2551         unsigned long bi;
2552         struct extent_buffer *leaf;
2553
2554         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2555         if (ret < 0)
2556                 goto fail;
2557         BUG_ON(ret);
2558
2559         leaf = path->nodes[0];
2560         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2561         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2562         btrfs_mark_buffer_dirty(leaf);
2563         btrfs_release_path(extent_root, path);
2564 fail:
2565         if (ret)
2566                 return ret;
2567         return 0;
2568
2569 }
2570
2571 static struct btrfs_block_group_cache *
2572 next_block_group(struct btrfs_root *root,
2573                  struct btrfs_block_group_cache *cache)
2574 {
2575         struct rb_node *node;
2576         spin_lock(&root->fs_info->block_group_cache_lock);
2577         node = rb_next(&cache->cache_node);
2578         btrfs_put_block_group(cache);
2579         if (node) {
2580                 cache = rb_entry(node, struct btrfs_block_group_cache,
2581                                  cache_node);
2582                 atomic_inc(&cache->count);
2583         } else
2584                 cache = NULL;
2585         spin_unlock(&root->fs_info->block_group_cache_lock);
2586         return cache;
2587 }
2588
2589 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2590                                    struct btrfs_root *root)
2591 {
2592         struct btrfs_block_group_cache *cache;
2593         int err = 0;
2594         struct btrfs_path *path;
2595         u64 last = 0;
2596
2597         path = btrfs_alloc_path();
2598         if (!path)
2599                 return -ENOMEM;
2600
2601         while (1) {
2602                 if (last == 0) {
2603                         err = btrfs_run_delayed_refs(trans, root,
2604                                                      (unsigned long)-1);
2605                         BUG_ON(err);
2606                 }
2607
2608                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2609                 while (cache) {
2610                         if (cache->dirty)
2611                                 break;
2612                         cache = next_block_group(root, cache);
2613                 }
2614                 if (!cache) {
2615                         if (last == 0)
2616                                 break;
2617                         last = 0;
2618                         continue;
2619                 }
2620
2621                 cache->dirty = 0;
2622                 last = cache->key.objectid + cache->key.offset;
2623
2624                 err = write_one_cache_group(trans, root, path, cache);
2625                 BUG_ON(err);
2626                 btrfs_put_block_group(cache);
2627         }
2628
2629         btrfs_free_path(path);
2630         return 0;
2631 }
2632
2633 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2634 {
2635         struct btrfs_block_group_cache *block_group;
2636         int readonly = 0;
2637
2638         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2639         if (!block_group || block_group->ro)
2640                 readonly = 1;
2641         if (block_group)
2642                 btrfs_put_block_group(block_group);
2643         return readonly;
2644 }
2645
2646 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2647                              u64 total_bytes, u64 bytes_used,
2648                              struct btrfs_space_info **space_info)
2649 {
2650         struct btrfs_space_info *found;
2651
2652         found = __find_space_info(info, flags);
2653         if (found) {
2654                 spin_lock(&found->lock);
2655                 found->total_bytes += total_bytes;
2656                 found->bytes_used += bytes_used;
2657                 found->full = 0;
2658                 spin_unlock(&found->lock);
2659                 *space_info = found;
2660                 return 0;
2661         }
2662         found = kzalloc(sizeof(*found), GFP_NOFS);
2663         if (!found)
2664                 return -ENOMEM;
2665
2666         INIT_LIST_HEAD(&found->block_groups);
2667         init_rwsem(&found->groups_sem);
2668         spin_lock_init(&found->lock);
2669         found->flags = flags;
2670         found->total_bytes = total_bytes;
2671         found->bytes_used = bytes_used;
2672         found->bytes_pinned = 0;
2673         found->bytes_reserved = 0;
2674         found->bytes_readonly = 0;
2675         found->bytes_delalloc = 0;
2676         found->full = 0;
2677         found->force_alloc = 0;
2678         *space_info = found;
2679         list_add_rcu(&found->list, &info->space_info);
2680         atomic_set(&found->caching_threads, 0);
2681         return 0;
2682 }
2683
2684 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2685 {
2686         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2687                                    BTRFS_BLOCK_GROUP_RAID1 |
2688                                    BTRFS_BLOCK_GROUP_RAID10 |
2689                                    BTRFS_BLOCK_GROUP_DUP);
2690         if (extra_flags) {
2691                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2692                         fs_info->avail_data_alloc_bits |= extra_flags;
2693                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2694                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2695                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2696                         fs_info->avail_system_alloc_bits |= extra_flags;
2697         }
2698 }
2699
2700 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2701 {
2702         spin_lock(&cache->space_info->lock);
2703         spin_lock(&cache->lock);
2704         if (!cache->ro) {
2705                 cache->space_info->bytes_readonly += cache->key.offset -
2706                                         btrfs_block_group_used(&cache->item);
2707                 cache->ro = 1;
2708         }
2709         spin_unlock(&cache->lock);
2710         spin_unlock(&cache->space_info->lock);
2711 }
2712
2713 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2714 {
2715         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2716
2717         if (num_devices == 1)
2718                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2719         if (num_devices < 4)
2720                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2721
2722         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2723             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2724                       BTRFS_BLOCK_GROUP_RAID10))) {
2725                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2726         }
2727
2728         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2729             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2730                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2731         }
2732
2733         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2734             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2735              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2736              (flags & BTRFS_BLOCK_GROUP_DUP)))
2737                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2738         return flags;
2739 }
2740
2741 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2742 {
2743         struct btrfs_fs_info *info = root->fs_info;
2744         u64 alloc_profile;
2745
2746         if (data) {
2747                 alloc_profile = info->avail_data_alloc_bits &
2748                         info->data_alloc_profile;
2749                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2750         } else if (root == root->fs_info->chunk_root) {
2751                 alloc_profile = info->avail_system_alloc_bits &
2752                         info->system_alloc_profile;
2753                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2754         } else {
2755                 alloc_profile = info->avail_metadata_alloc_bits &
2756                         info->metadata_alloc_profile;
2757                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2758         }
2759
2760         return btrfs_reduce_alloc_profile(root, data);
2761 }
2762
2763 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2764 {
2765         u64 alloc_target;
2766
2767         alloc_target = btrfs_get_alloc_profile(root, 1);
2768         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2769                                                        alloc_target);
2770 }
2771
2772 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2773 {
2774         u64 num_bytes;
2775         int level;
2776
2777         level = BTRFS_MAX_LEVEL - 2;
2778         /*
2779          * NOTE: these calculations are absolutely the worst possible case.
2780          * This assumes that _every_ item we insert will require a new leaf, and
2781          * that the tree has grown to its maximum level size.
2782          */
2783
2784         /*
2785          * for every item we insert we could insert both an extent item and a
2786          * extent ref item.  Then for ever item we insert, we will need to cow
2787          * both the original leaf, plus the leaf to the left and right of it.
2788          *
2789          * Unless we are talking about the extent root, then we just want the
2790          * number of items * 2, since we just need the extent item plus its ref.
2791          */
2792         if (root == root->fs_info->extent_root)
2793                 num_bytes = num_items * 2;
2794         else
2795                 num_bytes = (num_items + (2 * num_items)) * 3;
2796
2797         /*
2798          * num_bytes is total number of leaves we could need times the leaf
2799          * size, and then for every leaf we could end up cow'ing 2 nodes per
2800          * level, down to the leaf level.
2801          */
2802         num_bytes = (num_bytes * root->leafsize) +
2803                 (num_bytes * (level * 2)) * root->nodesize;
2804
2805         return num_bytes;
2806 }
2807
2808 /*
2809  * Unreserve metadata space for delalloc.  If we have less reserved credits than
2810  * we have extents, this function does nothing.
2811  */
2812 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2813                                           struct inode *inode, int num_items)
2814 {
2815         struct btrfs_fs_info *info = root->fs_info;
2816         struct btrfs_space_info *meta_sinfo;
2817         u64 num_bytes;
2818         u64 alloc_target;
2819         bool bug = false;
2820
2821         /* get the space info for where the metadata will live */
2822         alloc_target = btrfs_get_alloc_profile(root, 0);
2823         meta_sinfo = __find_space_info(info, alloc_target);
2824
2825         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2826                                            num_items);
2827
2828         spin_lock(&meta_sinfo->lock);
2829         spin_lock(&BTRFS_I(inode)->accounting_lock);
2830         if (BTRFS_I(inode)->reserved_extents <=
2831             BTRFS_I(inode)->outstanding_extents) {
2832                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2833                 spin_unlock(&meta_sinfo->lock);
2834                 return 0;
2835         }
2836         spin_unlock(&BTRFS_I(inode)->accounting_lock);
2837
2838         BTRFS_I(inode)->reserved_extents--;
2839         BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2840
2841         if (meta_sinfo->bytes_delalloc < num_bytes) {
2842                 bug = true;
2843                 meta_sinfo->bytes_delalloc = 0;
2844         } else {
2845                 meta_sinfo->bytes_delalloc -= num_bytes;
2846         }
2847         spin_unlock(&meta_sinfo->lock);
2848
2849         BUG_ON(bug);
2850
2851         return 0;
2852 }
2853
2854 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2855 {
2856         u64 thresh;
2857
2858         thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2859                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2860                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2861                 meta_sinfo->bytes_may_use;
2862
2863         thresh = meta_sinfo->total_bytes - thresh;
2864         thresh *= 80;
2865         do_div(thresh, 100);
2866         if (thresh <= meta_sinfo->bytes_delalloc)
2867                 meta_sinfo->force_delalloc = 1;
2868         else
2869                 meta_sinfo->force_delalloc = 0;
2870 }
2871
2872 struct async_flush {
2873         struct btrfs_root *root;
2874         struct btrfs_space_info *info;
2875         struct btrfs_work work;
2876 };
2877
2878 static noinline void flush_delalloc_async(struct btrfs_work *work)
2879 {
2880         struct async_flush *async;
2881         struct btrfs_root *root;
2882         struct btrfs_space_info *info;
2883
2884         async = container_of(work, struct async_flush, work);
2885         root = async->root;
2886         info = async->info;
2887
2888         btrfs_start_delalloc_inodes(root);
2889         wake_up(&info->flush_wait);
2890         btrfs_wait_ordered_extents(root, 0);
2891
2892         spin_lock(&info->lock);
2893         info->flushing = 0;
2894         spin_unlock(&info->lock);
2895         wake_up(&info->flush_wait);
2896
2897         kfree(async);
2898 }
2899
2900 static void wait_on_flush(struct btrfs_space_info *info)
2901 {
2902         DEFINE_WAIT(wait);
2903         u64 used;
2904
2905         while (1) {
2906                 prepare_to_wait(&info->flush_wait, &wait,
2907                                 TASK_UNINTERRUPTIBLE);
2908                 spin_lock(&info->lock);
2909                 if (!info->flushing) {
2910                         spin_unlock(&info->lock);
2911                         break;
2912                 }
2913
2914                 used = info->bytes_used + info->bytes_reserved +
2915                         info->bytes_pinned + info->bytes_readonly +
2916                         info->bytes_super + info->bytes_root +
2917                         info->bytes_may_use + info->bytes_delalloc;
2918                 if (used < info->total_bytes) {
2919                         spin_unlock(&info->lock);
2920                         break;
2921                 }
2922                 spin_unlock(&info->lock);
2923                 schedule();
2924         }
2925         finish_wait(&info->flush_wait, &wait);
2926 }
2927
2928 static void flush_delalloc(struct btrfs_root *root,
2929                                  struct btrfs_space_info *info)
2930 {
2931         struct async_flush *async;
2932         bool wait = false;
2933
2934         spin_lock(&info->lock);
2935
2936         if (!info->flushing) {
2937                 info->flushing = 1;
2938                 init_waitqueue_head(&info->flush_wait);
2939         } else {
2940                 wait = true;
2941         }
2942
2943         spin_unlock(&info->lock);
2944
2945         if (wait) {
2946                 wait_on_flush(info);
2947                 return;
2948         }
2949
2950         async = kzalloc(sizeof(*async), GFP_NOFS);
2951         if (!async)
2952                 goto flush;
2953
2954         async->root = root;
2955         async->info = info;
2956         async->work.func = flush_delalloc_async;
2957
2958         btrfs_queue_worker(&root->fs_info->enospc_workers,
2959                            &async->work);
2960         wait_on_flush(info);
2961         return;
2962
2963 flush:
2964         btrfs_start_delalloc_inodes(root);
2965         btrfs_wait_ordered_extents(root, 0);
2966
2967         spin_lock(&info->lock);
2968         info->flushing = 0;
2969         spin_unlock(&info->lock);
2970         wake_up(&info->flush_wait);
2971 }
2972
2973 static int maybe_allocate_chunk(struct btrfs_root *root,
2974                                  struct btrfs_space_info *info)
2975 {
2976         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
2977         struct btrfs_trans_handle *trans;
2978         bool wait = false;
2979         int ret = 0;
2980         u64 min_metadata;
2981         u64 free_space;
2982
2983         free_space = btrfs_super_total_bytes(disk_super);
2984         /*
2985          * we allow the metadata to grow to a max of either 5gb or 5% of the
2986          * space in the volume.
2987          */
2988         min_metadata = min((u64)5 * 1024 * 1024 * 1024,
2989                              div64_u64(free_space * 5, 100));
2990         if (info->total_bytes >= min_metadata) {
2991                 spin_unlock(&info->lock);
2992                 return 0;
2993         }
2994
2995         if (info->full) {
2996                 spin_unlock(&info->lock);
2997                 return 0;
2998         }
2999
3000         if (!info->allocating_chunk) {
3001                 info->force_alloc = 1;
3002                 info->allocating_chunk = 1;
3003                 init_waitqueue_head(&info->allocate_wait);
3004         } else {
3005                 wait = true;
3006         }
3007
3008         spin_unlock(&info->lock);
3009
3010         if (wait) {
3011                 wait_event(info->allocate_wait,
3012                            !info->allocating_chunk);
3013                 return 1;
3014         }
3015
3016         trans = btrfs_start_transaction(root, 1);
3017         if (!trans) {
3018                 ret = -ENOMEM;
3019                 goto out;
3020         }
3021
3022         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3023                              4096 + 2 * 1024 * 1024,
3024                              info->flags, 0);
3025         btrfs_end_transaction(trans, root);
3026         if (ret)
3027                 goto out;
3028 out:
3029         spin_lock(&info->lock);
3030         info->allocating_chunk = 0;
3031         spin_unlock(&info->lock);
3032         wake_up(&info->allocate_wait);
3033
3034         if (ret)
3035                 return 0;
3036         return 1;
3037 }
3038
3039 /*
3040  * Reserve metadata space for delalloc.
3041  */
3042 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
3043                                         struct inode *inode, int num_items)
3044 {
3045         struct btrfs_fs_info *info = root->fs_info;
3046         struct btrfs_space_info *meta_sinfo;
3047         u64 num_bytes;
3048         u64 used;
3049         u64 alloc_target;
3050         int flushed = 0;
3051         int force_delalloc;
3052
3053         /* get the space info for where the metadata will live */
3054         alloc_target = btrfs_get_alloc_profile(root, 0);
3055         meta_sinfo = __find_space_info(info, alloc_target);
3056
3057         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
3058                                            num_items);
3059 again:
3060         spin_lock(&meta_sinfo->lock);
3061
3062         force_delalloc = meta_sinfo->force_delalloc;
3063
3064         if (unlikely(!meta_sinfo->bytes_root))
3065                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3066
3067         if (!flushed)
3068                 meta_sinfo->bytes_delalloc += num_bytes;
3069
3070         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3071                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3072                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3073                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3074
3075         if (used > meta_sinfo->total_bytes) {
3076                 flushed++;
3077
3078                 if (flushed == 1) {
3079                         if (maybe_allocate_chunk(root, meta_sinfo))
3080                                 goto again;
3081                         flushed++;
3082                 } else {
3083                         spin_unlock(&meta_sinfo->lock);
3084                 }
3085
3086                 if (flushed == 2) {
3087                         filemap_flush(inode->i_mapping);
3088                         goto again;
3089                 } else if (flushed == 3) {
3090                         flush_delalloc(root, meta_sinfo);
3091                         goto again;
3092                 }
3093                 spin_lock(&meta_sinfo->lock);
3094                 meta_sinfo->bytes_delalloc -= num_bytes;
3095                 spin_unlock(&meta_sinfo->lock);
3096                 printk(KERN_ERR "enospc, has %d, reserved %d\n",
3097                        BTRFS_I(inode)->outstanding_extents,
3098                        BTRFS_I(inode)->reserved_extents);
3099                 dump_space_info(meta_sinfo, 0, 0);
3100                 return -ENOSPC;
3101         }
3102
3103         BTRFS_I(inode)->reserved_extents++;
3104         check_force_delalloc(meta_sinfo);
3105         spin_unlock(&meta_sinfo->lock);
3106
3107         if (!flushed && force_delalloc)
3108                 filemap_flush(inode->i_mapping);
3109
3110         return 0;
3111 }
3112
3113 /*
3114  * unreserve num_items number of items worth of metadata space.  This needs to
3115  * be paired with btrfs_reserve_metadata_space.
3116  *
3117  * NOTE: if you have the option, run this _AFTER_ you do a
3118  * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3119  * oprations which will result in more used metadata, so we want to make sure we
3120  * can do that without issue.
3121  */
3122 int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
3123 {
3124         struct btrfs_fs_info *info = root->fs_info;
3125         struct btrfs_space_info *meta_sinfo;
3126         u64 num_bytes;
3127         u64 alloc_target;
3128         bool bug = false;
3129
3130         /* get the space info for where the metadata will live */
3131         alloc_target = btrfs_get_alloc_profile(root, 0);
3132         meta_sinfo = __find_space_info(info, alloc_target);
3133
3134         num_bytes = calculate_bytes_needed(root, num_items);
3135
3136         spin_lock(&meta_sinfo->lock);
3137         if (meta_sinfo->bytes_may_use < num_bytes) {
3138                 bug = true;
3139                 meta_sinfo->bytes_may_use = 0;
3140         } else {
3141                 meta_sinfo->bytes_may_use -= num_bytes;
3142         }
3143         spin_unlock(&meta_sinfo->lock);
3144
3145         BUG_ON(bug);
3146
3147         return 0;
3148 }
3149
3150 /*
3151  * Reserve some metadata space for use.  We'll calculate the worste case number
3152  * of bytes that would be needed to modify num_items number of items.  If we
3153  * have space, fantastic, if not, you get -ENOSPC.  Please call
3154  * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3155  * items you reserved, since whatever metadata you needed should have already
3156  * been allocated.
3157  *
3158  * This will commit the transaction to make more space if we don't have enough
3159  * metadata space.  THe only time we don't do this is if we're reserving space
3160  * inside of a transaction, then we will just return -ENOSPC and it is the
3161  * callers responsibility to handle it properly.
3162  */
3163 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3164 {
3165         struct btrfs_fs_info *info = root->fs_info;
3166         struct btrfs_space_info *meta_sinfo;
3167         u64 num_bytes;
3168         u64 used;
3169         u64 alloc_target;
3170         int retries = 0;
3171
3172         /* get the space info for where the metadata will live */
3173         alloc_target = btrfs_get_alloc_profile(root, 0);
3174         meta_sinfo = __find_space_info(info, alloc_target);
3175
3176         num_bytes = calculate_bytes_needed(root, num_items);
3177 again:
3178         spin_lock(&meta_sinfo->lock);
3179
3180         if (unlikely(!meta_sinfo->bytes_root))
3181                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3182
3183         if (!retries)
3184                 meta_sinfo->bytes_may_use += num_bytes;
3185
3186         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3187                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3188                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3189                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3190
3191         if (used > meta_sinfo->total_bytes) {
3192                 retries++;
3193                 if (retries == 1) {
3194                         if (maybe_allocate_chunk(root, meta_sinfo))
3195                                 goto again;
3196                         retries++;
3197                 } else {
3198                         spin_unlock(&meta_sinfo->lock);
3199                 }
3200
3201                 if (retries == 2) {
3202                         flush_delalloc(root, meta_sinfo);
3203                         goto again;
3204                 }
3205                 spin_lock(&meta_sinfo->lock);
3206                 meta_sinfo->bytes_may_use -= num_bytes;
3207                 spin_unlock(&meta_sinfo->lock);
3208
3209                 dump_space_info(meta_sinfo, 0, 0);
3210                 return -ENOSPC;
3211         }
3212
3213         check_force_delalloc(meta_sinfo);
3214         spin_unlock(&meta_sinfo->lock);
3215
3216         return 0;
3217 }
3218
3219 /*
3220  * This will check the space that the inode allocates from to make sure we have
3221  * enough space for bytes.
3222  */
3223 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3224                                 u64 bytes)
3225 {
3226         struct btrfs_space_info *data_sinfo;
3227         int ret = 0, committed = 0;
3228
3229         /* make sure bytes are sectorsize aligned */
3230         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3231
3232         data_sinfo = BTRFS_I(inode)->space_info;
3233         if (!data_sinfo)
3234                 goto alloc;
3235
3236 again:
3237         /* make sure we have enough space to handle the data first */
3238         spin_lock(&data_sinfo->lock);
3239         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
3240             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
3241             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
3242             data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
3243                 struct btrfs_trans_handle *trans;
3244
3245                 /*
3246                  * if we don't have enough free bytes in this space then we need
3247                  * to alloc a new chunk.
3248                  */
3249                 if (!data_sinfo->full) {
3250                         u64 alloc_target;
3251
3252                         data_sinfo->force_alloc = 1;
3253                         spin_unlock(&data_sinfo->lock);
3254 alloc:
3255                         alloc_target = btrfs_get_alloc_profile(root, 1);
3256                         trans = btrfs_start_transaction(root, 1);
3257                         if (!trans)
3258                                 return -ENOMEM;
3259
3260                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3261                                              bytes + 2 * 1024 * 1024,
3262                                              alloc_target, 0);
3263                         btrfs_end_transaction(trans, root);
3264                         if (ret)
3265                                 return ret;
3266
3267                         if (!data_sinfo) {
3268                                 btrfs_set_inode_space_info(root, inode);
3269                                 data_sinfo = BTRFS_I(inode)->space_info;
3270                         }
3271                         goto again;
3272                 }
3273                 spin_unlock(&data_sinfo->lock);
3274
3275                 /* commit the current transaction and try again */
3276                 if (!committed && !root->fs_info->open_ioctl_trans) {
3277                         committed = 1;
3278                         trans = btrfs_join_transaction(root, 1);
3279                         if (!trans)
3280                                 return -ENOMEM;
3281                         ret = btrfs_commit_transaction(trans, root);
3282                         if (ret)
3283                                 return ret;
3284                         goto again;
3285                 }
3286
3287                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3288                        ", %llu bytes_used, %llu bytes_reserved, "
3289                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3290                        "%llu total\n", (unsigned long long)bytes,
3291                        (unsigned long long)data_sinfo->bytes_delalloc,
3292                        (unsigned long long)data_sinfo->bytes_used,
3293                        (unsigned long long)data_sinfo->bytes_reserved,
3294                        (unsigned long long)data_sinfo->bytes_pinned,
3295                        (unsigned long long)data_sinfo->bytes_readonly,
3296                        (unsigned long long)data_sinfo->bytes_may_use,
3297                        (unsigned long long)data_sinfo->total_bytes);
3298                 return -ENOSPC;
3299         }
3300         data_sinfo->bytes_may_use += bytes;
3301         BTRFS_I(inode)->reserved_bytes += bytes;
3302         spin_unlock(&data_sinfo->lock);
3303
3304         return 0;
3305 }
3306
3307 /*
3308  * if there was an error for whatever reason after calling
3309  * btrfs_check_data_free_space, call this so we can cleanup the counters.
3310  */
3311 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3312                                     struct inode *inode, u64 bytes)
3313 {
3314         struct btrfs_space_info *data_sinfo;
3315
3316         /* make sure bytes are sectorsize aligned */
3317         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3318
3319         data_sinfo = BTRFS_I(inode)->space_info;
3320         spin_lock(&data_sinfo->lock);
3321         data_sinfo->bytes_may_use -= bytes;
3322         BTRFS_I(inode)->reserved_bytes -= bytes;
3323         spin_unlock(&data_sinfo->lock);
3324 }
3325
3326 /* called when we are adding a delalloc extent to the inode's io_tree */
3327 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3328                                   u64 bytes)
3329 {
3330         struct btrfs_space_info *data_sinfo;
3331
3332         /* get the space info for where this inode will be storing its data */
3333         data_sinfo = BTRFS_I(inode)->space_info;
3334
3335         /* make sure we have enough space to handle the data first */
3336         spin_lock(&data_sinfo->lock);
3337         data_sinfo->bytes_delalloc += bytes;
3338
3339         /*
3340          * we are adding a delalloc extent without calling
3341          * btrfs_check_data_free_space first.  This happens on a weird
3342          * writepage condition, but shouldn't hurt our accounting
3343          */
3344         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3345                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3346                 BTRFS_I(inode)->reserved_bytes = 0;
3347         } else {
3348                 data_sinfo->bytes_may_use -= bytes;
3349                 BTRFS_I(inode)->reserved_bytes -= bytes;
3350         }
3351
3352         spin_unlock(&data_sinfo->lock);
3353 }
3354
3355 /* called when we are clearing an delalloc extent from the inode's io_tree */
3356 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3357                               u64 bytes)
3358 {
3359         struct btrfs_space_info *info;
3360
3361         info = BTRFS_I(inode)->space_info;
3362
3363         spin_lock(&info->lock);
3364         info->bytes_delalloc -= bytes;
3365         spin_unlock(&info->lock);
3366 }
3367
3368 static void force_metadata_allocation(struct btrfs_fs_info *info)
3369 {
3370         struct list_head *head = &info->space_info;
3371         struct btrfs_space_info *found;
3372
3373         rcu_read_lock();
3374         list_for_each_entry_rcu(found, head, list) {
3375                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3376                         found->force_alloc = 1;
3377         }
3378         rcu_read_unlock();
3379 }
3380
3381 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3382                           struct btrfs_root *extent_root, u64 alloc_bytes,
3383                           u64 flags, int force)
3384 {
3385         struct btrfs_space_info *space_info;
3386         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3387         u64 thresh;
3388         int ret = 0;
3389
3390         mutex_lock(&fs_info->chunk_mutex);
3391
3392         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3393
3394         space_info = __find_space_info(extent_root->fs_info, flags);
3395         if (!space_info) {
3396                 ret = update_space_info(extent_root->fs_info, flags,
3397                                         0, 0, &space_info);
3398                 BUG_ON(ret);
3399         }
3400         BUG_ON(!space_info);
3401
3402         spin_lock(&space_info->lock);
3403         if (space_info->force_alloc)
3404                 force = 1;
3405         if (space_info->full) {
3406                 spin_unlock(&space_info->lock);
3407                 goto out;
3408         }
3409
3410         thresh = space_info->total_bytes - space_info->bytes_readonly;
3411         thresh = div_factor(thresh, 8);
3412         if (!force &&
3413            (space_info->bytes_used + space_info->bytes_pinned +
3414             space_info->bytes_reserved + alloc_bytes) < thresh) {
3415                 spin_unlock(&space_info->lock);
3416                 goto out;
3417         }
3418         spin_unlock(&space_info->lock);
3419
3420         /*
3421          * if we're doing a data chunk, go ahead and make sure that
3422          * we keep a reasonable number of metadata chunks allocated in the
3423          * FS as well.
3424          */
3425         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3426                 fs_info->data_chunk_allocations++;
3427                 if (!(fs_info->data_chunk_allocations %
3428                       fs_info->metadata_ratio))
3429                         force_metadata_allocation(fs_info);
3430         }
3431
3432         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3433         spin_lock(&space_info->lock);
3434         if (ret)
3435                 space_info->full = 1;
3436         space_info->force_alloc = 0;
3437         spin_unlock(&space_info->lock);
3438 out:
3439         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3440         return ret;
3441 }
3442
3443 static int update_block_group(struct btrfs_trans_handle *trans,
3444                               struct btrfs_root *root,
3445                               u64 bytenr, u64 num_bytes, int alloc,
3446                               int mark_free)
3447 {
3448         struct btrfs_block_group_cache *cache;
3449         struct btrfs_fs_info *info = root->fs_info;
3450         u64 total = num_bytes;
3451         u64 old_val;
3452         u64 byte_in_group;
3453
3454         /* block accounting for super block */
3455         spin_lock(&info->delalloc_lock);
3456         old_val = btrfs_super_bytes_used(&info->super_copy);
3457         if (alloc)
3458                 old_val += num_bytes;
3459         else
3460                 old_val -= num_bytes;
3461         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3462
3463         /* block accounting for root item */
3464         old_val = btrfs_root_used(&root->root_item);
3465         if (alloc)
3466                 old_val += num_bytes;
3467         else
3468                 old_val -= num_bytes;
3469         btrfs_set_root_used(&root->root_item, old_val);
3470         spin_unlock(&info->delalloc_lock);
3471
3472         while (total) {
3473                 cache = btrfs_lookup_block_group(info, bytenr);
3474                 if (!cache)
3475                         return -1;
3476                 byte_in_group = bytenr - cache->key.objectid;
3477                 WARN_ON(byte_in_group > cache->key.offset);
3478
3479                 spin_lock(&cache->space_info->lock);
3480                 spin_lock(&cache->lock);
3481                 cache->dirty = 1;
3482                 old_val = btrfs_block_group_used(&cache->item);
3483                 num_bytes = min(total, cache->key.offset - byte_in_group);
3484                 if (alloc) {
3485                         old_val += num_bytes;
3486                         btrfs_set_block_group_used(&cache->item, old_val);
3487                         cache->reserved -= num_bytes;
3488                         cache->space_info->bytes_used += num_bytes;
3489                         cache->space_info->bytes_reserved -= num_bytes;
3490                         if (cache->ro)
3491                                 cache->space_info->bytes_readonly -= num_bytes;
3492                         spin_unlock(&cache->lock);
3493                         spin_unlock(&cache->space_info->lock);
3494                 } else {
3495                         old_val -= num_bytes;
3496                         cache->space_info->bytes_used -= num_bytes;
3497                         if (cache->ro)
3498                                 cache->space_info->bytes_readonly += num_bytes;
3499                         btrfs_set_block_group_used(&cache->item, old_val);
3500                         spin_unlock(&cache->lock);
3501                         spin_unlock(&cache->space_info->lock);
3502                         if (mark_free) {
3503                                 int ret;
3504
3505                                 ret = btrfs_discard_extent(root, bytenr,
3506                                                            num_bytes);
3507                                 WARN_ON(ret);
3508
3509                                 ret = btrfs_add_free_space(cache, bytenr,
3510                                                            num_bytes);
3511                                 WARN_ON(ret);
3512                         }
3513                 }
3514                 btrfs_put_block_group(cache);
3515                 total -= num_bytes;
3516                 bytenr += num_bytes;
3517         }
3518         return 0;
3519 }
3520
3521 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3522 {
3523         struct btrfs_block_group_cache *cache;
3524         u64 bytenr;
3525
3526         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3527         if (!cache)
3528                 return 0;
3529
3530         bytenr = cache->key.objectid;
3531         btrfs_put_block_group(cache);
3532
3533         return bytenr;
3534 }
3535
3536 /*
3537  * this function must be called within transaction
3538  */
3539 int btrfs_pin_extent(struct btrfs_root *root,
3540                      u64 bytenr, u64 num_bytes, int reserved)
3541 {
3542         struct btrfs_fs_info *fs_info = root->fs_info;
3543         struct btrfs_block_group_cache *cache;
3544
3545         cache = btrfs_lookup_block_group(fs_info, bytenr);
3546         BUG_ON(!cache);
3547
3548         spin_lock(&cache->space_info->lock);
3549         spin_lock(&cache->lock);
3550         cache->pinned += num_bytes;
3551         cache->space_info->bytes_pinned += num_bytes;
3552         if (reserved) {
3553                 cache->reserved -= num_bytes;
3554                 cache->space_info->bytes_reserved -= num_bytes;
3555         }
3556         spin_unlock(&cache->lock);
3557         spin_unlock(&cache->space_info->lock);
3558
3559         btrfs_put_block_group(cache);
3560
3561         set_extent_dirty(fs_info->pinned_extents,
3562                          bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3563         return 0;
3564 }
3565
3566 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3567                                    u64 num_bytes, int reserve)
3568 {
3569         spin_lock(&cache->space_info->lock);
3570         spin_lock(&cache->lock);
3571         if (reserve) {
3572                 cache->reserved += num_bytes;
3573                 cache->space_info->bytes_reserved += num_bytes;
3574         } else {
3575                 cache->reserved -= num_bytes;
3576                 cache->space_info->bytes_reserved -= num_bytes;
3577         }
3578         spin_unlock(&cache->lock);
3579         spin_unlock(&cache->space_info->lock);
3580         return 0;
3581 }
3582
3583 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3584                                 struct btrfs_root *root)
3585 {
3586         struct btrfs_fs_info *fs_info = root->fs_info;
3587         struct btrfs_caching_control *next;
3588         struct btrfs_caching_control *caching_ctl;
3589         struct btrfs_block_group_cache *cache;
3590
3591         down_write(&fs_info->extent_commit_sem);
3592
3593         list_for_each_entry_safe(caching_ctl, next,
3594                                  &fs_info->caching_block_groups, list) {
3595                 cache = caching_ctl->block_group;
3596                 if (block_group_cache_done(cache)) {
3597                         cache->last_byte_to_unpin = (u64)-1;
3598                         list_del_init(&caching_ctl->list);
3599                         put_caching_control(caching_ctl);
3600                 } else {
3601                         cache->last_byte_to_unpin = caching_ctl->progress;
3602                 }
3603         }
3604
3605         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3606                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3607         else
3608                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3609
3610         up_write(&fs_info->extent_commit_sem);
3611         return 0;
3612 }
3613
3614 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3615 {
3616         struct btrfs_fs_info *fs_info = root->fs_info;
3617         struct btrfs_block_group_cache *cache = NULL;
3618         u64 len;
3619
3620         while (start <= end) {
3621                 if (!cache ||
3622                     start >= cache->key.objectid + cache->key.offset) {
3623                         if (cache)
3624                                 btrfs_put_block_group(cache);
3625                         cache = btrfs_lookup_block_group(fs_info, start);
3626                         BUG_ON(!cache);
3627                 }
3628
3629                 len = cache->key.objectid + cache->key.offset - start;
3630                 len = min(len, end + 1 - start);
3631
3632                 if (start < cache->last_byte_to_unpin) {
3633                         len = min(len, cache->last_byte_to_unpin - start);
3634                         btrfs_add_free_space(cache, start, len);
3635                 }
3636
3637                 spin_lock(&cache->space_info->lock);
3638                 spin_lock(&cache->lock);
3639                 cache->pinned -= len;
3640                 cache->space_info->bytes_pinned -= len;
3641                 spin_unlock(&cache->lock);
3642                 spin_unlock(&cache->space_info->lock);
3643
3644                 start += len;
3645         }
3646
3647         if (cache)
3648                 btrfs_put_block_group(cache);
3649         return 0;
3650 }
3651
3652 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3653                                struct btrfs_root *root)
3654 {
3655         struct btrfs_fs_info *fs_info = root->fs_info;
3656         struct extent_io_tree *unpin;
3657         u64 start;
3658         u64 end;
3659         int ret;
3660
3661         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3662                 unpin = &fs_info->freed_extents[1];
3663         else
3664                 unpin = &fs_info->freed_extents[0];
3665
3666         while (1) {
3667                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3668                                             EXTENT_DIRTY);
3669                 if (ret)
3670                         break;
3671
3672                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3673
3674                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3675                 unpin_extent_range(root, start, end);
3676                 cond_resched();
3677         }
3678
3679         return ret;
3680 }
3681
3682 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3683                           struct btrfs_root *root,
3684                           struct btrfs_path *path,
3685                           u64 bytenr, u64 num_bytes,
3686                           int is_data, int reserved,
3687                           struct extent_buffer **must_clean)
3688 {
3689         int err = 0;
3690         struct extent_buffer *buf;
3691
3692         if (is_data)
3693                 goto pinit;
3694
3695         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3696         if (!buf)
3697                 goto pinit;
3698
3699         /* we can reuse a block if it hasn't been written
3700          * and it is from this transaction.  We can't
3701          * reuse anything from the tree log root because
3702          * it has tiny sub-transactions.
3703          */
3704         if (btrfs_buffer_uptodate(buf, 0) &&
3705             btrfs_try_tree_lock(buf)) {
3706                 u64 header_owner = btrfs_header_owner(buf);
3707                 u64 header_transid = btrfs_header_generation(buf);
3708                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3709                     header_transid == trans->transid &&
3710                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3711                         *must_clean = buf;
3712                         return 1;
3713                 }
3714                 btrfs_tree_unlock(buf);
3715         }
3716         free_extent_buffer(buf);
3717 pinit:
3718         if (path)
3719                 btrfs_set_path_blocking(path);
3720         /* unlocks the pinned mutex */
3721         btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3722
3723         BUG_ON(err < 0);
3724         return 0;
3725 }
3726
3727 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3728                                 struct btrfs_root *root,
3729                                 u64 bytenr, u64 num_bytes, u64 parent,
3730                                 u64 root_objectid, u64 owner_objectid,
3731                                 u64 owner_offset, int refs_to_drop,
3732                                 struct btrfs_delayed_extent_op *extent_op)
3733 {
3734         struct btrfs_key key;
3735         struct btrfs_path *path;
3736         struct btrfs_fs_info *info = root->fs_info;
3737         struct btrfs_root *extent_root = info->extent_root;
3738         struct extent_buffer *leaf;
3739         struct btrfs_extent_item *ei;
3740         struct btrfs_extent_inline_ref *iref;
3741         int ret;
3742         int is_data;
3743         int extent_slot = 0;
3744         int found_extent = 0;
3745         int num_to_del = 1;
3746         u32 item_size;
3747         u64 refs;
3748
3749         path = btrfs_alloc_path();
3750         if (!path)
3751                 return -ENOMEM;
3752
3753         path->reada = 1;
3754         path->leave_spinning = 1;
3755
3756         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3757         BUG_ON(!is_data && refs_to_drop != 1);
3758
3759         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3760                                     bytenr, num_bytes, parent,
3761                                     root_objectid, owner_objectid,
3762                                     owner_offset);
3763         if (ret == 0) {
3764                 extent_slot = path->slots[0];
3765                 while (extent_slot >= 0) {
3766                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3767                                               extent_slot);
3768                         if (key.objectid != bytenr)
3769                                 break;
3770                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3771                             key.offset == num_bytes) {
3772                                 found_extent = 1;
3773                                 break;
3774                         }
3775                         if (path->slots[0] - extent_slot > 5)
3776                                 break;
3777                         extent_slot--;
3778                 }
3779 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3780                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3781                 if (found_extent && item_size < sizeof(*ei))
3782                         found_extent = 0;
3783 #endif
3784                 if (!found_extent) {
3785                         BUG_ON(iref);
3786                         ret = remove_extent_backref(trans, extent_root, path,
3787                                                     NULL, refs_to_drop,
3788                                                     is_data);
3789                         BUG_ON(ret);
3790                         btrfs_release_path(extent_root, path);
3791                         path->leave_spinning = 1;
3792
3793                         key.objectid = bytenr;
3794                         key.type = BTRFS_EXTENT_ITEM_KEY;
3795                         key.offset = num_bytes;
3796
3797                         ret = btrfs_search_slot(trans, extent_root,
3798                                                 &key, path, -1, 1);
3799                         if (ret) {
3800                                 printk(KERN_ERR "umm, got %d back from search"
3801                                        ", was looking for %llu\n", ret,
3802                                        (unsigned long long)bytenr);
3803                                 btrfs_print_leaf(extent_root, path->nodes[0]);
3804                         }
3805                         BUG_ON(ret);
3806                         extent_slot = path->slots[0];
3807                 }
3808         } else {
3809                 btrfs_print_leaf(extent_root, path->nodes[0]);
3810                 WARN_ON(1);
3811                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3812                        "parent %llu root %llu  owner %llu offset %llu\n",
3813                        (unsigned long long)bytenr,
3814                        (unsigned long long)parent,
3815                        (unsigned long long)root_objectid,
3816                        (unsigned long long)owner_objectid,
3817                        (unsigned long long)owner_offset);
3818         }
3819
3820         leaf = path->nodes[0];
3821         item_size = btrfs_item_size_nr(leaf, extent_slot);
3822 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3823         if (item_size < sizeof(*ei)) {
3824                 BUG_ON(found_extent || extent_slot != path->slots[0]);
3825                 ret = convert_extent_item_v0(trans, extent_root, path,
3826                                              owner_objectid, 0);
3827                 BUG_ON(ret < 0);
3828
3829                 btrfs_release_path(extent_root, path);
3830                 path->leave_spinning = 1;
3831
3832                 key.objectid = bytenr;
3833                 key.type = BTRFS_EXTENT_ITEM_KEY;
3834                 key.offset = num_bytes;
3835
3836                 ret = btrfs_search_slot(trans, extent_root, &key, path,
3837                                         -1, 1);
3838                 if (ret) {
3839                         printk(KERN_ERR "umm, got %d back from search"
3840                                ", was looking for %llu\n", ret,
3841                                (unsigned long long)bytenr);
3842                         btrfs_print_leaf(extent_root, path->nodes[0]);
3843                 }
3844                 BUG_ON(ret);
3845                 extent_slot = path->slots[0];
3846                 leaf = path->nodes[0];
3847                 item_size = btrfs_item_size_nr(leaf, extent_slot);
3848         }
3849 #endif
3850         BUG_ON(item_size < sizeof(*ei));
3851         ei = btrfs_item_ptr(leaf, extent_slot,
3852                             struct btrfs_extent_item);
3853         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3854                 struct btrfs_tree_block_info *bi;
3855                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3856                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3857                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3858         }
3859
3860         refs = btrfs_extent_refs(leaf, ei);
3861         BUG_ON(refs < refs_to_drop);
3862         refs -= refs_to_drop;
3863
3864         if (refs > 0) {
3865                 if (extent_op)
3866                         __run_delayed_extent_op(extent_op, leaf, ei);
3867                 /*
3868                  * In the case of inline back ref, reference count will
3869                  * be updated by remove_extent_backref
3870                  */
3871                 if (iref) {
3872                         BUG_ON(!found_extent);
3873                 } else {
3874                         btrfs_set_extent_refs(leaf, ei, refs);
3875                         btrfs_mark_buffer_dirty(leaf);
3876                 }
3877                 if (found_extent) {
3878                         ret = remove_extent_backref(trans, extent_root, path,
3879                                                     iref, refs_to_drop,
3880                                                     is_data);
3881                         BUG_ON(ret);
3882                 }
3883         } else {
3884                 int mark_free = 0;
3885                 struct extent_buffer *must_clean = NULL;
3886
3887                 if (found_extent) {
3888                         BUG_ON(is_data && refs_to_drop !=
3889                                extent_data_ref_count(root, path, iref));
3890                         if (iref) {
3891                                 BUG_ON(path->slots[0] != extent_slot);
3892                         } else {
3893                                 BUG_ON(path->slots[0] != extent_slot + 1);
3894                                 path->slots[0] = extent_slot;
3895                                 num_to_del = 2;
3896                         }
3897                 }
3898
3899                 ret = pin_down_bytes(trans, root, path, bytenr,
3900                                      num_bytes, is_data, 0, &must_clean);
3901                 if (ret > 0)
3902                         mark_free = 1;
3903                 BUG_ON(ret < 0);
3904                 /*
3905                  * it is going to be very rare for someone to be waiting
3906                  * on the block we're freeing.  del_items might need to
3907                  * schedule, so rather than get fancy, just force it
3908                  * to blocking here
3909                  */
3910                 if (must_clean)
3911                         btrfs_set_lock_blocking(must_clean);
3912
3913                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3914                                       num_to_del);
3915                 BUG_ON(ret);
3916                 btrfs_release_path(extent_root, path);
3917
3918                 if (must_clean) {
3919                         clean_tree_block(NULL, root, must_clean);
3920                         btrfs_tree_unlock(must_clean);
3921                         free_extent_buffer(must_clean);
3922                 }
3923
3924                 if (is_data) {
3925                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3926                         BUG_ON(ret);
3927                 } else {
3928                         invalidate_mapping_pages(info->btree_inode->i_mapping,
3929                              bytenr >> PAGE_CACHE_SHIFT,
3930                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3931                 }
3932
3933                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3934                                          mark_free);
3935                 BUG_ON(ret);
3936         }
3937         btrfs_free_path(path);
3938         return ret;
3939 }
3940
3941 /*
3942  * when we free an extent, it is possible (and likely) that we free the last
3943  * delayed ref for that extent as well.  This searches the delayed ref tree for
3944  * a given extent, and if there are no other delayed refs to be processed, it
3945  * removes it from the tree.
3946  */
3947 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3948                                       struct btrfs_root *root, u64 bytenr)
3949 {
3950         struct btrfs_delayed_ref_head *head;
3951         struct btrfs_delayed_ref_root *delayed_refs;
3952         struct btrfs_delayed_ref_node *ref;
3953         struct rb_node *node;
3954         int ret;
3955
3956         delayed_refs = &trans->transaction->delayed_refs;
3957         spin_lock(&delayed_refs->lock);
3958         head = btrfs_find_delayed_ref_head(trans, bytenr);
3959         if (!head)
3960                 goto out;
3961
3962         node = rb_prev(&head->node.rb_node);
3963         if (!node)
3964                 goto out;
3965
3966         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3967
3968         /* there are still entries for this ref, we can't drop it */
3969         if (ref->bytenr == bytenr)
3970                 goto out;
3971
3972         if (head->extent_op) {
3973                 if (!head->must_insert_reserved)
3974                         goto out;
3975                 kfree(head->extent_op);
3976                 head->extent_op = NULL;
3977         }
3978
3979         /*
3980          * waiting for the lock here would deadlock.  If someone else has it
3981          * locked they are already in the process of dropping it anyway
3982          */
3983         if (!mutex_trylock(&head->mutex))
3984                 goto out;
3985
3986         /*
3987          * at this point we have a head with no other entries.  Go
3988          * ahead and process it.
3989          */
3990         head->node.in_tree = 0;
3991         rb_erase(&head->node.rb_node, &delayed_refs->root);
3992
3993         delayed_refs->num_entries--;
3994
3995         /*
3996          * we don't take a ref on the node because we're removing it from the
3997          * tree, so we just steal the ref the tree was holding.
3998          */
3999         delayed_refs->num_heads--;
4000         if (list_empty(&head->cluster))
4001                 delayed_refs->num_heads_ready--;
4002
4003         list_del_init(&head->cluster);
4004         spin_unlock(&delayed_refs->lock);
4005
4006         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
4007                                   &head->node, head->extent_op,
4008                                   head->must_insert_reserved);
4009         BUG_ON(ret);
4010         btrfs_put_delayed_ref(&head->node);
4011         return 0;
4012 out:
4013         spin_unlock(&delayed_refs->lock);
4014         return 0;
4015 }
4016
4017 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4018                       struct btrfs_root *root,
4019                       u64 bytenr, u64 num_bytes, u64 parent,
4020                       u64 root_objectid, u64 owner, u64 offset)
4021 {
4022         int ret;
4023
4024         /*
4025          * tree log blocks never actually go into the extent allocation
4026          * tree, just update pinning info and exit early.
4027          */
4028         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4029                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4030                 /* unlocks the pinned mutex */
4031                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4032                 ret = 0;
4033         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4034                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4035                                         parent, root_objectid, (int)owner,
4036                                         BTRFS_DROP_DELAYED_REF, NULL);
4037                 BUG_ON(ret);
4038                 ret = check_ref_cleanup(trans, root, bytenr);
4039                 BUG_ON(ret);
4040         } else {
4041                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4042                                         parent, root_objectid, owner,
4043                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4044                 BUG_ON(ret);
4045         }
4046         return ret;
4047 }
4048
4049 static u64 stripe_align(struct btrfs_root *root, u64 val)
4050 {
4051         u64 mask = ((u64)root->stripesize - 1);
4052         u64 ret = (val + mask) & ~mask;
4053         return ret;
4054 }
4055
4056 /*
4057  * when we wait for progress in the block group caching, its because
4058  * our allocation attempt failed at least once.  So, we must sleep
4059  * and let some progress happen before we try again.
4060  *
4061  * This function will sleep at least once waiting for new free space to
4062  * show up, and then it will check the block group free space numbers
4063  * for our min num_bytes.  Another option is to have it go ahead
4064  * and look in the rbtree for a free extent of a given size, but this
4065  * is a good start.
4066  */
4067 static noinline int
4068 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4069                                 u64 num_bytes)
4070 {
4071         struct btrfs_caching_control *caching_ctl;
4072         DEFINE_WAIT(wait);
4073
4074         caching_ctl = get_caching_control(cache);
4075         if (!caching_ctl)
4076                 return 0;
4077
4078         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4079                    (cache->free_space >= num_bytes));
4080
4081         put_caching_control(caching_ctl);
4082         return 0;
4083 }
4084
4085 static noinline int
4086 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4087 {
4088         struct btrfs_caching_control *caching_ctl;
4089         DEFINE_WAIT(wait);
4090
4091         caching_ctl = get_caching_control(cache);
4092         if (!caching_ctl)
4093                 return 0;
4094
4095         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4096
4097         put_caching_control(caching_ctl);
4098         return 0;
4099 }
4100
4101 enum btrfs_loop_type {
4102         LOOP_CACHED_ONLY = 0,
4103         LOOP_CACHING_NOWAIT = 1,
4104         LOOP_CACHING_WAIT = 2,
4105         LOOP_ALLOC_CHUNK = 3,
4106         LOOP_NO_EMPTY_SIZE = 4,
4107 };
4108
4109 /*
4110  * walks the btree of allocated extents and find a hole of a given size.
4111  * The key ins is changed to record the hole:
4112  * ins->objectid == block start
4113  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4114  * ins->offset == number of blocks
4115  * Any available blocks before search_start are skipped.
4116  */
4117 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4118                                      struct btrfs_root *orig_root,
4119                                      u64 num_bytes, u64 empty_size,
4120                                      u64 search_start, u64 search_end,
4121                                      u64 hint_byte, struct btrfs_key *ins,
4122                                      u64 exclude_start, u64 exclude_nr,
4123                                      int data)
4124 {
4125         int ret = 0;
4126         struct btrfs_root *root = orig_root->fs_info->extent_root;
4127         struct btrfs_free_cluster *last_ptr = NULL;
4128         struct btrfs_block_group_cache *block_group = NULL;
4129         int empty_cluster = 2 * 1024 * 1024;
4130         int allowed_chunk_alloc = 0;
4131         struct btrfs_space_info *space_info;
4132         int last_ptr_loop = 0;
4133         int loop = 0;
4134         bool found_uncached_bg = false;
4135         bool failed_cluster_refill = false;
4136         bool failed_alloc = false;
4137
4138         WARN_ON(num_bytes < root->sectorsize);
4139         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4140         ins->objectid = 0;
4141         ins->offset = 0;
4142
4143         space_info = __find_space_info(root->fs_info, data);
4144
4145         if (orig_root->ref_cows || empty_size)
4146                 allowed_chunk_alloc = 1;
4147
4148         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4149                 last_ptr = &root->fs_info->meta_alloc_cluster;
4150                 if (!btrfs_test_opt(root, SSD))
4151                         empty_cluster = 64 * 1024;
4152         }
4153
4154         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4155                 last_ptr = &root->fs_info->data_alloc_cluster;
4156         }
4157
4158         if (last_ptr) {
4159                 spin_lock(&last_ptr->lock);
4160                 if (last_ptr->block_group)
4161                         hint_byte = last_ptr->window_start;
4162                 spin_unlock(&last_ptr->lock);
4163         }
4164
4165         search_start = max(search_start, first_logical_byte(root, 0));
4166         search_start = max(search_start, hint_byte);
4167
4168         if (!last_ptr)
4169                 empty_cluster = 0;
4170
4171         if (search_start == hint_byte) {
4172                 block_group = btrfs_lookup_block_group(root->fs_info,
4173                                                        search_start);
4174                 /*
4175                  * we don't want to use the block group if it doesn't match our
4176                  * allocation bits, or if its not cached.
4177                  */
4178                 if (block_group && block_group_bits(block_group, data) &&
4179                     block_group_cache_done(block_group)) {
4180                         down_read(&space_info->groups_sem);
4181                         if (list_empty(&block_group->list) ||
4182                             block_group->ro) {
4183                                 /*
4184                                  * someone is removing this block group,
4185                                  * we can't jump into the have_block_group
4186                                  * target because our list pointers are not
4187                                  * valid
4188                                  */
4189                                 btrfs_put_block_group(block_group);
4190                                 up_read(&space_info->groups_sem);
4191                         } else
4192                                 goto have_block_group;
4193                 } else if (block_group) {
4194                         btrfs_put_block_group(block_group);
4195                 }
4196         }
4197
4198 search:
4199         down_read(&space_info->groups_sem);
4200         list_for_each_entry(block_group, &space_info->block_groups, list) {
4201                 u64 offset;
4202                 int cached;
4203
4204                 atomic_inc(&block_group->count);
4205                 search_start = block_group->key.objectid;
4206
4207 have_block_group:
4208                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4209                         /*
4210                          * we want to start caching kthreads, but not too many
4211                          * right off the bat so we don't overwhelm the system,
4212                          * so only start them if there are less than 2 and we're
4213                          * in the initial allocation phase.
4214                          */
4215                         if (loop > LOOP_CACHING_NOWAIT ||
4216                             atomic_read(&space_info->caching_threads) < 2) {
4217                                 ret = cache_block_group(block_group);
4218                                 BUG_ON(ret);
4219                         }
4220                 }
4221
4222                 cached = block_group_cache_done(block_group);
4223                 if (unlikely(!cached)) {
4224                         found_uncached_bg = true;
4225
4226                         /* if we only want cached bgs, loop */
4227                         if (loop == LOOP_CACHED_ONLY)
4228                                 goto loop;
4229                 }
4230
4231                 if (unlikely(block_group->ro))
4232                         goto loop;
4233
4234                 /*
4235                  * Ok we want to try and use the cluster allocator, so lets look
4236                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4237                  * have tried the cluster allocator plenty of times at this
4238                  * point and not have found anything, so we are likely way too
4239                  * fragmented for the clustering stuff to find anything, so lets
4240                  * just skip it and let the allocator find whatever block it can
4241                  * find
4242                  */
4243                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4244                         /*
4245                          * the refill lock keeps out other
4246                          * people trying to start a new cluster
4247                          */
4248                         spin_lock(&last_ptr->refill_lock);
4249                         if (last_ptr->block_group &&
4250                             (last_ptr->block_group->ro ||
4251                             !block_group_bits(last_ptr->block_group, data))) {
4252                                 offset = 0;
4253                                 goto refill_cluster;
4254                         }
4255
4256                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4257                                                  num_bytes, search_start);
4258                         if (offset) {
4259                                 /* we have a block, we're done */
4260                                 spin_unlock(&last_ptr->refill_lock);
4261                                 goto checks;
4262                         }
4263
4264                         spin_lock(&last_ptr->lock);
4265                         /*
4266                          * whoops, this cluster doesn't actually point to
4267                          * this block group.  Get a ref on the block
4268                          * group is does point to and try again
4269                          */
4270                         if (!last_ptr_loop && last_ptr->block_group &&
4271                             last_ptr->block_group != block_group) {
4272
4273                                 btrfs_put_block_group(block_group);
4274                                 block_group = last_ptr->block_group;
4275                                 atomic_inc(&block_group->count);
4276                                 spin_unlock(&last_ptr->lock);
4277                                 spin_unlock(&last_ptr->refill_lock);
4278
4279                                 last_ptr_loop = 1;
4280                                 search_start = block_group->key.objectid;
4281                                 /*
4282                                  * we know this block group is properly
4283                                  * in the list because
4284                                  * btrfs_remove_block_group, drops the
4285                                  * cluster before it removes the block
4286                                  * group from the list
4287                                  */
4288                                 goto have_block_group;
4289                         }
4290                         spin_unlock(&last_ptr->lock);
4291 refill_cluster:
4292                         /*
4293                          * this cluster didn't work out, free it and
4294                          * start over
4295                          */
4296                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4297
4298                         last_ptr_loop = 0;
4299
4300                         /* allocate a cluster in this block group */
4301                         ret = btrfs_find_space_cluster(trans, root,
4302                                                block_group, last_ptr,
4303                                                offset, num_bytes,
4304                                                empty_cluster + empty_size);
4305                         if (ret == 0) {
4306                                 /*
4307                                  * now pull our allocation out of this
4308                                  * cluster
4309                                  */
4310                                 offset = btrfs_alloc_from_cluster(block_group,
4311                                                   last_ptr, num_bytes,
4312                                                   search_start);
4313                                 if (offset) {
4314                                         /* we found one, proceed */
4315                                         spin_unlock(&last_ptr->refill_lock);
4316                                         goto checks;
4317                                 }
4318                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4319                                    && !failed_cluster_refill) {
4320                                 spin_unlock(&last_ptr->refill_lock);
4321
4322                                 failed_cluster_refill = true;
4323                                 wait_block_group_cache_progress(block_group,
4324                                        num_bytes + empty_cluster + empty_size);
4325                                 goto have_block_group;
4326                         }
4327
4328                         /*
4329                          * at this point we either didn't find a cluster
4330                          * or we weren't able to allocate a block from our
4331                          * cluster.  Free the cluster we've been trying
4332                          * to use, and go to the next block group
4333                          */
4334                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4335                         spin_unlock(&last_ptr->refill_lock);
4336                         goto loop;
4337                 }
4338
4339                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4340                                                     num_bytes, empty_size);
4341                 /*
4342                  * If we didn't find a chunk, and we haven't failed on this
4343                  * block group before, and this block group is in the middle of
4344                  * caching and we are ok with waiting, then go ahead and wait
4345                  * for progress to be made, and set failed_alloc to true.
4346                  *
4347                  * If failed_alloc is true then we've already waited on this
4348                  * block group once and should move on to the next block group.
4349                  */
4350                 if (!offset && !failed_alloc && !cached &&
4351                     loop > LOOP_CACHING_NOWAIT) {
4352                         wait_block_group_cache_progress(block_group,
4353                                                 num_bytes + empty_size);
4354                         failed_alloc = true;
4355                         goto have_block_group;
4356                 } else if (!offset) {
4357                         goto loop;
4358                 }
4359 checks:
4360                 search_start = stripe_align(root, offset);
4361                 /* move on to the next group */
4362                 if (search_start + num_bytes >= search_end) {
4363                         btrfs_add_free_space(block_group, offset, num_bytes);
4364                         goto loop;
4365                 }
4366
4367                 /* move on to the next group */
4368                 if (search_start + num_bytes >
4369                     block_group->key.objectid + block_group->key.offset) {
4370                         btrfs_add_free_space(block_group, offset, num_bytes);
4371                         goto loop;
4372                 }
4373
4374                 if (exclude_nr > 0 &&
4375                     (search_start + num_bytes > exclude_start &&
4376                      search_start < exclude_start + exclude_nr)) {
4377                         search_start = exclude_start + exclude_nr;
4378
4379                         btrfs_add_free_space(block_group, offset, num_bytes);
4380                         /*
4381                          * if search_start is still in this block group
4382                          * then we just re-search this block group
4383                          */
4384                         if (search_start >= block_group->key.objectid &&
4385                             search_start < (block_group->key.objectid +
4386                                             block_group->key.offset))
4387                                 goto have_block_group;
4388                         goto loop;
4389                 }
4390
4391                 ins->objectid = search_start;
4392                 ins->offset = num_bytes;
4393
4394                 if (offset < search_start)
4395                         btrfs_add_free_space(block_group, offset,
4396                                              search_start - offset);
4397                 BUG_ON(offset > search_start);
4398
4399                 update_reserved_extents(block_group, num_bytes, 1);
4400
4401                 /* we are all good, lets return */
4402                 break;
4403 loop:
4404                 failed_cluster_refill = false;
4405                 failed_alloc = false;
4406                 btrfs_put_block_group(block_group);
4407         }
4408         up_read(&space_info->groups_sem);
4409
4410         /* LOOP_CACHED_ONLY, only search fully cached block groups
4411          * LOOP_CACHING_NOWAIT, search partially cached block groups, but
4412          *                      dont wait foR them to finish caching
4413          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4414          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4415          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4416          *                      again
4417          */
4418         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4419             (found_uncached_bg || empty_size || empty_cluster ||
4420              allowed_chunk_alloc)) {
4421                 if (found_uncached_bg) {
4422                         found_uncached_bg = false;
4423                         if (loop < LOOP_CACHING_WAIT) {
4424                                 loop++;
4425                                 goto search;
4426                         }
4427                 }
4428
4429                 if (loop == LOOP_ALLOC_CHUNK) {
4430                         empty_size = 0;
4431                         empty_cluster = 0;
4432                 }
4433
4434                 if (allowed_chunk_alloc) {
4435                         ret = do_chunk_alloc(trans, root, num_bytes +
4436                                              2 * 1024 * 1024, data, 1);
4437                         allowed_chunk_alloc = 0;
4438                 } else {
4439                         space_info->force_alloc = 1;
4440                 }
4441
4442                 if (loop < LOOP_NO_EMPTY_SIZE) {
4443                         loop++;
4444                         goto search;
4445                 }
4446                 ret = -ENOSPC;
4447         } else if (!ins->objectid) {
4448                 ret = -ENOSPC;
4449         }
4450
4451         /* we found what we needed */
4452         if (ins->objectid) {
4453                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4454                         trans->block_group = block_group->key.objectid;
4455
4456                 btrfs_put_block_group(block_group);
4457                 ret = 0;
4458         }
4459
4460         return ret;
4461 }
4462
4463 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4464                             int dump_block_groups)
4465 {
4466         struct btrfs_block_group_cache *cache;
4467
4468         spin_lock(&info->lock);
4469         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4470                (unsigned long long)(info->total_bytes - info->bytes_used -
4471                                     info->bytes_pinned - info->bytes_reserved -
4472                                     info->bytes_super),
4473                (info->full) ? "" : "not ");
4474         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4475                " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4476                "\n",
4477                (unsigned long long)info->total_bytes,
4478                (unsigned long long)info->bytes_pinned,
4479                (unsigned long long)info->bytes_delalloc,
4480                (unsigned long long)info->bytes_may_use,
4481                (unsigned long long)info->bytes_used,
4482                (unsigned long long)info->bytes_root,
4483                (unsigned long long)info->bytes_super,
4484                (unsigned long long)info->bytes_reserved);
4485         spin_unlock(&info->lock);
4486
4487         if (!dump_block_groups)
4488                 return;
4489
4490         down_read(&info->groups_sem);
4491         list_for_each_entry(cache, &info->block_groups, list) {
4492                 spin_lock(&cache->lock);
4493                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4494                        "%llu pinned %llu reserved\n",
4495                        (unsigned long long)cache->key.objectid,
4496                        (unsigned long long)cache->key.offset,
4497                        (unsigned long long)btrfs_block_group_used(&cache->item),
4498                        (unsigned long long)cache->pinned,
4499                        (unsigned long long)cache->reserved);
4500                 btrfs_dump_free_space(cache, bytes);
4501                 spin_unlock(&cache->lock);
4502         }
4503         up_read(&info->groups_sem);
4504 }
4505
4506 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4507                          struct btrfs_root *root,
4508                          u64 num_bytes, u64 min_alloc_size,
4509                          u64 empty_size, u64 hint_byte,
4510                          u64 search_end, struct btrfs_key *ins,
4511                          u64 data)
4512 {
4513         int ret;
4514         u64 search_start = 0;
4515         struct btrfs_fs_info *info = root->fs_info;
4516
4517         data = btrfs_get_alloc_profile(root, data);
4518 again:
4519         /*
4520          * the only place that sets empty_size is btrfs_realloc_node, which
4521          * is not called recursively on allocations
4522          */
4523         if (empty_size || root->ref_cows) {
4524                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
4525                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4526                                      2 * 1024 * 1024,
4527                                      BTRFS_BLOCK_GROUP_METADATA |
4528                                      (info->metadata_alloc_profile &
4529                                       info->avail_metadata_alloc_bits), 0);
4530                 }
4531                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4532                                      num_bytes + 2 * 1024 * 1024, data, 0);
4533         }
4534
4535         WARN_ON(num_bytes < root->sectorsize);
4536         ret = find_free_extent(trans, root, num_bytes, empty_size,
4537                                search_start, search_end, hint_byte, ins,
4538                                trans->alloc_exclude_start,
4539                                trans->alloc_exclude_nr, data);
4540
4541         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4542                 num_bytes = num_bytes >> 1;
4543                 num_bytes = num_bytes & ~(root->sectorsize - 1);
4544                 num_bytes = max(num_bytes, min_alloc_size);
4545                 do_chunk_alloc(trans, root->fs_info->extent_root,
4546                                num_bytes, data, 1);
4547                 goto again;
4548         }
4549         if (ret == -ENOSPC) {
4550                 struct btrfs_space_info *sinfo;
4551
4552                 sinfo = __find_space_info(root->fs_info, data);
4553                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4554                        "wanted %llu\n", (unsigned long long)data,
4555                        (unsigned long long)num_bytes);
4556                 dump_space_info(sinfo, num_bytes, 1);
4557         }
4558
4559         return ret;
4560 }
4561
4562 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4563 {
4564         struct btrfs_block_group_cache *cache;
4565         int ret = 0;
4566
4567         cache = btrfs_lookup_block_group(root->fs_info, start);
4568         if (!cache) {
4569                 printk(KERN_ERR "Unable to find block group for %llu\n",
4570                        (unsigned long long)start);
4571                 return -ENOSPC;
4572         }
4573
4574         ret = btrfs_discard_extent(root, start, len);
4575
4576         btrfs_add_free_space(cache, start, len);
4577         update_reserved_extents(cache, len, 0);
4578         btrfs_put_block_group(cache);
4579
4580         return ret;
4581 }
4582
4583 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4584                                       struct btrfs_root *root,
4585                                       u64 parent, u64 root_objectid,
4586                                       u64 flags, u64 owner, u64 offset,
4587                                       struct btrfs_key *ins, int ref_mod)
4588 {
4589         int ret;
4590         struct btrfs_fs_info *fs_info = root->fs_info;
4591         struct btrfs_extent_item *extent_item;
4592         struct btrfs_extent_inline_ref *iref;
4593         struct btrfs_path *path;
4594         struct extent_buffer *leaf;
4595         int type;
4596         u32 size;
4597
4598         if (parent > 0)
4599                 type = BTRFS_SHARED_DATA_REF_KEY;
4600         else
4601                 type = BTRFS_EXTENT_DATA_REF_KEY;
4602
4603         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4604
4605         path = btrfs_alloc_path();
4606         BUG_ON(!path);
4607
4608         path->leave_spinning = 1;
4609         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4610                                       ins, size);
4611         BUG_ON(ret);
4612
4613         leaf = path->nodes[0];
4614         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4615                                      struct btrfs_extent_item);
4616         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4617         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4618         btrfs_set_extent_flags(leaf, extent_item,
4619                                flags | BTRFS_EXTENT_FLAG_DATA);
4620
4621         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4622         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4623         if (parent > 0) {
4624                 struct btrfs_shared_data_ref *ref;
4625                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4626                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4627                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4628         } else {
4629                 struct btrfs_extent_data_ref *ref;
4630                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4631                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4632                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4633                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4634                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4635         }
4636
4637         btrfs_mark_buffer_dirty(path->nodes[0]);
4638         btrfs_free_path(path);
4639
4640         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4641                                  1, 0);
4642         if (ret) {
4643                 printk(KERN_ERR "btrfs update block group failed for %llu "
4644                        "%llu\n", (unsigned long long)ins->objectid,
4645                        (unsigned long long)ins->offset);
4646                 BUG();
4647         }
4648         return ret;
4649 }
4650
4651 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4652                                      struct btrfs_root *root,
4653                                      u64 parent, u64 root_objectid,
4654                                      u64 flags, struct btrfs_disk_key *key,
4655                                      int level, struct btrfs_key *ins)
4656 {
4657         int ret;
4658         struct btrfs_fs_info *fs_info = root->fs_info;
4659         struct btrfs_extent_item *extent_item;
4660         struct btrfs_tree_block_info *block_info;
4661         struct btrfs_extent_inline_ref *iref;
4662         struct btrfs_path *path;
4663         struct extent_buffer *leaf;
4664         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4665
4666         path = btrfs_alloc_path();
4667         BUG_ON(!path);
4668
4669         path->leave_spinning = 1;
4670         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4671                                       ins, size);
4672         BUG_ON(ret);
4673
4674         leaf = path->nodes[0];
4675         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4676                                      struct btrfs_extent_item);
4677         btrfs_set_extent_refs(leaf, extent_item, 1);
4678         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4679         btrfs_set_extent_flags(leaf, extent_item,
4680                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4681         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4682
4683         btrfs_set_tree_block_key(leaf, block_info, key);
4684         btrfs_set_tree_block_level(leaf, block_info, level);
4685
4686         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4687         if (parent > 0) {
4688                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4689                 btrfs_set_extent_inline_ref_type(leaf, iref,
4690                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4691                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4692         } else {
4693                 btrfs_set_extent_inline_ref_type(leaf, iref,
4694                                                  BTRFS_TREE_BLOCK_REF_KEY);
4695                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4696         }
4697
4698         btrfs_mark_buffer_dirty(leaf);
4699         btrfs_free_path(path);
4700
4701         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4702                                  1, 0);
4703         if (ret) {
4704                 printk(KERN_ERR "btrfs update block group failed for %llu "
4705                        "%llu\n", (unsigned long long)ins->objectid,
4706                        (unsigned long long)ins->offset);
4707                 BUG();
4708         }
4709         return ret;
4710 }
4711
4712 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4713                                      struct btrfs_root *root,
4714                                      u64 root_objectid, u64 owner,
4715                                      u64 offset, struct btrfs_key *ins)
4716 {
4717         int ret;
4718
4719         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4720
4721         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4722                                          0, root_objectid, owner, offset,
4723                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4724         return ret;
4725 }
4726
4727 /*
4728  * this is used by the tree logging recovery code.  It records that
4729  * an extent has been allocated and makes sure to clear the free
4730  * space cache bits as well
4731  */
4732 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4733                                    struct btrfs_root *root,
4734                                    u64 root_objectid, u64 owner, u64 offset,
4735                                    struct btrfs_key *ins)
4736 {
4737         int ret;
4738         struct btrfs_block_group_cache *block_group;
4739         struct btrfs_caching_control *caching_ctl;
4740         u64 start = ins->objectid;
4741         u64 num_bytes = ins->offset;
4742
4743         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4744         cache_block_group(block_group);
4745         caching_ctl = get_caching_control(block_group);
4746
4747         if (!caching_ctl) {
4748                 BUG_ON(!block_group_cache_done(block_group));
4749                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4750                 BUG_ON(ret);
4751         } else {
4752                 mutex_lock(&caching_ctl->mutex);
4753
4754                 if (start >= caching_ctl->progress) {
4755                         ret = add_excluded_extent(root, start, num_bytes);
4756                         BUG_ON(ret);
4757                 } else if (start + num_bytes <= caching_ctl->progress) {
4758                         ret = btrfs_remove_free_space(block_group,
4759                                                       start, num_bytes);
4760                         BUG_ON(ret);
4761                 } else {
4762                         num_bytes = caching_ctl->progress - start;
4763                         ret = btrfs_remove_free_space(block_group,
4764                                                       start, num_bytes);
4765                         BUG_ON(ret);
4766
4767                         start = caching_ctl->progress;
4768                         num_bytes = ins->objectid + ins->offset -
4769                                     caching_ctl->progress;
4770                         ret = add_excluded_extent(root, start, num_bytes);
4771                         BUG_ON(ret);
4772                 }
4773
4774                 mutex_unlock(&caching_ctl->mutex);
4775                 put_caching_control(caching_ctl);
4776         }
4777
4778         update_reserved_extents(block_group, ins->offset, 1);
4779         btrfs_put_block_group(block_group);
4780         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4781                                          0, owner, offset, ins, 1);
4782         return ret;
4783 }
4784
4785 /*
4786  * finds a free extent and does all the dirty work required for allocation
4787  * returns the key for the extent through ins, and a tree buffer for
4788  * the first block of the extent through buf.
4789  *
4790  * returns 0 if everything worked, non-zero otherwise.
4791  */
4792 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4793                             struct btrfs_root *root,
4794                             u64 num_bytes, u64 parent, u64 root_objectid,
4795                             struct btrfs_disk_key *key, int level,
4796                             u64 empty_size, u64 hint_byte, u64 search_end,
4797                             struct btrfs_key *ins)
4798 {
4799         int ret;
4800         u64 flags = 0;
4801
4802         ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4803                                    empty_size, hint_byte, search_end,
4804                                    ins, 0);
4805         if (ret)
4806                 return ret;
4807
4808         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4809                 if (parent == 0)
4810                         parent = ins->objectid;
4811                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4812         } else
4813                 BUG_ON(parent > 0);
4814
4815         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4816                 struct btrfs_delayed_extent_op *extent_op;
4817                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4818                 BUG_ON(!extent_op);
4819                 if (key)
4820                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4821                 else
4822                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4823                 extent_op->flags_to_set = flags;
4824                 extent_op->update_key = 1;
4825                 extent_op->update_flags = 1;
4826                 extent_op->is_data = 0;
4827
4828                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4829                                         ins->offset, parent, root_objectid,
4830                                         level, BTRFS_ADD_DELAYED_EXTENT,
4831                                         extent_op);
4832                 BUG_ON(ret);
4833         }
4834         return ret;
4835 }
4836
4837 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4838                                             struct btrfs_root *root,
4839                                             u64 bytenr, u32 blocksize,
4840                                             int level)
4841 {
4842         struct extent_buffer *buf;
4843
4844         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4845         if (!buf)
4846                 return ERR_PTR(-ENOMEM);
4847         btrfs_set_header_generation(buf, trans->transid);
4848         btrfs_set_buffer_lockdep_class(buf, level);
4849         btrfs_tree_lock(buf);
4850         clean_tree_block(trans, root, buf);
4851
4852         btrfs_set_lock_blocking(buf);
4853         btrfs_set_buffer_uptodate(buf);
4854
4855         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4856                 set_extent_dirty(&root->dirty_log_pages, buf->start,
4857                          buf->start + buf->len - 1, GFP_NOFS);
4858         } else {
4859                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4860                          buf->start + buf->len - 1, GFP_NOFS);
4861         }
4862         trans->blocks_used++;
4863         /* this returns a buffer locked for blocking */
4864         return buf;
4865 }
4866
4867 /*
4868  * helper function to allocate a block for a given tree
4869  * returns the tree buffer or NULL.
4870  */
4871 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4872                                         struct btrfs_root *root, u32 blocksize,
4873                                         u64 parent, u64 root_objectid,
4874                                         struct btrfs_disk_key *key, int level,
4875                                         u64 hint, u64 empty_size)
4876 {
4877         struct btrfs_key ins;
4878         int ret;
4879         struct extent_buffer *buf;
4880
4881         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4882                                key, level, empty_size, hint, (u64)-1, &ins);
4883         if (ret) {
4884                 BUG_ON(ret > 0);
4885                 return ERR_PTR(ret);
4886         }
4887
4888         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4889                                     blocksize, level);
4890         return buf;
4891 }
4892
4893 struct walk_control {
4894         u64 refs[BTRFS_MAX_LEVEL];
4895         u64 flags[BTRFS_MAX_LEVEL];
4896         struct btrfs_key update_progress;
4897         int stage;
4898         int level;
4899         int shared_level;
4900         int update_ref;
4901         int keep_locks;
4902         int reada_slot;
4903         int reada_count;
4904 };
4905
4906 #define DROP_REFERENCE  1
4907 #define UPDATE_BACKREF  2
4908
4909 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4910                                      struct btrfs_root *root,
4911                                      struct walk_control *wc,
4912                                      struct btrfs_path *path)
4913 {
4914         u64 bytenr;
4915         u64 generation;
4916         u64 refs;
4917         u64 flags;
4918         u64 last = 0;
4919         u32 nritems;
4920         u32 blocksize;
4921         struct btrfs_key key;
4922         struct extent_buffer *eb;
4923         int ret;
4924         int slot;
4925         int nread = 0;
4926
4927         if (path->slots[wc->level] < wc->reada_slot) {
4928                 wc->reada_count = wc->reada_count * 2 / 3;
4929                 wc->reada_count = max(wc->reada_count, 2);
4930         } else {
4931                 wc->reada_count = wc->reada_count * 3 / 2;
4932                 wc->reada_count = min_t(int, wc->reada_count,
4933                                         BTRFS_NODEPTRS_PER_BLOCK(root));
4934         }
4935
4936         eb = path->nodes[wc->level];
4937         nritems = btrfs_header_nritems(eb);
4938         blocksize = btrfs_level_size(root, wc->level - 1);
4939
4940         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4941                 if (nread >= wc->reada_count)
4942                         break;
4943
4944                 cond_resched();
4945                 bytenr = btrfs_node_blockptr(eb, slot);
4946                 generation = btrfs_node_ptr_generation(eb, slot);
4947
4948                 if (slot == path->slots[wc->level])
4949                         goto reada;
4950
4951                 if (wc->stage == UPDATE_BACKREF &&
4952                     generation <= root->root_key.offset)
4953                         continue;
4954
4955                 /* We don't lock the tree block, it's OK to be racy here */
4956                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
4957                                                &refs, &flags);
4958                 BUG_ON(ret);
4959                 BUG_ON(refs == 0);
4960
4961                 if (wc->stage == DROP_REFERENCE) {
4962                         if (refs == 1)
4963                                 goto reada;
4964
4965                         if (wc->level == 1 &&
4966                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4967                                 continue;
4968                         if (!wc->update_ref ||
4969                             generation <= root->root_key.offset)
4970                                 continue;
4971                         btrfs_node_key_to_cpu(eb, &key, slot);
4972                         ret = btrfs_comp_cpu_keys(&key,
4973                                                   &wc->update_progress);
4974                         if (ret < 0)
4975                                 continue;
4976                 } else {
4977                         if (wc->level == 1 &&
4978                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4979                                 continue;
4980                 }
4981 reada:
4982                 ret = readahead_tree_block(root, bytenr, blocksize,
4983                                            generation);
4984                 if (ret)
4985                         break;
4986                 last = bytenr + blocksize;
4987                 nread++;
4988         }
4989         wc->reada_slot = slot;
4990 }
4991
4992 /*
4993  * hepler to process tree block while walking down the tree.
4994  *
4995  * when wc->stage == UPDATE_BACKREF, this function updates
4996  * back refs for pointers in the block.
4997  *
4998  * NOTE: return value 1 means we should stop walking down.
4999  */
5000 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5001                                    struct btrfs_root *root,
5002                                    struct btrfs_path *path,
5003                                    struct walk_control *wc, int lookup_info)
5004 {
5005         int level = wc->level;
5006         struct extent_buffer *eb = path->nodes[level];
5007         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5008         int ret;
5009
5010         if (wc->stage == UPDATE_BACKREF &&
5011             btrfs_header_owner(eb) != root->root_key.objectid)
5012                 return 1;
5013
5014         /*
5015          * when reference count of tree block is 1, it won't increase
5016          * again. once full backref flag is set, we never clear it.
5017          */
5018         if (lookup_info &&
5019             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5020              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5021                 BUG_ON(!path->locks[level]);
5022                 ret = btrfs_lookup_extent_info(trans, root,
5023                                                eb->start, eb->len,
5024                                                &wc->refs[level],
5025                                                &wc->flags[level]);
5026                 BUG_ON(ret);
5027                 BUG_ON(wc->refs[level] == 0);
5028         }
5029
5030         if (wc->stage == DROP_REFERENCE) {
5031                 if (wc->refs[level] > 1)
5032                         return 1;
5033
5034                 if (path->locks[level] && !wc->keep_locks) {
5035                         btrfs_tree_unlock(eb);
5036                         path->locks[level] = 0;
5037                 }
5038                 return 0;
5039         }
5040
5041         /* wc->stage == UPDATE_BACKREF */
5042         if (!(wc->flags[level] & flag)) {
5043                 BUG_ON(!path->locks[level]);
5044                 ret = btrfs_inc_ref(trans, root, eb, 1);
5045                 BUG_ON(ret);
5046                 ret = btrfs_dec_ref(trans, root, eb, 0);
5047                 BUG_ON(ret);
5048                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5049                                                   eb->len, flag, 0);
5050                 BUG_ON(ret);
5051                 wc->flags[level] |= flag;
5052         }
5053
5054         /*
5055          * the block is shared by multiple trees, so it's not good to
5056          * keep the tree lock
5057          */
5058         if (path->locks[level] && level > 0) {
5059                 btrfs_tree_unlock(eb);
5060                 path->locks[level] = 0;
5061         }
5062         return 0;
5063 }
5064
5065 /*
5066  * hepler to process tree block pointer.
5067  *
5068  * when wc->stage == DROP_REFERENCE, this function checks
5069  * reference count of the block pointed to. if the block
5070  * is shared and we need update back refs for the subtree
5071  * rooted at the block, this function changes wc->stage to
5072  * UPDATE_BACKREF. if the block is shared and there is no
5073  * need to update back, this function drops the reference
5074  * to the block.
5075  *
5076  * NOTE: return value 1 means we should stop walking down.
5077  */
5078 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5079                                  struct btrfs_root *root,
5080                                  struct btrfs_path *path,
5081                                  struct walk_control *wc, int *lookup_info)
5082 {
5083         u64 bytenr;
5084         u64 generation;
5085         u64 parent;
5086         u32 blocksize;
5087         struct btrfs_key key;
5088         struct extent_buffer *next;
5089         int level = wc->level;
5090         int reada = 0;
5091         int ret = 0;
5092
5093         generation = btrfs_node_ptr_generation(path->nodes[level],
5094                                                path->slots[level]);
5095         /*
5096          * if the lower level block was created before the snapshot
5097          * was created, we know there is no need to update back refs
5098          * for the subtree
5099          */
5100         if (wc->stage == UPDATE_BACKREF &&
5101             generation <= root->root_key.offset) {
5102                 *lookup_info = 1;
5103                 return 1;
5104         }
5105
5106         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5107         blocksize = btrfs_level_size(root, level - 1);
5108
5109         next = btrfs_find_tree_block(root, bytenr, blocksize);
5110         if (!next) {
5111                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5112                 reada = 1;
5113         }
5114         btrfs_tree_lock(next);
5115         btrfs_set_lock_blocking(next);
5116
5117         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5118                                        &wc->refs[level - 1],
5119                                        &wc->flags[level - 1]);
5120         BUG_ON(ret);
5121         BUG_ON(wc->refs[level - 1] == 0);
5122         *lookup_info = 0;
5123
5124         if (wc->stage == DROP_REFERENCE) {
5125                 if (wc->refs[level - 1] > 1) {
5126                         if (level == 1 &&
5127                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5128                                 goto skip;
5129
5130                         if (!wc->update_ref ||
5131                             generation <= root->root_key.offset)
5132                                 goto skip;
5133
5134                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5135                                               path->slots[level]);
5136                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5137                         if (ret < 0)
5138                                 goto skip;
5139
5140                         wc->stage = UPDATE_BACKREF;
5141                         wc->shared_level = level - 1;
5142                 }
5143         } else {
5144                 if (level == 1 &&
5145                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5146                         goto skip;
5147         }
5148
5149         if (!btrfs_buffer_uptodate(next, generation)) {
5150                 btrfs_tree_unlock(next);
5151                 free_extent_buffer(next);
5152                 next = NULL;
5153                 *lookup_info = 1;
5154         }
5155
5156         if (!next) {
5157                 if (reada && level == 1)
5158                         reada_walk_down(trans, root, wc, path);
5159                 next = read_tree_block(root, bytenr, blocksize, generation);
5160                 btrfs_tree_lock(next);
5161                 btrfs_set_lock_blocking(next);
5162         }
5163
5164         level--;
5165         BUG_ON(level != btrfs_header_level(next));
5166         path->nodes[level] = next;
5167         path->slots[level] = 0;
5168         path->locks[level] = 1;
5169         wc->level = level;
5170         if (wc->level == 1)
5171                 wc->reada_slot = 0;
5172         return 0;
5173 skip:
5174         wc->refs[level - 1] = 0;
5175         wc->flags[level - 1] = 0;
5176         if (wc->stage == DROP_REFERENCE) {
5177                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5178                         parent = path->nodes[level]->start;
5179                 } else {
5180                         BUG_ON(root->root_key.objectid !=
5181                                btrfs_header_owner(path->nodes[level]));
5182                         parent = 0;
5183                 }
5184
5185                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5186                                         root->root_key.objectid, level - 1, 0);
5187                 BUG_ON(ret);
5188         }
5189         btrfs_tree_unlock(next);
5190         free_extent_buffer(next);
5191         *lookup_info = 1;
5192         return 1;
5193 }
5194
5195 /*
5196  * hepler to process tree block while walking up the tree.
5197  *
5198  * when wc->stage == DROP_REFERENCE, this function drops
5199  * reference count on the block.
5200  *
5201  * when wc->stage == UPDATE_BACKREF, this function changes
5202  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5203  * to UPDATE_BACKREF previously while processing the block.
5204  *
5205  * NOTE: return value 1 means we should stop walking up.
5206  */
5207 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5208                                  struct btrfs_root *root,
5209                                  struct btrfs_path *path,
5210                                  struct walk_control *wc)
5211 {
5212         int ret = 0;
5213         int level = wc->level;
5214         struct extent_buffer *eb = path->nodes[level];
5215         u64 parent = 0;
5216
5217         if (wc->stage == UPDATE_BACKREF) {
5218                 BUG_ON(wc->shared_level < level);
5219                 if (level < wc->shared_level)
5220                         goto out;
5221
5222                 ret = find_next_key(path, level + 1, &wc->update_progress);
5223                 if (ret > 0)
5224                         wc->update_ref = 0;
5225
5226                 wc->stage = DROP_REFERENCE;
5227                 wc->shared_level = -1;
5228                 path->slots[level] = 0;
5229
5230                 /*
5231                  * check reference count again if the block isn't locked.
5232                  * we should start walking down the tree again if reference
5233                  * count is one.
5234                  */
5235                 if (!path->locks[level]) {
5236                         BUG_ON(level == 0);
5237                         btrfs_tree_lock(eb);
5238                         btrfs_set_lock_blocking(eb);
5239                         path->locks[level] = 1;
5240
5241                         ret = btrfs_lookup_extent_info(trans, root,
5242                                                        eb->start, eb->len,
5243                                                        &wc->refs[level],
5244                                                        &wc->flags[level]);
5245                         BUG_ON(ret);
5246                         BUG_ON(wc->refs[level] == 0);
5247                         if (wc->refs[level] == 1) {
5248                                 btrfs_tree_unlock(eb);
5249                                 path->locks[level] = 0;
5250                                 return 1;
5251                         }
5252                 }
5253         }
5254
5255         /* wc->stage == DROP_REFERENCE */
5256         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5257
5258         if (wc->refs[level] == 1) {
5259                 if (level == 0) {
5260                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5261                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5262                         else
5263                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5264                         BUG_ON(ret);
5265                 }
5266                 /* make block locked assertion in clean_tree_block happy */
5267                 if (!path->locks[level] &&
5268                     btrfs_header_generation(eb) == trans->transid) {
5269                         btrfs_tree_lock(eb);
5270                         btrfs_set_lock_blocking(eb);
5271                         path->locks[level] = 1;
5272                 }
5273                 clean_tree_block(trans, root, eb);
5274         }
5275
5276         if (eb == root->node) {
5277                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5278                         parent = eb->start;
5279                 else
5280                         BUG_ON(root->root_key.objectid !=
5281                                btrfs_header_owner(eb));
5282         } else {
5283                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5284                         parent = path->nodes[level + 1]->start;
5285                 else
5286                         BUG_ON(root->root_key.objectid !=
5287                                btrfs_header_owner(path->nodes[level + 1]));
5288         }
5289
5290         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5291                                 root->root_key.objectid, level, 0);
5292         BUG_ON(ret);
5293 out:
5294         wc->refs[level] = 0;
5295         wc->flags[level] = 0;
5296         return ret;
5297 }
5298
5299 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5300                                    struct btrfs_root *root,
5301                                    struct btrfs_path *path,
5302                                    struct walk_control *wc)
5303 {
5304         int level = wc->level;
5305         int lookup_info = 1;
5306         int ret;
5307
5308         while (level >= 0) {
5309                 if (path->slots[level] >=
5310                     btrfs_header_nritems(path->nodes[level]))
5311                         break;
5312
5313                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5314                 if (ret > 0)
5315                         break;
5316
5317                 if (level == 0)
5318                         break;
5319
5320                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5321                 if (ret > 0) {
5322                         path->slots[level]++;
5323                         continue;
5324                 }
5325                 level = wc->level;
5326         }
5327         return 0;
5328 }
5329
5330 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5331                                  struct btrfs_root *root,
5332                                  struct btrfs_path *path,
5333                                  struct walk_control *wc, int max_level)
5334 {
5335         int level = wc->level;
5336         int ret;
5337
5338         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5339         while (level < max_level && path->nodes[level]) {
5340                 wc->level = level;
5341                 if (path->slots[level] + 1 <
5342                     btrfs_header_nritems(path->nodes[level])) {
5343                         path->slots[level]++;
5344                         return 0;
5345                 } else {
5346                         ret = walk_up_proc(trans, root, path, wc);
5347                         if (ret > 0)
5348                                 return 0;
5349
5350                         if (path->locks[level]) {
5351                                 btrfs_tree_unlock(path->nodes[level]);
5352                                 path->locks[level] = 0;
5353                         }
5354                         free_extent_buffer(path->nodes[level]);
5355                         path->nodes[level] = NULL;
5356                         level++;
5357                 }
5358         }
5359         return 1;
5360 }
5361
5362 /*
5363  * drop a subvolume tree.
5364  *
5365  * this function traverses the tree freeing any blocks that only
5366  * referenced by the tree.
5367  *
5368  * when a shared tree block is found. this function decreases its
5369  * reference count by one. if update_ref is true, this function
5370  * also make sure backrefs for the shared block and all lower level
5371  * blocks are properly updated.
5372  */
5373 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5374 {
5375         struct btrfs_path *path;
5376         struct btrfs_trans_handle *trans;
5377         struct btrfs_root *tree_root = root->fs_info->tree_root;
5378         struct btrfs_root_item *root_item = &root->root_item;
5379         struct walk_control *wc;
5380         struct btrfs_key key;
5381         int err = 0;
5382         int ret;
5383         int level;
5384
5385         path = btrfs_alloc_path();
5386         BUG_ON(!path);
5387
5388         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5389         BUG_ON(!wc);
5390
5391         trans = btrfs_start_transaction(tree_root, 1);
5392
5393         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5394                 level = btrfs_header_level(root->node);
5395                 path->nodes[level] = btrfs_lock_root_node(root);
5396                 btrfs_set_lock_blocking(path->nodes[level]);
5397                 path->slots[level] = 0;
5398                 path->locks[level] = 1;
5399                 memset(&wc->update_progress, 0,
5400                        sizeof(wc->update_progress));
5401         } else {
5402                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5403                 memcpy(&wc->update_progress, &key,
5404                        sizeof(wc->update_progress));
5405
5406                 level = root_item->drop_level;
5407                 BUG_ON(level == 0);
5408                 path->lowest_level = level;
5409                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5410                 path->lowest_level = 0;
5411                 if (ret < 0) {
5412                         err = ret;
5413                         goto out;
5414                 }
5415                 WARN_ON(ret > 0);
5416
5417                 /*
5418                  * unlock our path, this is safe because only this
5419                  * function is allowed to delete this snapshot
5420                  */
5421                 btrfs_unlock_up_safe(path, 0);
5422
5423                 level = btrfs_header_level(root->node);
5424                 while (1) {
5425                         btrfs_tree_lock(path->nodes[level]);
5426                         btrfs_set_lock_blocking(path->nodes[level]);
5427
5428                         ret = btrfs_lookup_extent_info(trans, root,
5429                                                 path->nodes[level]->start,
5430                                                 path->nodes[level]->len,
5431                                                 &wc->refs[level],
5432                                                 &wc->flags[level]);
5433                         BUG_ON(ret);
5434                         BUG_ON(wc->refs[level] == 0);
5435
5436                         if (level == root_item->drop_level)
5437                                 break;
5438
5439                         btrfs_tree_unlock(path->nodes[level]);
5440                         WARN_ON(wc->refs[level] != 1);
5441                         level--;
5442                 }
5443         }
5444
5445         wc->level = level;
5446         wc->shared_level = -1;
5447         wc->stage = DROP_REFERENCE;
5448         wc->update_ref = update_ref;
5449         wc->keep_locks = 0;
5450         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5451
5452         while (1) {
5453                 ret = walk_down_tree(trans, root, path, wc);
5454                 if (ret < 0) {
5455                         err = ret;
5456                         break;
5457                 }
5458
5459                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5460                 if (ret < 0) {
5461                         err = ret;
5462                         break;
5463                 }
5464
5465                 if (ret > 0) {
5466                         BUG_ON(wc->stage != DROP_REFERENCE);
5467                         break;
5468                 }
5469
5470                 if (wc->stage == DROP_REFERENCE) {
5471                         level = wc->level;
5472                         btrfs_node_key(path->nodes[level],
5473                                        &root_item->drop_progress,
5474                                        path->slots[level]);
5475                         root_item->drop_level = level;
5476                 }
5477
5478                 BUG_ON(wc->level == 0);
5479                 if (trans->transaction->in_commit ||
5480                     trans->transaction->delayed_refs.flushing) {
5481                         ret = btrfs_update_root(trans, tree_root,
5482                                                 &root->root_key,
5483                                                 root_item);
5484                         BUG_ON(ret);
5485
5486                         btrfs_end_transaction(trans, tree_root);
5487                         trans = btrfs_start_transaction(tree_root, 1);
5488                 } else {
5489                         unsigned long update;
5490                         update = trans->delayed_ref_updates;
5491                         trans->delayed_ref_updates = 0;
5492                         if (update)
5493                                 btrfs_run_delayed_refs(trans, tree_root,
5494                                                        update);
5495                 }
5496         }
5497         btrfs_release_path(root, path);
5498         BUG_ON(err);
5499
5500         ret = btrfs_del_root(trans, tree_root, &root->root_key);
5501         BUG_ON(ret);
5502
5503         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5504                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5505                                            NULL, NULL);
5506                 BUG_ON(ret < 0);
5507                 if (ret > 0) {
5508                         ret = btrfs_del_orphan_item(trans, tree_root,
5509                                                     root->root_key.objectid);
5510                         BUG_ON(ret);
5511                 }
5512         }
5513
5514         if (root->in_radix) {
5515                 btrfs_free_fs_root(tree_root->fs_info, root);
5516         } else {
5517                 free_extent_buffer(root->node);
5518                 free_extent_buffer(root->commit_root);
5519                 kfree(root);
5520         }
5521 out:
5522         btrfs_end_transaction(trans, tree_root);
5523         kfree(wc);
5524         btrfs_free_path(path);
5525         return err;
5526 }
5527
5528 /*
5529  * drop subtree rooted at tree block 'node'.
5530  *
5531  * NOTE: this function will unlock and release tree block 'node'
5532  */
5533 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5534                         struct btrfs_root *root,
5535                         struct extent_buffer *node,
5536                         struct extent_buffer *parent)
5537 {
5538         struct btrfs_path *path;
5539         struct walk_control *wc;
5540         int level;
5541         int parent_level;
5542         int ret = 0;
5543         int wret;
5544
5545         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5546
5547         path = btrfs_alloc_path();
5548         BUG_ON(!path);
5549
5550         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5551         BUG_ON(!wc);
5552
5553         btrfs_assert_tree_locked(parent);
5554         parent_level = btrfs_header_level(parent);
5555         extent_buffer_get(parent);
5556         path->nodes[parent_level] = parent;
5557         path->slots[parent_level] = btrfs_header_nritems(parent);
5558
5559         btrfs_assert_tree_locked(node);
5560         level = btrfs_header_level(node);
5561         path->nodes[level] = node;
5562         path->slots[level] = 0;
5563         path->locks[level] = 1;
5564
5565         wc->refs[parent_level] = 1;
5566         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5567         wc->level = level;
5568         wc->shared_level = -1;
5569         wc->stage = DROP_REFERENCE;
5570         wc->update_ref = 0;
5571         wc->keep_locks = 1;
5572         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5573
5574         while (1) {
5575                 wret = walk_down_tree(trans, root, path, wc);
5576                 if (wret < 0) {
5577                         ret = wret;
5578                         break;
5579                 }
5580
5581                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5582                 if (wret < 0)
5583                         ret = wret;
5584                 if (wret != 0)
5585                         break;
5586         }
5587
5588         kfree(wc);
5589         btrfs_free_path(path);
5590         return ret;
5591 }
5592
5593 #if 0
5594 static unsigned long calc_ra(unsigned long start, unsigned long last,
5595                              unsigned long nr)
5596 {
5597         return min(last, start + nr - 1);
5598 }
5599
5600 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5601                                          u64 len)
5602 {
5603         u64 page_start;
5604         u64 page_end;
5605         unsigned long first_index;
5606         unsigned long last_index;
5607         unsigned long i;
5608         struct page *page;
5609         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5610         struct file_ra_state *ra;
5611         struct btrfs_ordered_extent *ordered;
5612         unsigned int total_read = 0;
5613         unsigned int total_dirty = 0;
5614         int ret = 0;
5615
5616         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5617
5618         mutex_lock(&inode->i_mutex);
5619         first_index = start >> PAGE_CACHE_SHIFT;
5620         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5621
5622         /* make sure the dirty trick played by the caller work */
5623         ret = invalidate_inode_pages2_range(inode->i_mapping,
5624                                             first_index, last_index);
5625         if (ret)
5626                 goto out_unlock;
5627
5628         file_ra_state_init(ra, inode->i_mapping);
5629
5630         for (i = first_index ; i <= last_index; i++) {
5631                 if (total_read % ra->ra_pages == 0) {
5632                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5633                                        calc_ra(i, last_index, ra->ra_pages));
5634                 }
5635                 total_read++;
5636 again:
5637                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5638                         BUG_ON(1);
5639                 page = grab_cache_page(inode->i_mapping, i);
5640                 if (!page) {
5641                         ret = -ENOMEM;
5642                         goto out_unlock;
5643                 }
5644                 if (!PageUptodate(page)) {
5645                         btrfs_readpage(NULL, page);
5646                         lock_page(page);
5647                         if (!PageUptodate(page)) {
5648                                 unlock_page(page);
5649                                 page_cache_release(page);
5650                                 ret = -EIO;
5651                                 goto out_unlock;
5652                         }
5653                 }
5654                 wait_on_page_writeback(page);
5655
5656                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5657                 page_end = page_start + PAGE_CACHE_SIZE - 1;
5658                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5659
5660                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5661                 if (ordered) {
5662                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5663                         unlock_page(page);
5664                         page_cache_release(page);
5665                         btrfs_start_ordered_extent(inode, ordered, 1);
5666                         btrfs_put_ordered_extent(ordered);
5667                         goto again;
5668                 }
5669                 set_page_extent_mapped(page);
5670
5671                 if (i == first_index)
5672                         set_extent_bits(io_tree, page_start, page_end,
5673                                         EXTENT_BOUNDARY, GFP_NOFS);
5674                 btrfs_set_extent_delalloc(inode, page_start, page_end);
5675
5676                 set_page_dirty(page);
5677                 total_dirty++;
5678
5679                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5680                 unlock_page(page);
5681                 page_cache_release(page);
5682         }
5683
5684 out_unlock:
5685         kfree(ra);
5686         mutex_unlock(&inode->i_mutex);
5687         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5688         return ret;
5689 }
5690
5691 static noinline int relocate_data_extent(struct inode *reloc_inode,
5692                                          struct btrfs_key *extent_key,
5693                                          u64 offset)
5694 {
5695         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5696         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5697         struct extent_map *em;
5698         u64 start = extent_key->objectid - offset;
5699         u64 end = start + extent_key->offset - 1;
5700
5701         em = alloc_extent_map(GFP_NOFS);
5702         BUG_ON(!em || IS_ERR(em));
5703
5704         em->start = start;
5705         em->len = extent_key->offset;
5706         em->block_len = extent_key->offset;
5707         em->block_start = extent_key->objectid;
5708         em->bdev = root->fs_info->fs_devices->latest_bdev;
5709         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5710
5711         /* setup extent map to cheat btrfs_readpage */
5712         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5713         while (1) {
5714                 int ret;
5715                 write_lock(&em_tree->lock);
5716                 ret = add_extent_mapping(em_tree, em);
5717                 write_unlock(&em_tree->lock);
5718                 if (ret != -EEXIST) {
5719                         free_extent_map(em);
5720                         break;
5721                 }
5722                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5723         }
5724         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5725
5726         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5727 }
5728
5729 struct btrfs_ref_path {
5730         u64 extent_start;
5731         u64 nodes[BTRFS_MAX_LEVEL];
5732         u64 root_objectid;
5733         u64 root_generation;
5734         u64 owner_objectid;
5735         u32 num_refs;
5736         int lowest_level;
5737         int current_level;
5738         int shared_level;
5739
5740         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5741         u64 new_nodes[BTRFS_MAX_LEVEL];
5742 };
5743
5744 struct disk_extent {
5745         u64 ram_bytes;
5746         u64 disk_bytenr;
5747         u64 disk_num_bytes;
5748         u64 offset;
5749         u64 num_bytes;
5750         u8 compression;
5751         u8 encryption;
5752         u16 other_encoding;
5753 };
5754
5755 static int is_cowonly_root(u64 root_objectid)
5756 {
5757         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5758             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5759             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5760             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5761             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5762             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5763                 return 1;
5764         return 0;
5765 }
5766
5767 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5768                                     struct btrfs_root *extent_root,
5769                                     struct btrfs_ref_path *ref_path,
5770                                     int first_time)
5771 {
5772         struct extent_buffer *leaf;
5773         struct btrfs_path *path;
5774         struct btrfs_extent_ref *ref;
5775         struct btrfs_key key;
5776         struct btrfs_key found_key;
5777         u64 bytenr;
5778         u32 nritems;
5779         int level;
5780         int ret = 1;
5781
5782         path = btrfs_alloc_path();
5783         if (!path)
5784                 return -ENOMEM;
5785
5786         if (first_time) {
5787                 ref_path->lowest_level = -1;
5788                 ref_path->current_level = -1;
5789                 ref_path->shared_level = -1;
5790                 goto walk_up;
5791         }
5792 walk_down:
5793         level = ref_path->current_level - 1;
5794         while (level >= -1) {
5795                 u64 parent;
5796                 if (level < ref_path->lowest_level)
5797                         break;
5798
5799                 if (level >= 0)
5800                         bytenr = ref_path->nodes[level];
5801                 else
5802                         bytenr = ref_path->extent_start;
5803                 BUG_ON(bytenr == 0);
5804
5805                 parent = ref_path->nodes[level + 1];
5806                 ref_path->nodes[level + 1] = 0;
5807                 ref_path->current_level = level;
5808                 BUG_ON(parent == 0);
5809
5810                 key.objectid = bytenr;
5811                 key.offset = parent + 1;
5812                 key.type = BTRFS_EXTENT_REF_KEY;
5813
5814                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5815                 if (ret < 0)
5816                         goto out;
5817                 BUG_ON(ret == 0);
5818
5819                 leaf = path->nodes[0];
5820                 nritems = btrfs_header_nritems(leaf);
5821                 if (path->slots[0] >= nritems) {
5822                         ret = btrfs_next_leaf(extent_root, path);
5823                         if (ret < 0)
5824                                 goto out;
5825                         if (ret > 0)
5826                                 goto next;
5827                         leaf = path->nodes[0];
5828                 }
5829
5830                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5831                 if (found_key.objectid == bytenr &&
5832                     found_key.type == BTRFS_EXTENT_REF_KEY) {
5833                         if (level < ref_path->shared_level)
5834                                 ref_path->shared_level = level;
5835                         goto found;
5836                 }
5837 next:
5838                 level--;
5839                 btrfs_release_path(extent_root, path);
5840                 cond_resched();
5841         }
5842         /* reached lowest level */
5843         ret = 1;
5844         goto out;
5845 walk_up:
5846         level = ref_path->current_level;
5847         while (level < BTRFS_MAX_LEVEL - 1) {
5848                 u64 ref_objectid;
5849
5850                 if (level >= 0)
5851                         bytenr = ref_path->nodes[level];
5852                 else
5853                         bytenr = ref_path->extent_start;
5854
5855                 BUG_ON(bytenr == 0);
5856
5857                 key.objectid = bytenr;
5858                 key.offset = 0;
5859                 key.type = BTRFS_EXTENT_REF_KEY;
5860
5861                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5862                 if (ret < 0)
5863                         goto out;
5864
5865                 leaf = path->nodes[0];
5866                 nritems = btrfs_header_nritems(leaf);
5867                 if (path->slots[0] >= nritems) {
5868                         ret = btrfs_next_leaf(extent_root, path);
5869                         if (ret < 0)
5870                                 goto out;
5871                         if (ret > 0) {
5872                                 /* the extent was freed by someone */
5873                                 if (ref_path->lowest_level == level)
5874                                         goto out;
5875                                 btrfs_release_path(extent_root, path);
5876                                 goto walk_down;
5877                         }
5878                         leaf = path->nodes[0];
5879                 }
5880
5881                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5882                 if (found_key.objectid != bytenr ||
5883                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
5884                         /* the extent was freed by someone */
5885                         if (ref_path->lowest_level == level) {
5886                                 ret = 1;
5887                                 goto out;
5888                         }
5889                         btrfs_release_path(extent_root, path);
5890                         goto walk_down;
5891                 }
5892 found:
5893                 ref = btrfs_item_ptr(leaf, path->slots[0],
5894                                 struct btrfs_extent_ref);
5895                 ref_objectid = btrfs_ref_objectid(leaf, ref);
5896                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5897                         if (first_time) {
5898                                 level = (int)ref_objectid;
5899                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
5900                                 ref_path->lowest_level = level;
5901                                 ref_path->current_level = level;
5902                                 ref_path->nodes[level] = bytenr;
5903                         } else {
5904                                 WARN_ON(ref_objectid != level);
5905                         }
5906                 } else {
5907                         WARN_ON(level != -1);
5908                 }
5909                 first_time = 0;
5910
5911                 if (ref_path->lowest_level == level) {
5912                         ref_path->owner_objectid = ref_objectid;
5913                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5914                 }
5915
5916                 /*
5917                  * the block is tree root or the block isn't in reference
5918                  * counted tree.
5919                  */
5920                 if (found_key.objectid == found_key.offset ||
5921                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5922                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5923                         ref_path->root_generation =
5924                                 btrfs_ref_generation(leaf, ref);
5925                         if (level < 0) {
5926                                 /* special reference from the tree log */
5927                                 ref_path->nodes[0] = found_key.offset;
5928                                 ref_path->current_level = 0;
5929                         }
5930                         ret = 0;
5931                         goto out;
5932                 }
5933
5934                 level++;
5935                 BUG_ON(ref_path->nodes[level] != 0);
5936                 ref_path->nodes[level] = found_key.offset;
5937                 ref_path->current_level = level;
5938
5939                 /*
5940                  * the reference was created in the running transaction,
5941                  * no need to continue walking up.
5942                  */
5943                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5944                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5945                         ref_path->root_generation =
5946                                 btrfs_ref_generation(leaf, ref);
5947                         ret = 0;
5948                         goto out;
5949                 }
5950
5951                 btrfs_release_path(extent_root, path);
5952                 cond_resched();
5953         }
5954         /* reached max tree level, but no tree root found. */
5955         BUG();
5956 out:
5957         btrfs_free_path(path);
5958         return ret;
5959 }
5960
5961 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5962                                 struct btrfs_root *extent_root,
5963                                 struct btrfs_ref_path *ref_path,
5964                                 u64 extent_start)
5965 {
5966         memset(ref_path, 0, sizeof(*ref_path));
5967         ref_path->extent_start = extent_start;
5968
5969         return __next_ref_path(trans, extent_root, ref_path, 1);
5970 }
5971
5972 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5973                                struct btrfs_root *extent_root,
5974                                struct btrfs_ref_path *ref_path)
5975 {
5976         return __next_ref_path(trans, extent_root, ref_path, 0);
5977 }
5978
5979 static noinline int get_new_locations(struct inode *reloc_inode,
5980                                       struct btrfs_key *extent_key,
5981                                       u64 offset, int no_fragment,
5982                                       struct disk_extent **extents,
5983                                       int *nr_extents)
5984 {
5985         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5986         struct btrfs_path *path;
5987         struct btrfs_file_extent_item *fi;
5988         struct extent_buffer *leaf;
5989         struct disk_extent *exts = *extents;
5990         struct btrfs_key found_key;
5991         u64 cur_pos;
5992         u64 last_byte;
5993         u32 nritems;
5994         int nr = 0;
5995         int max = *nr_extents;
5996         int ret;
5997
5998         WARN_ON(!no_fragment && *extents);
5999         if (!exts) {
6000                 max = 1;
6001                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6002                 if (!exts)
6003                         return -ENOMEM;
6004         }
6005
6006         path = btrfs_alloc_path();
6007         BUG_ON(!path);
6008
6009         cur_pos = extent_key->objectid - offset;
6010         last_byte = extent_key->objectid + extent_key->offset;
6011         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6012                                        cur_pos, 0);
6013         if (ret < 0)
6014                 goto out;
6015         if (ret > 0) {
6016                 ret = -ENOENT;
6017                 goto out;
6018         }
6019
6020         while (1) {
6021                 leaf = path->nodes[0];
6022                 nritems = btrfs_header_nritems(leaf);
6023                 if (path->slots[0] >= nritems) {
6024                         ret = btrfs_next_leaf(root, path);
6025                         if (ret < 0)
6026                                 goto out;
6027                         if (ret > 0)
6028                                 break;
6029                         leaf = path->nodes[0];
6030                 }
6031
6032                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6033                 if (found_key.offset != cur_pos ||
6034                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6035                     found_key.objectid != reloc_inode->i_ino)
6036                         break;
6037
6038                 fi = btrfs_item_ptr(leaf, path->slots[0],
6039                                     struct btrfs_file_extent_item);
6040                 if (btrfs_file_extent_type(leaf, fi) !=
6041                     BTRFS_FILE_EXTENT_REG ||
6042                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6043                         break;
6044
6045                 if (nr == max) {
6046                         struct disk_extent *old = exts;
6047                         max *= 2;
6048                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6049                         memcpy(exts, old, sizeof(*exts) * nr);
6050                         if (old != *extents)
6051                                 kfree(old);
6052                 }
6053
6054                 exts[nr].disk_bytenr =
6055                         btrfs_file_extent_disk_bytenr(leaf, fi);
6056                 exts[nr].disk_num_bytes =
6057                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6058                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6059                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6060                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6061                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6062                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6063                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6064                                                                            fi);
6065                 BUG_ON(exts[nr].offset > 0);
6066                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6067                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6068
6069                 cur_pos += exts[nr].num_bytes;
6070                 nr++;
6071
6072                 if (cur_pos + offset >= last_byte)
6073                         break;
6074
6075                 if (no_fragment) {
6076                         ret = 1;
6077                         goto out;
6078                 }
6079                 path->slots[0]++;
6080         }
6081
6082         BUG_ON(cur_pos + offset > last_byte);
6083         if (cur_pos + offset < last_byte) {
6084                 ret = -ENOENT;
6085                 goto out;
6086         }
6087         ret = 0;
6088 out:
6089         btrfs_free_path(path);
6090         if (ret) {
6091                 if (exts != *extents)
6092                         kfree(exts);
6093         } else {
6094                 *extents = exts;
6095                 *nr_extents = nr;
6096         }
6097         return ret;
6098 }
6099
6100 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6101                                         struct btrfs_root *root,
6102                                         struct btrfs_path *path,
6103                                         struct btrfs_key *extent_key,
6104                                         struct btrfs_key *leaf_key,
6105                                         struct btrfs_ref_path *ref_path,
6106                                         struct disk_extent *new_extents,
6107                                         int nr_extents)
6108 {
6109         struct extent_buffer *leaf;
6110         struct btrfs_file_extent_item *fi;
6111         struct inode *inode = NULL;
6112         struct btrfs_key key;
6113         u64 lock_start = 0;
6114         u64 lock_end = 0;
6115         u64 num_bytes;
6116         u64 ext_offset;
6117         u64 search_end = (u64)-1;
6118         u32 nritems;
6119         int nr_scaned = 0;
6120         int extent_locked = 0;
6121         int extent_type;
6122         int ret;
6123
6124         memcpy(&key, leaf_key, sizeof(key));
6125         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6126                 if (key.objectid < ref_path->owner_objectid ||
6127                     (key.objectid == ref_path->owner_objectid &&
6128                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6129                         key.objectid = ref_path->owner_objectid;
6130                         key.type = BTRFS_EXTENT_DATA_KEY;
6131                         key.offset = 0;
6132                 }
6133         }
6134
6135         while (1) {
6136                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6137                 if (ret < 0)
6138                         goto out;
6139
6140                 leaf = path->nodes[0];
6141                 nritems = btrfs_header_nritems(leaf);
6142 next:
6143                 if (extent_locked && ret > 0) {
6144                         /*
6145                          * the file extent item was modified by someone
6146                          * before the extent got locked.
6147                          */
6148                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6149                                       lock_end, GFP_NOFS);
6150                         extent_locked = 0;
6151                 }
6152
6153                 if (path->slots[0] >= nritems) {
6154                         if (++nr_scaned > 2)
6155                                 break;
6156
6157                         BUG_ON(extent_locked);
6158                         ret = btrfs_next_leaf(root, path);
6159                         if (ret < 0)
6160                                 goto out;
6161                         if (ret > 0)
6162                                 break;
6163                         leaf = path->nodes[0];
6164                         nritems = btrfs_header_nritems(leaf);
6165                 }
6166
6167                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6168
6169                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6170                         if ((key.objectid > ref_path->owner_objectid) ||
6171                             (key.objectid == ref_path->owner_objectid &&
6172                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6173                             key.offset >= search_end)
6174                                 break;
6175                 }
6176
6177                 if (inode && key.objectid != inode->i_ino) {
6178                         BUG_ON(extent_locked);
6179                         btrfs_release_path(root, path);
6180                         mutex_unlock(&inode->i_mutex);
6181                         iput(inode);
6182                         inode = NULL;
6183                         continue;
6184                 }
6185
6186                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6187                         path->slots[0]++;
6188                         ret = 1;
6189                         goto next;
6190                 }
6191                 fi = btrfs_item_ptr(leaf, path->slots[0],
6192                                     struct btrfs_file_extent_item);
6193                 extent_type = btrfs_file_extent_type(leaf, fi);
6194                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6195                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6196                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6197                      extent_key->objectid)) {
6198                         path->slots[0]++;
6199                         ret = 1;
6200                         goto next;
6201                 }
6202
6203                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6204                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6205
6206                 if (search_end == (u64)-1) {
6207                         search_end = key.offset - ext_offset +
6208                                 btrfs_file_extent_ram_bytes(leaf, fi);
6209                 }
6210
6211                 if (!extent_locked) {
6212                         lock_start = key.offset;
6213                         lock_end = lock_start + num_bytes - 1;
6214                 } else {
6215                         if (lock_start > key.offset ||
6216                             lock_end + 1 < key.offset + num_bytes) {
6217                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6218                                               lock_start, lock_end, GFP_NOFS);
6219                                 extent_locked = 0;
6220                         }
6221                 }
6222
6223                 if (!inode) {
6224                         btrfs_release_path(root, path);
6225
6226                         inode = btrfs_iget_locked(root->fs_info->sb,
6227                                                   key.objectid, root);
6228                         if (inode->i_state & I_NEW) {
6229                                 BTRFS_I(inode)->root = root;
6230                                 BTRFS_I(inode)->location.objectid =
6231                                         key.objectid;
6232                                 BTRFS_I(inode)->location.type =
6233                                         BTRFS_INODE_ITEM_KEY;
6234                                 BTRFS_I(inode)->location.offset = 0;
6235                                 btrfs_read_locked_inode(inode);
6236                                 unlock_new_inode(inode);
6237                         }
6238                         /*
6239                          * some code call btrfs_commit_transaction while
6240                          * holding the i_mutex, so we can't use mutex_lock
6241                          * here.
6242                          */
6243                         if (is_bad_inode(inode) ||
6244                             !mutex_trylock(&inode->i_mutex)) {
6245                                 iput(inode);
6246                                 inode = NULL;
6247                                 key.offset = (u64)-1;
6248                                 goto skip;
6249                         }
6250                 }
6251
6252                 if (!extent_locked) {
6253                         struct btrfs_ordered_extent *ordered;
6254
6255                         btrfs_release_path(root, path);
6256
6257                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6258                                     lock_end, GFP_NOFS);
6259                         ordered = btrfs_lookup_first_ordered_extent(inode,
6260                                                                     lock_end);
6261                         if (ordered &&
6262                             ordered->file_offset <= lock_end &&
6263                             ordered->file_offset + ordered->len > lock_start) {
6264                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6265                                               lock_start, lock_end, GFP_NOFS);
6266                                 btrfs_start_ordered_extent(inode, ordered, 1);
6267                                 btrfs_put_ordered_extent(ordered);
6268                                 key.offset += num_bytes;
6269                                 goto skip;
6270                         }
6271                         if (ordered)
6272                                 btrfs_put_ordered_extent(ordered);
6273
6274                         extent_locked = 1;
6275                         continue;
6276                 }
6277
6278                 if (nr_extents == 1) {
6279                         /* update extent pointer in place */
6280                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6281                                                 new_extents[0].disk_bytenr);
6282                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6283                                                 new_extents[0].disk_num_bytes);
6284                         btrfs_mark_buffer_dirty(leaf);
6285
6286                         btrfs_drop_extent_cache(inode, key.offset,
6287                                                 key.offset + num_bytes - 1, 0);
6288
6289                         ret = btrfs_inc_extent_ref(trans, root,
6290                                                 new_extents[0].disk_bytenr,
6291                                                 new_extents[0].disk_num_bytes,
6292                                                 leaf->start,
6293                                                 root->root_key.objectid,
6294                                                 trans->transid,
6295                                                 key.objectid);
6296                         BUG_ON(ret);
6297
6298                         ret = btrfs_free_extent(trans, root,
6299                                                 extent_key->objectid,
6300                                                 extent_key->offset,
6301                                                 leaf->start,
6302                                                 btrfs_header_owner(leaf),
6303                                                 btrfs_header_generation(leaf),
6304                                                 key.objectid, 0);
6305                         BUG_ON(ret);
6306
6307                         btrfs_release_path(root, path);
6308                         key.offset += num_bytes;
6309                 } else {
6310                         BUG_ON(1);
6311 #if 0
6312                         u64 alloc_hint;
6313                         u64 extent_len;
6314                         int i;
6315                         /*
6316                          * drop old extent pointer at first, then insert the
6317                          * new pointers one bye one
6318                          */
6319                         btrfs_release_path(root, path);
6320                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
6321                                                  key.offset + num_bytes,
6322                                                  key.offset, &alloc_hint);
6323                         BUG_ON(ret);
6324
6325                         for (i = 0; i < nr_extents; i++) {
6326                                 if (ext_offset >= new_extents[i].num_bytes) {
6327                                         ext_offset -= new_extents[i].num_bytes;
6328                                         continue;
6329                                 }
6330                                 extent_len = min(new_extents[i].num_bytes -
6331                                                  ext_offset, num_bytes);
6332
6333                                 ret = btrfs_insert_empty_item(trans, root,
6334                                                               path, &key,
6335                                                               sizeof(*fi));
6336                                 BUG_ON(ret);
6337
6338                                 leaf = path->nodes[0];
6339                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6340                                                 struct btrfs_file_extent_item);
6341                                 btrfs_set_file_extent_generation(leaf, fi,
6342                                                         trans->transid);
6343                                 btrfs_set_file_extent_type(leaf, fi,
6344                                                         BTRFS_FILE_EXTENT_REG);
6345                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6346                                                 new_extents[i].disk_bytenr);
6347                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6348                                                 new_extents[i].disk_num_bytes);
6349                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6350                                                 new_extents[i].ram_bytes);
6351
6352                                 btrfs_set_file_extent_compression(leaf, fi,
6353                                                 new_extents[i].compression);
6354                                 btrfs_set_file_extent_encryption(leaf, fi,
6355                                                 new_extents[i].encryption);
6356                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6357                                                 new_extents[i].other_encoding);
6358
6359                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6360                                                         extent_len);
6361                                 ext_offset += new_extents[i].offset;
6362                                 btrfs_set_file_extent_offset(leaf, fi,
6363                                                         ext_offset);
6364                                 btrfs_mark_buffer_dirty(leaf);
6365
6366                                 btrfs_drop_extent_cache(inode, key.offset,
6367                                                 key.offset + extent_len - 1, 0);
6368
6369                                 ret = btrfs_inc_extent_ref(trans, root,
6370                                                 new_extents[i].disk_bytenr,
6371                                                 new_extents[i].disk_num_bytes,
6372                                                 leaf->start,
6373                                                 root->root_key.objectid,
6374                                                 trans->transid, key.objectid);
6375                                 BUG_ON(ret);
6376                                 btrfs_release_path(root, path);
6377
6378                                 inode_add_bytes(inode, extent_len);
6379
6380                                 ext_offset = 0;
6381                                 num_bytes -= extent_len;
6382                                 key.offset += extent_len;
6383
6384                                 if (num_bytes == 0)
6385                                         break;
6386                         }
6387                         BUG_ON(i >= nr_extents);
6388 #endif
6389                 }
6390
6391                 if (extent_locked) {
6392                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6393                                       lock_end, GFP_NOFS);
6394                         extent_locked = 0;
6395                 }
6396 skip:
6397                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6398                     key.offset >= search_end)
6399                         break;
6400
6401                 cond_resched();
6402         }
6403         ret = 0;
6404 out:
6405         btrfs_release_path(root, path);
6406         if (inode) {
6407                 mutex_unlock(&inode->i_mutex);
6408                 if (extent_locked) {
6409                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6410                                       lock_end, GFP_NOFS);
6411                 }
6412                 iput(inode);
6413         }
6414         return ret;
6415 }
6416
6417 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6418                                struct btrfs_root *root,
6419                                struct extent_buffer *buf, u64 orig_start)
6420 {
6421         int level;
6422         int ret;
6423
6424         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6425         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6426
6427         level = btrfs_header_level(buf);
6428         if (level == 0) {
6429                 struct btrfs_leaf_ref *ref;
6430                 struct btrfs_leaf_ref *orig_ref;
6431
6432                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6433                 if (!orig_ref)
6434                         return -ENOENT;
6435
6436                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6437                 if (!ref) {
6438                         btrfs_free_leaf_ref(root, orig_ref);
6439                         return -ENOMEM;
6440                 }
6441
6442                 ref->nritems = orig_ref->nritems;
6443                 memcpy(ref->extents, orig_ref->extents,
6444                         sizeof(ref->extents[0]) * ref->nritems);
6445
6446                 btrfs_free_leaf_ref(root, orig_ref);
6447
6448                 ref->root_gen = trans->transid;
6449                 ref->bytenr = buf->start;
6450                 ref->owner = btrfs_header_owner(buf);
6451                 ref->generation = btrfs_header_generation(buf);
6452
6453                 ret = btrfs_add_leaf_ref(root, ref, 0);
6454                 WARN_ON(ret);
6455                 btrfs_free_leaf_ref(root, ref);
6456         }
6457         return 0;
6458 }
6459
6460 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6461                                         struct extent_buffer *leaf,
6462                                         struct btrfs_block_group_cache *group,
6463                                         struct btrfs_root *target_root)
6464 {
6465         struct btrfs_key key;
6466         struct inode *inode = NULL;
6467         struct btrfs_file_extent_item *fi;
6468         u64 num_bytes;
6469         u64 skip_objectid = 0;
6470         u32 nritems;
6471         u32 i;
6472
6473         nritems = btrfs_header_nritems(leaf);
6474         for (i = 0; i < nritems; i++) {
6475                 btrfs_item_key_to_cpu(leaf, &key, i);
6476                 if (key.objectid == skip_objectid ||
6477                     key.type != BTRFS_EXTENT_DATA_KEY)
6478                         continue;
6479                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6480                 if (btrfs_file_extent_type(leaf, fi) ==
6481                     BTRFS_FILE_EXTENT_INLINE)
6482                         continue;
6483                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6484                         continue;
6485                 if (!inode || inode->i_ino != key.objectid) {
6486                         iput(inode);
6487                         inode = btrfs_ilookup(target_root->fs_info->sb,
6488                                               key.objectid, target_root, 1);
6489                 }
6490                 if (!inode) {
6491                         skip_objectid = key.objectid;
6492                         continue;
6493                 }
6494                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6495
6496                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6497                             key.offset + num_bytes - 1, GFP_NOFS);
6498                 btrfs_drop_extent_cache(inode, key.offset,
6499                                         key.offset + num_bytes - 1, 1);
6500                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6501                               key.offset + num_bytes - 1, GFP_NOFS);
6502                 cond_resched();
6503         }
6504         iput(inode);
6505         return 0;
6506 }
6507
6508 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6509                                         struct btrfs_root *root,
6510                                         struct extent_buffer *leaf,
6511                                         struct btrfs_block_group_cache *group,
6512                                         struct inode *reloc_inode)
6513 {
6514         struct btrfs_key key;
6515         struct btrfs_key extent_key;
6516         struct btrfs_file_extent_item *fi;
6517         struct btrfs_leaf_ref *ref;
6518         struct disk_extent *new_extent;
6519         u64 bytenr;
6520         u64 num_bytes;
6521         u32 nritems;
6522         u32 i;
6523         int ext_index;
6524         int nr_extent;
6525         int ret;
6526
6527         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6528         BUG_ON(!new_extent);
6529
6530         ref = btrfs_lookup_leaf_ref(root, leaf->start);
6531         BUG_ON(!ref);
6532
6533         ext_index = -1;
6534         nritems = btrfs_header_nritems(leaf);
6535         for (i = 0; i < nritems; i++) {
6536                 btrfs_item_key_to_cpu(leaf, &key, i);
6537                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6538                         continue;
6539                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6540                 if (btrfs_file_extent_type(leaf, fi) ==
6541                     BTRFS_FILE_EXTENT_INLINE)
6542                         continue;
6543                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6544                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6545                 if (bytenr == 0)
6546                         continue;
6547
6548                 ext_index++;
6549                 if (bytenr >= group->key.objectid + group->key.offset ||
6550                     bytenr + num_bytes <= group->key.objectid)
6551                         continue;
6552
6553                 extent_key.objectid = bytenr;
6554                 extent_key.offset = num_bytes;
6555                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6556                 nr_extent = 1;
6557                 ret = get_new_locations(reloc_inode, &extent_key,
6558                                         group->key.objectid, 1,
6559                                         &new_extent, &nr_extent);
6560                 if (ret > 0)
6561                         continue;
6562                 BUG_ON(ret < 0);
6563
6564                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6565                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6566                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6567                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6568
6569                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6570                                                 new_extent->disk_bytenr);
6571                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6572                                                 new_extent->disk_num_bytes);
6573                 btrfs_mark_buffer_dirty(leaf);
6574
6575                 ret = btrfs_inc_extent_ref(trans, root,
6576                                         new_extent->disk_bytenr,
6577                                         new_extent->disk_num_bytes,
6578                                         leaf->start,
6579                                         root->root_key.objectid,
6580                                         trans->transid, key.objectid);
6581                 BUG_ON(ret);
6582
6583                 ret = btrfs_free_extent(trans, root,
6584                                         bytenr, num_bytes, leaf->start,
6585                                         btrfs_header_owner(leaf),
6586                                         btrfs_header_generation(leaf),
6587                                         key.objectid, 0);
6588                 BUG_ON(ret);
6589                 cond_resched();
6590         }
6591         kfree(new_extent);
6592         BUG_ON(ext_index + 1 != ref->nritems);
6593         btrfs_free_leaf_ref(root, ref);
6594         return 0;
6595 }
6596
6597 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6598                           struct btrfs_root *root)
6599 {
6600         struct btrfs_root *reloc_root;
6601         int ret;
6602
6603         if (root->reloc_root) {
6604                 reloc_root = root->reloc_root;
6605                 root->reloc_root = NULL;
6606                 list_add(&reloc_root->dead_list,
6607                          &root->fs_info->dead_reloc_roots);
6608
6609                 btrfs_set_root_bytenr(&reloc_root->root_item,
6610                                       reloc_root->node->start);
6611                 btrfs_set_root_level(&root->root_item,
6612                                      btrfs_header_level(reloc_root->node));
6613                 memset(&reloc_root->root_item.drop_progress, 0,
6614                         sizeof(struct btrfs_disk_key));
6615                 reloc_root->root_item.drop_level = 0;
6616
6617                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6618                                         &reloc_root->root_key,
6619                                         &reloc_root->root_item);
6620                 BUG_ON(ret);
6621         }
6622         return 0;
6623 }
6624
6625 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6626 {
6627         struct btrfs_trans_handle *trans;
6628         struct btrfs_root *reloc_root;
6629         struct btrfs_root *prev_root = NULL;
6630         struct list_head dead_roots;
6631         int ret;
6632         unsigned long nr;
6633
6634         INIT_LIST_HEAD(&dead_roots);
6635         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6636
6637         while (!list_empty(&dead_roots)) {
6638                 reloc_root = list_entry(dead_roots.prev,
6639                                         struct btrfs_root, dead_list);
6640                 list_del_init(&reloc_root->dead_list);
6641
6642                 BUG_ON(reloc_root->commit_root != NULL);
6643                 while (1) {
6644                         trans = btrfs_join_transaction(root, 1);
6645                         BUG_ON(!trans);
6646
6647                         mutex_lock(&root->fs_info->drop_mutex);
6648                         ret = btrfs_drop_snapshot(trans, reloc_root);
6649                         if (ret != -EAGAIN)
6650                                 break;
6651                         mutex_unlock(&root->fs_info->drop_mutex);
6652
6653                         nr = trans->blocks_used;
6654                         ret = btrfs_end_transaction(trans, root);
6655                         BUG_ON(ret);
6656                         btrfs_btree_balance_dirty(root, nr);
6657                 }
6658
6659                 free_extent_buffer(reloc_root->node);
6660
6661                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6662                                      &reloc_root->root_key);
6663                 BUG_ON(ret);
6664                 mutex_unlock(&root->fs_info->drop_mutex);
6665
6666                 nr = trans->blocks_used;
6667                 ret = btrfs_end_transaction(trans, root);
6668                 BUG_ON(ret);
6669                 btrfs_btree_balance_dirty(root, nr);
6670
6671                 kfree(prev_root);
6672                 prev_root = reloc_root;
6673         }
6674         if (prev_root) {
6675                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6676                 kfree(prev_root);
6677         }
6678         return 0;
6679 }
6680
6681 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6682 {
6683         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6684         return 0;
6685 }
6686
6687 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6688 {
6689         struct btrfs_root *reloc_root;
6690         struct btrfs_trans_handle *trans;
6691         struct btrfs_key location;
6692         int found;
6693         int ret;
6694
6695         mutex_lock(&root->fs_info->tree_reloc_mutex);
6696         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6697         BUG_ON(ret);
6698         found = !list_empty(&root->fs_info->dead_reloc_roots);
6699         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6700
6701         if (found) {
6702                 trans = btrfs_start_transaction(root, 1);
6703                 BUG_ON(!trans);
6704                 ret = btrfs_commit_transaction(trans, root);
6705                 BUG_ON(ret);
6706         }
6707
6708         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6709         location.offset = (u64)-1;
6710         location.type = BTRFS_ROOT_ITEM_KEY;
6711
6712         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6713         BUG_ON(!reloc_root);
6714         btrfs_orphan_cleanup(reloc_root);
6715         return 0;
6716 }
6717
6718 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6719                                     struct btrfs_root *root)
6720 {
6721         struct btrfs_root *reloc_root;
6722         struct extent_buffer *eb;
6723         struct btrfs_root_item *root_item;
6724         struct btrfs_key root_key;
6725         int ret;
6726
6727         BUG_ON(!root->ref_cows);
6728         if (root->reloc_root)
6729                 return 0;
6730
6731         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6732         BUG_ON(!root_item);
6733
6734         ret = btrfs_copy_root(trans, root, root->commit_root,
6735                               &eb, BTRFS_TREE_RELOC_OBJECTID);
6736         BUG_ON(ret);
6737
6738         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6739         root_key.offset = root->root_key.objectid;
6740         root_key.type = BTRFS_ROOT_ITEM_KEY;
6741
6742         memcpy(root_item, &root->root_item, sizeof(root_item));
6743         btrfs_set_root_refs(root_item, 0);
6744         btrfs_set_root_bytenr(root_item, eb->start);
6745         btrfs_set_root_level(root_item, btrfs_header_level(eb));
6746         btrfs_set_root_generation(root_item, trans->transid);
6747
6748         btrfs_tree_unlock(eb);
6749         free_extent_buffer(eb);
6750
6751         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6752                                 &root_key, root_item);
6753         BUG_ON(ret);
6754         kfree(root_item);
6755
6756         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6757                                                  &root_key);
6758         BUG_ON(!reloc_root);
6759         reloc_root->last_trans = trans->transid;
6760         reloc_root->commit_root = NULL;
6761         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6762
6763         root->reloc_root = reloc_root;
6764         return 0;
6765 }
6766
6767 /*
6768  * Core function of space balance.
6769  *
6770  * The idea is using reloc trees to relocate tree blocks in reference
6771  * counted roots. There is one reloc tree for each subvol, and all
6772  * reloc trees share same root key objectid. Reloc trees are snapshots
6773  * of the latest committed roots of subvols (root->commit_root).
6774  *
6775  * To relocate a tree block referenced by a subvol, there are two steps.
6776  * COW the block through subvol's reloc tree, then update block pointer
6777  * in the subvol to point to the new block. Since all reloc trees share
6778  * same root key objectid, doing special handing for tree blocks owned
6779  * by them is easy. Once a tree block has been COWed in one reloc tree,
6780  * we can use the resulting new block directly when the same block is
6781  * required to COW again through other reloc trees. By this way, relocated
6782  * tree blocks are shared between reloc trees, so they are also shared
6783  * between subvols.
6784  */
6785 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6786                                       struct btrfs_root *root,
6787                                       struct btrfs_path *path,
6788                                       struct btrfs_key *first_key,
6789                                       struct btrfs_ref_path *ref_path,
6790                                       struct btrfs_block_group_cache *group,
6791                                       struct inode *reloc_inode)
6792 {
6793         struct btrfs_root *reloc_root;
6794         struct extent_buffer *eb = NULL;
6795         struct btrfs_key *keys;
6796         u64 *nodes;
6797         int level;
6798         int shared_level;
6799         int lowest_level = 0;
6800         int ret;
6801
6802         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6803                 lowest_level = ref_path->owner_objectid;
6804
6805         if (!root->ref_cows) {
6806                 path->lowest_level = lowest_level;
6807                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6808                 BUG_ON(ret < 0);
6809                 path->lowest_level = 0;
6810                 btrfs_release_path(root, path);
6811                 return 0;
6812         }
6813
6814         mutex_lock(&root->fs_info->tree_reloc_mutex);
6815         ret = init_reloc_tree(trans, root);
6816         BUG_ON(ret);
6817         reloc_root = root->reloc_root;
6818
6819         shared_level = ref_path->shared_level;
6820         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6821
6822         keys = ref_path->node_keys;
6823         nodes = ref_path->new_nodes;
6824         memset(&keys[shared_level + 1], 0,
6825                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6826         memset(&nodes[shared_level + 1], 0,
6827                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6828
6829         if (nodes[lowest_level] == 0) {
6830                 path->lowest_level = lowest_level;
6831                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6832                                         0, 1);
6833                 BUG_ON(ret);
6834                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6835                         eb = path->nodes[level];
6836                         if (!eb || eb == reloc_root->node)
6837                                 break;
6838                         nodes[level] = eb->start;
6839                         if (level == 0)
6840                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6841                         else
6842                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6843                 }
6844                 if (nodes[0] &&
6845                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6846                         eb = path->nodes[0];
6847                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
6848                                                       group, reloc_inode);
6849                         BUG_ON(ret);
6850                 }
6851                 btrfs_release_path(reloc_root, path);
6852         } else {
6853                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6854                                        lowest_level);
6855                 BUG_ON(ret);
6856         }
6857
6858         /*
6859          * replace tree blocks in the fs tree with tree blocks in
6860          * the reloc tree.
6861          */
6862         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6863         BUG_ON(ret < 0);
6864
6865         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6866                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6867                                         0, 0);
6868                 BUG_ON(ret);
6869                 extent_buffer_get(path->nodes[0]);
6870                 eb = path->nodes[0];
6871                 btrfs_release_path(reloc_root, path);
6872                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6873                 BUG_ON(ret);
6874                 free_extent_buffer(eb);
6875         }
6876
6877         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6878         path->lowest_level = 0;
6879         return 0;
6880 }
6881
6882 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6883                                         struct btrfs_root *root,
6884                                         struct btrfs_path *path,
6885                                         struct btrfs_key *first_key,
6886                                         struct btrfs_ref_path *ref_path)
6887 {
6888         int ret;
6889
6890         ret = relocate_one_path(trans, root, path, first_key,
6891                                 ref_path, NULL, NULL);
6892         BUG_ON(ret);
6893
6894         return 0;
6895 }
6896
6897 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6898                                     struct btrfs_root *extent_root,
6899                                     struct btrfs_path *path,
6900                                     struct btrfs_key *extent_key)
6901 {
6902         int ret;
6903
6904         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6905         if (ret)
6906                 goto out;
6907         ret = btrfs_del_item(trans, extent_root, path);
6908 out:
6909         btrfs_release_path(extent_root, path);
6910         return ret;
6911 }
6912
6913 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6914                                                 struct btrfs_ref_path *ref_path)
6915 {
6916         struct btrfs_key root_key;
6917
6918         root_key.objectid = ref_path->root_objectid;
6919         root_key.type = BTRFS_ROOT_ITEM_KEY;
6920         if (is_cowonly_root(ref_path->root_objectid))
6921                 root_key.offset = 0;
6922         else
6923                 root_key.offset = (u64)-1;
6924
6925         return btrfs_read_fs_root_no_name(fs_info, &root_key);
6926 }
6927
6928 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6929                                         struct btrfs_path *path,
6930                                         struct btrfs_key *extent_key,
6931                                         struct btrfs_block_group_cache *group,
6932                                         struct inode *reloc_inode, int pass)
6933 {
6934         struct btrfs_trans_handle *trans;
6935         struct btrfs_root *found_root;
6936         struct btrfs_ref_path *ref_path = NULL;
6937         struct disk_extent *new_extents = NULL;
6938         int nr_extents = 0;
6939         int loops;
6940         int ret;
6941         int level;
6942         struct btrfs_key first_key;
6943         u64 prev_block = 0;
6944
6945
6946         trans = btrfs_start_transaction(extent_root, 1);
6947         BUG_ON(!trans);
6948
6949         if (extent_key->objectid == 0) {
6950                 ret = del_extent_zero(trans, extent_root, path, extent_key);
6951                 goto out;
6952         }
6953
6954         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6955         if (!ref_path) {
6956                 ret = -ENOMEM;
6957                 goto out;
6958         }
6959
6960         for (loops = 0; ; loops++) {
6961                 if (loops == 0) {
6962                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6963                                                    extent_key->objectid);
6964                 } else {
6965                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6966                 }
6967                 if (ret < 0)
6968                         goto out;
6969                 if (ret > 0)
6970                         break;
6971
6972                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6973                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6974                         continue;
6975
6976                 found_root = read_ref_root(extent_root->fs_info, ref_path);
6977                 BUG_ON(!found_root);
6978                 /*
6979                  * for reference counted tree, only process reference paths
6980                  * rooted at the latest committed root.
6981                  */
6982                 if (found_root->ref_cows &&
6983                     ref_path->root_generation != found_root->root_key.offset)
6984                         continue;
6985
6986                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6987                         if (pass == 0) {
6988                                 /*
6989                                  * copy data extents to new locations
6990                                  */
6991                                 u64 group_start = group->key.objectid;
6992                                 ret = relocate_data_extent(reloc_inode,
6993                                                            extent_key,
6994                                                            group_start);
6995                                 if (ret < 0)
6996                                         goto out;
6997                                 break;
6998                         }
6999                         level = 0;
7000                 } else {
7001                         level = ref_path->owner_objectid;
7002                 }
7003
7004                 if (prev_block != ref_path->nodes[level]) {
7005                         struct extent_buffer *eb;
7006                         u64 block_start = ref_path->nodes[level];
7007                         u64 block_size = btrfs_level_size(found_root, level);
7008
7009                         eb = read_tree_block(found_root, block_start,
7010                                              block_size, 0);
7011                         btrfs_tree_lock(eb);
7012                         BUG_ON(level != btrfs_header_level(eb));
7013
7014                         if (level == 0)
7015                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7016                         else
7017                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7018
7019                         btrfs_tree_unlock(eb);
7020                         free_extent_buffer(eb);
7021                         prev_block = block_start;
7022                 }
7023
7024                 mutex_lock(&extent_root->fs_info->trans_mutex);
7025                 btrfs_record_root_in_trans(found_root);
7026                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7027                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7028                         /*
7029                          * try to update data extent references while
7030                          * keeping metadata shared between snapshots.
7031                          */
7032                         if (pass == 1) {
7033                                 ret = relocate_one_path(trans, found_root,
7034                                                 path, &first_key, ref_path,
7035                                                 group, reloc_inode);
7036                                 if (ret < 0)
7037                                         goto out;
7038                                 continue;
7039                         }
7040                         /*
7041                          * use fallback method to process the remaining
7042                          * references.
7043                          */
7044                         if (!new_extents) {
7045                                 u64 group_start = group->key.objectid;
7046                                 new_extents = kmalloc(sizeof(*new_extents),
7047                                                       GFP_NOFS);
7048                                 nr_extents = 1;
7049                                 ret = get_new_locations(reloc_inode,
7050                                                         extent_key,
7051                                                         group_start, 1,
7052                                                         &new_extents,
7053                                                         &nr_extents);
7054                                 if (ret)
7055                                         goto out;
7056                         }
7057                         ret = replace_one_extent(trans, found_root,
7058                                                 path, extent_key,
7059                                                 &first_key, ref_path,
7060                                                 new_extents, nr_extents);
7061                 } else {
7062                         ret = relocate_tree_block(trans, found_root, path,
7063                                                   &first_key, ref_path);
7064                 }
7065                 if (ret < 0)
7066                         goto out;
7067         }
7068         ret = 0;
7069 out:
7070         btrfs_end_transaction(trans, extent_root);
7071         kfree(new_extents);
7072         kfree(ref_path);
7073         return ret;
7074 }
7075 #endif
7076
7077 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7078 {
7079         u64 num_devices;
7080         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7081                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7082
7083         num_devices = root->fs_info->fs_devices->rw_devices;
7084         if (num_devices == 1) {
7085                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7086                 stripped = flags & ~stripped;
7087
7088                 /* turn raid0 into single device chunks */
7089                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7090                         return stripped;
7091
7092                 /* turn mirroring into duplication */
7093                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7094                              BTRFS_BLOCK_GROUP_RAID10))
7095                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7096                 return flags;
7097         } else {
7098                 /* they already had raid on here, just return */
7099                 if (flags & stripped)
7100                         return flags;
7101
7102                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7103                 stripped = flags & ~stripped;
7104
7105                 /* switch duplicated blocks with raid1 */
7106                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7107                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7108
7109                 /* turn single device chunks into raid0 */
7110                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7111         }
7112         return flags;
7113 }
7114
7115 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
7116                      struct btrfs_block_group_cache *shrink_block_group,
7117                      int force)
7118 {
7119         struct btrfs_trans_handle *trans;
7120         u64 new_alloc_flags;
7121         u64 calc;
7122
7123         spin_lock(&shrink_block_group->lock);
7124         if (btrfs_block_group_used(&shrink_block_group->item) +
7125             shrink_block_group->reserved > 0) {
7126                 spin_unlock(&shrink_block_group->lock);
7127
7128                 trans = btrfs_start_transaction(root, 1);
7129                 spin_lock(&shrink_block_group->lock);
7130
7131                 new_alloc_flags = update_block_group_flags(root,
7132                                                    shrink_block_group->flags);
7133                 if (new_alloc_flags != shrink_block_group->flags) {
7134                         calc =
7135                              btrfs_block_group_used(&shrink_block_group->item);
7136                 } else {
7137                         calc = shrink_block_group->key.offset;
7138                 }
7139                 spin_unlock(&shrink_block_group->lock);
7140
7141                 do_chunk_alloc(trans, root->fs_info->extent_root,
7142                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
7143
7144                 btrfs_end_transaction(trans, root);
7145         } else
7146                 spin_unlock(&shrink_block_group->lock);
7147         return 0;
7148 }
7149
7150
7151 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
7152                                          struct btrfs_block_group_cache *group)
7153
7154 {
7155         __alloc_chunk_for_shrink(root, group, 1);
7156         set_block_group_readonly(group);
7157         return 0;
7158 }
7159
7160 /*
7161  * checks to see if its even possible to relocate this block group.
7162  *
7163  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7164  * ok to go ahead and try.
7165  */
7166 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7167 {
7168         struct btrfs_block_group_cache *block_group;
7169         struct btrfs_space_info *space_info;
7170         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7171         struct btrfs_device *device;
7172         int full = 0;
7173         int ret = 0;
7174
7175         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7176
7177         /* odd, couldn't find the block group, leave it alone */
7178         if (!block_group)
7179                 return -1;
7180
7181         /* no bytes used, we're good */
7182         if (!btrfs_block_group_used(&block_group->item))
7183                 goto out;
7184
7185         space_info = block_group->space_info;
7186         spin_lock(&space_info->lock);
7187
7188         full = space_info->full;
7189
7190         /*
7191          * if this is the last block group we have in this space, we can't
7192          * relocate it unless we're able to allocate a new chunk below.
7193          *
7194          * Otherwise, we need to make sure we have room in the space to handle
7195          * all of the extents from this block group.  If we can, we're good
7196          */
7197         if ((space_info->total_bytes != block_group->key.offset) &&
7198            (space_info->bytes_used + space_info->bytes_reserved +
7199             space_info->bytes_pinned + space_info->bytes_readonly +
7200             btrfs_block_group_used(&block_group->item) <
7201             space_info->total_bytes)) {
7202                 spin_unlock(&space_info->lock);
7203                 goto out;
7204         }
7205         spin_unlock(&space_info->lock);
7206
7207         /*
7208          * ok we don't have enough space, but maybe we have free space on our
7209          * devices to allocate new chunks for relocation, so loop through our
7210          * alloc devices and guess if we have enough space.  However, if we
7211          * were marked as full, then we know there aren't enough chunks, and we
7212          * can just return.
7213          */
7214         ret = -1;
7215         if (full)
7216                 goto out;
7217
7218         mutex_lock(&root->fs_info->chunk_mutex);
7219         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7220                 u64 min_free = btrfs_block_group_used(&block_group->item);
7221                 u64 dev_offset, max_avail;
7222
7223                 /*
7224                  * check to make sure we can actually find a chunk with enough
7225                  * space to fit our block group in.
7226                  */
7227                 if (device->total_bytes > device->bytes_used + min_free) {
7228                         ret = find_free_dev_extent(NULL, device, min_free,
7229                                                    &dev_offset, &max_avail);
7230                         if (!ret)
7231                                 break;
7232                         ret = -1;
7233                 }
7234         }
7235         mutex_unlock(&root->fs_info->chunk_mutex);
7236 out:
7237         btrfs_put_block_group(block_group);
7238         return ret;
7239 }
7240
7241 static int find_first_block_group(struct btrfs_root *root,
7242                 struct btrfs_path *path, struct btrfs_key *key)
7243 {
7244         int ret = 0;
7245         struct btrfs_key found_key;
7246         struct extent_buffer *leaf;
7247         int slot;
7248
7249         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7250         if (ret < 0)
7251                 goto out;
7252
7253         while (1) {
7254                 slot = path->slots[0];
7255                 leaf = path->nodes[0];
7256                 if (slot >= btrfs_header_nritems(leaf)) {
7257                         ret = btrfs_next_leaf(root, path);
7258                         if (ret == 0)
7259                                 continue;
7260                         if (ret < 0)
7261                                 goto out;
7262                         break;
7263                 }
7264                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7265
7266                 if (found_key.objectid >= key->objectid &&
7267                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7268                         ret = 0;
7269                         goto out;
7270                 }
7271                 path->slots[0]++;
7272         }
7273         ret = -ENOENT;
7274 out:
7275         return ret;
7276 }
7277
7278 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7279 {
7280         struct btrfs_block_group_cache *block_group;
7281         struct btrfs_space_info *space_info;
7282         struct btrfs_caching_control *caching_ctl;
7283         struct rb_node *n;
7284
7285         down_write(&info->extent_commit_sem);
7286         while (!list_empty(&info->caching_block_groups)) {
7287                 caching_ctl = list_entry(info->caching_block_groups.next,
7288                                          struct btrfs_caching_control, list);
7289                 list_del(&caching_ctl->list);
7290                 put_caching_control(caching_ctl);
7291         }
7292         up_write(&info->extent_commit_sem);
7293
7294         spin_lock(&info->block_group_cache_lock);
7295         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7296                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7297                                        cache_node);
7298                 rb_erase(&block_group->cache_node,
7299                          &info->block_group_cache_tree);
7300                 spin_unlock(&info->block_group_cache_lock);
7301
7302                 down_write(&block_group->space_info->groups_sem);
7303                 list_del(&block_group->list);
7304                 up_write(&block_group->space_info->groups_sem);
7305
7306                 if (block_group->cached == BTRFS_CACHE_STARTED)
7307                         wait_block_group_cache_done(block_group);
7308
7309                 btrfs_remove_free_space_cache(block_group);
7310
7311                 WARN_ON(atomic_read(&block_group->count) != 1);
7312                 kfree(block_group);
7313
7314                 spin_lock(&info->block_group_cache_lock);
7315         }
7316         spin_unlock(&info->block_group_cache_lock);
7317
7318         /* now that all the block groups are freed, go through and
7319          * free all the space_info structs.  This is only called during
7320          * the final stages of unmount, and so we know nobody is
7321          * using them.  We call synchronize_rcu() once before we start,
7322          * just to be on the safe side.
7323          */
7324         synchronize_rcu();
7325
7326         while(!list_empty(&info->space_info)) {
7327                 space_info = list_entry(info->space_info.next,
7328                                         struct btrfs_space_info,
7329                                         list);
7330
7331                 list_del(&space_info->list);
7332                 kfree(space_info);
7333         }
7334         return 0;
7335 }
7336
7337 int btrfs_read_block_groups(struct btrfs_root *root)
7338 {
7339         struct btrfs_path *path;
7340         int ret;
7341         struct btrfs_block_group_cache *cache;
7342         struct btrfs_fs_info *info = root->fs_info;
7343         struct btrfs_space_info *space_info;
7344         struct btrfs_key key;
7345         struct btrfs_key found_key;
7346         struct extent_buffer *leaf;
7347
7348         root = info->extent_root;
7349         key.objectid = 0;
7350         key.offset = 0;
7351         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7352         path = btrfs_alloc_path();
7353         if (!path)
7354                 return -ENOMEM;
7355
7356         while (1) {
7357                 ret = find_first_block_group(root, path, &key);
7358                 if (ret > 0) {
7359                         ret = 0;
7360                         goto error;
7361                 }
7362                 if (ret != 0)
7363                         goto error;
7364
7365                 leaf = path->nodes[0];
7366                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7367                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7368                 if (!cache) {
7369                         ret = -ENOMEM;
7370                         break;
7371                 }
7372
7373                 atomic_set(&cache->count, 1);
7374                 spin_lock_init(&cache->lock);
7375                 spin_lock_init(&cache->tree_lock);
7376                 cache->fs_info = info;
7377                 INIT_LIST_HEAD(&cache->list);
7378                 INIT_LIST_HEAD(&cache->cluster_list);
7379
7380                 /*
7381                  * we only want to have 32k of ram per block group for keeping
7382                  * track of free space, and if we pass 1/2 of that we want to
7383                  * start converting things over to using bitmaps
7384                  */
7385                 cache->extents_thresh = ((1024 * 32) / 2) /
7386                         sizeof(struct btrfs_free_space);
7387
7388                 read_extent_buffer(leaf, &cache->item,
7389                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7390                                    sizeof(cache->item));
7391                 memcpy(&cache->key, &found_key, sizeof(found_key));
7392
7393                 key.objectid = found_key.objectid + found_key.offset;
7394                 btrfs_release_path(root, path);
7395                 cache->flags = btrfs_block_group_flags(&cache->item);
7396                 cache->sectorsize = root->sectorsize;
7397
7398                 /*
7399                  * check for two cases, either we are full, and therefore
7400                  * don't need to bother with the caching work since we won't
7401                  * find any space, or we are empty, and we can just add all
7402                  * the space in and be done with it.  This saves us _alot_ of
7403                  * time, particularly in the full case.
7404                  */
7405                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7406                         exclude_super_stripes(root, cache);
7407                         cache->last_byte_to_unpin = (u64)-1;
7408                         cache->cached = BTRFS_CACHE_FINISHED;
7409                         free_excluded_extents(root, cache);
7410                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7411                         exclude_super_stripes(root, cache);
7412                         cache->last_byte_to_unpin = (u64)-1;
7413                         cache->cached = BTRFS_CACHE_FINISHED;
7414                         add_new_free_space(cache, root->fs_info,
7415                                            found_key.objectid,
7416                                            found_key.objectid +
7417                                            found_key.offset);
7418                         free_excluded_extents(root, cache);
7419                 }
7420
7421                 ret = update_space_info(info, cache->flags, found_key.offset,
7422                                         btrfs_block_group_used(&cache->item),
7423                                         &space_info);
7424                 BUG_ON(ret);
7425                 cache->space_info = space_info;
7426                 spin_lock(&cache->space_info->lock);
7427                 cache->space_info->bytes_super += cache->bytes_super;
7428                 spin_unlock(&cache->space_info->lock);
7429
7430                 down_write(&space_info->groups_sem);
7431                 list_add_tail(&cache->list, &space_info->block_groups);
7432                 up_write(&space_info->groups_sem);
7433
7434                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7435                 BUG_ON(ret);
7436
7437                 set_avail_alloc_bits(root->fs_info, cache->flags);
7438                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7439                         set_block_group_readonly(cache);
7440         }
7441         ret = 0;
7442 error:
7443         btrfs_free_path(path);
7444         return ret;
7445 }
7446
7447 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7448                            struct btrfs_root *root, u64 bytes_used,
7449                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7450                            u64 size)
7451 {
7452         int ret;
7453         struct btrfs_root *extent_root;
7454         struct btrfs_block_group_cache *cache;
7455
7456         extent_root = root->fs_info->extent_root;
7457
7458         root->fs_info->last_trans_log_full_commit = trans->transid;
7459
7460         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7461         if (!cache)
7462                 return -ENOMEM;
7463
7464         cache->key.objectid = chunk_offset;
7465         cache->key.offset = size;
7466         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7467         cache->sectorsize = root->sectorsize;
7468
7469         /*
7470          * we only want to have 32k of ram per block group for keeping track
7471          * of free space, and if we pass 1/2 of that we want to start
7472          * converting things over to using bitmaps
7473          */
7474         cache->extents_thresh = ((1024 * 32) / 2) /
7475                 sizeof(struct btrfs_free_space);
7476         atomic_set(&cache->count, 1);
7477         spin_lock_init(&cache->lock);
7478         spin_lock_init(&cache->tree_lock);
7479         INIT_LIST_HEAD(&cache->list);
7480         INIT_LIST_HEAD(&cache->cluster_list);
7481
7482         btrfs_set_block_group_used(&cache->item, bytes_used);
7483         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7484         cache->flags = type;
7485         btrfs_set_block_group_flags(&cache->item, type);
7486
7487         cache->last_byte_to_unpin = (u64)-1;
7488         cache->cached = BTRFS_CACHE_FINISHED;
7489         exclude_super_stripes(root, cache);
7490
7491         add_new_free_space(cache, root->fs_info, chunk_offset,
7492                            chunk_offset + size);
7493
7494         free_excluded_extents(root, cache);
7495
7496         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7497                                 &cache->space_info);
7498         BUG_ON(ret);
7499
7500         spin_lock(&cache->space_info->lock);
7501         cache->space_info->bytes_super += cache->bytes_super;
7502         spin_unlock(&cache->space_info->lock);
7503
7504         down_write(&cache->space_info->groups_sem);
7505         list_add_tail(&cache->list, &cache->space_info->block_groups);
7506         up_write(&cache->space_info->groups_sem);
7507
7508         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7509         BUG_ON(ret);
7510
7511         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7512                                 sizeof(cache->item));
7513         BUG_ON(ret);
7514
7515         set_avail_alloc_bits(extent_root->fs_info, type);
7516
7517         return 0;
7518 }
7519
7520 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7521                              struct btrfs_root *root, u64 group_start)
7522 {
7523         struct btrfs_path *path;
7524         struct btrfs_block_group_cache *block_group;
7525         struct btrfs_free_cluster *cluster;
7526         struct btrfs_key key;
7527         int ret;
7528
7529         root = root->fs_info->extent_root;
7530
7531         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7532         BUG_ON(!block_group);
7533         BUG_ON(!block_group->ro);
7534
7535         memcpy(&key, &block_group->key, sizeof(key));
7536
7537         /* make sure this block group isn't part of an allocation cluster */
7538         cluster = &root->fs_info->data_alloc_cluster;
7539         spin_lock(&cluster->refill_lock);
7540         btrfs_return_cluster_to_free_space(block_group, cluster);
7541         spin_unlock(&cluster->refill_lock);
7542
7543         /*
7544          * make sure this block group isn't part of a metadata
7545          * allocation cluster
7546          */
7547         cluster = &root->fs_info->meta_alloc_cluster;
7548         spin_lock(&cluster->refill_lock);
7549         btrfs_return_cluster_to_free_space(block_group, cluster);
7550         spin_unlock(&cluster->refill_lock);
7551
7552         path = btrfs_alloc_path();
7553         BUG_ON(!path);
7554
7555         spin_lock(&root->fs_info->block_group_cache_lock);
7556         rb_erase(&block_group->cache_node,
7557                  &root->fs_info->block_group_cache_tree);
7558         spin_unlock(&root->fs_info->block_group_cache_lock);
7559
7560         down_write(&block_group->space_info->groups_sem);
7561         /*
7562          * we must use list_del_init so people can check to see if they
7563          * are still on the list after taking the semaphore
7564          */
7565         list_del_init(&block_group->list);
7566         up_write(&block_group->space_info->groups_sem);
7567
7568         if (block_group->cached == BTRFS_CACHE_STARTED)
7569                 wait_block_group_cache_done(block_group);
7570
7571         btrfs_remove_free_space_cache(block_group);
7572
7573         spin_lock(&block_group->space_info->lock);
7574         block_group->space_info->total_bytes -= block_group->key.offset;
7575         block_group->space_info->bytes_readonly -= block_group->key.offset;
7576         spin_unlock(&block_group->space_info->lock);
7577
7578         btrfs_clear_space_info_full(root->fs_info);
7579
7580         btrfs_put_block_group(block_group);
7581         btrfs_put_block_group(block_group);
7582
7583         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7584         if (ret > 0)
7585                 ret = -EIO;
7586         if (ret < 0)
7587                 goto out;
7588
7589         ret = btrfs_del_item(trans, root, path);
7590 out:
7591         btrfs_free_path(path);
7592         return ret;
7593 }