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         /* Tell the block device(s) that the sectors can be discarded */
1588         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1589                               bytenr, &map_length, &multi, 0);
1590         if (!ret) {
1591                 struct btrfs_bio_stripe *stripe = multi->stripes;
1592                 int i;
1593
1594                 if (map_length > num_bytes)
1595                         map_length = num_bytes;
1596
1597                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1598                         btrfs_issue_discard(stripe->dev->bdev,
1599                                             stripe->physical,
1600                                             map_length);
1601                 }
1602                 kfree(multi);
1603         }
1604
1605         return ret;
1606 #else
1607         return 0;
1608 #endif
1609 }
1610
1611 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1612                          struct btrfs_root *root,
1613                          u64 bytenr, u64 num_bytes, u64 parent,
1614                          u64 root_objectid, u64 owner, u64 offset)
1615 {
1616         int ret;
1617         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1618                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1619
1620         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1621                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1622                                         parent, root_objectid, (int)owner,
1623                                         BTRFS_ADD_DELAYED_REF, NULL);
1624         } else {
1625                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1626                                         parent, root_objectid, owner, offset,
1627                                         BTRFS_ADD_DELAYED_REF, NULL);
1628         }
1629         return ret;
1630 }
1631
1632 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1633                                   struct btrfs_root *root,
1634                                   u64 bytenr, u64 num_bytes,
1635                                   u64 parent, u64 root_objectid,
1636                                   u64 owner, u64 offset, int refs_to_add,
1637                                   struct btrfs_delayed_extent_op *extent_op)
1638 {
1639         struct btrfs_path *path;
1640         struct extent_buffer *leaf;
1641         struct btrfs_extent_item *item;
1642         u64 refs;
1643         int ret;
1644         int err = 0;
1645
1646         path = btrfs_alloc_path();
1647         if (!path)
1648                 return -ENOMEM;
1649
1650         path->reada = 1;
1651         path->leave_spinning = 1;
1652         /* this will setup the path even if it fails to insert the back ref */
1653         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1654                                            path, bytenr, num_bytes, parent,
1655                                            root_objectid, owner, offset,
1656                                            refs_to_add, extent_op);
1657         if (ret == 0)
1658                 goto out;
1659
1660         if (ret != -EAGAIN) {
1661                 err = ret;
1662                 goto out;
1663         }
1664
1665         leaf = path->nodes[0];
1666         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1667         refs = btrfs_extent_refs(leaf, item);
1668         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1669         if (extent_op)
1670                 __run_delayed_extent_op(extent_op, leaf, item);
1671
1672         btrfs_mark_buffer_dirty(leaf);
1673         btrfs_release_path(root->fs_info->extent_root, path);
1674
1675         path->reada = 1;
1676         path->leave_spinning = 1;
1677
1678         /* now insert the actual backref */
1679         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1680                                     path, bytenr, parent, root_objectid,
1681                                     owner, offset, refs_to_add);
1682         BUG_ON(ret);
1683 out:
1684         btrfs_free_path(path);
1685         return err;
1686 }
1687
1688 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1689                                 struct btrfs_root *root,
1690                                 struct btrfs_delayed_ref_node *node,
1691                                 struct btrfs_delayed_extent_op *extent_op,
1692                                 int insert_reserved)
1693 {
1694         int ret = 0;
1695         struct btrfs_delayed_data_ref *ref;
1696         struct btrfs_key ins;
1697         u64 parent = 0;
1698         u64 ref_root = 0;
1699         u64 flags = 0;
1700
1701         ins.objectid = node->bytenr;
1702         ins.offset = node->num_bytes;
1703         ins.type = BTRFS_EXTENT_ITEM_KEY;
1704
1705         ref = btrfs_delayed_node_to_data_ref(node);
1706         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1707                 parent = ref->parent;
1708         else
1709                 ref_root = ref->root;
1710
1711         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1712                 if (extent_op) {
1713                         BUG_ON(extent_op->update_key);
1714                         flags |= extent_op->flags_to_set;
1715                 }
1716                 ret = alloc_reserved_file_extent(trans, root,
1717                                                  parent, ref_root, flags,
1718                                                  ref->objectid, ref->offset,
1719                                                  &ins, node->ref_mod);
1720         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1721                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1722                                              node->num_bytes, parent,
1723                                              ref_root, ref->objectid,
1724                                              ref->offset, node->ref_mod,
1725                                              extent_op);
1726         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1727                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1728                                           node->num_bytes, parent,
1729                                           ref_root, ref->objectid,
1730                                           ref->offset, node->ref_mod,
1731                                           extent_op);
1732         } else {
1733                 BUG();
1734         }
1735         return ret;
1736 }
1737
1738 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1739                                     struct extent_buffer *leaf,
1740                                     struct btrfs_extent_item *ei)
1741 {
1742         u64 flags = btrfs_extent_flags(leaf, ei);
1743         if (extent_op->update_flags) {
1744                 flags |= extent_op->flags_to_set;
1745                 btrfs_set_extent_flags(leaf, ei, flags);
1746         }
1747
1748         if (extent_op->update_key) {
1749                 struct btrfs_tree_block_info *bi;
1750                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1751                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1752                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1753         }
1754 }
1755
1756 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1757                                  struct btrfs_root *root,
1758                                  struct btrfs_delayed_ref_node *node,
1759                                  struct btrfs_delayed_extent_op *extent_op)
1760 {
1761         struct btrfs_key key;
1762         struct btrfs_path *path;
1763         struct btrfs_extent_item *ei;
1764         struct extent_buffer *leaf;
1765         u32 item_size;
1766         int ret;
1767         int err = 0;
1768
1769         path = btrfs_alloc_path();
1770         if (!path)
1771                 return -ENOMEM;
1772
1773         key.objectid = node->bytenr;
1774         key.type = BTRFS_EXTENT_ITEM_KEY;
1775         key.offset = node->num_bytes;
1776
1777         path->reada = 1;
1778         path->leave_spinning = 1;
1779         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1780                                 path, 0, 1);
1781         if (ret < 0) {
1782                 err = ret;
1783                 goto out;
1784         }
1785         if (ret > 0) {
1786                 err = -EIO;
1787                 goto out;
1788         }
1789
1790         leaf = path->nodes[0];
1791         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1792 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1793         if (item_size < sizeof(*ei)) {
1794                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1795                                              path, (u64)-1, 0);
1796                 if (ret < 0) {
1797                         err = ret;
1798                         goto out;
1799                 }
1800                 leaf = path->nodes[0];
1801                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1802         }
1803 #endif
1804         BUG_ON(item_size < sizeof(*ei));
1805         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1806         __run_delayed_extent_op(extent_op, leaf, ei);
1807
1808         btrfs_mark_buffer_dirty(leaf);
1809 out:
1810         btrfs_free_path(path);
1811         return err;
1812 }
1813
1814 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1815                                 struct btrfs_root *root,
1816                                 struct btrfs_delayed_ref_node *node,
1817                                 struct btrfs_delayed_extent_op *extent_op,
1818                                 int insert_reserved)
1819 {
1820         int ret = 0;
1821         struct btrfs_delayed_tree_ref *ref;
1822         struct btrfs_key ins;
1823         u64 parent = 0;
1824         u64 ref_root = 0;
1825
1826         ins.objectid = node->bytenr;
1827         ins.offset = node->num_bytes;
1828         ins.type = BTRFS_EXTENT_ITEM_KEY;
1829
1830         ref = btrfs_delayed_node_to_tree_ref(node);
1831         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1832                 parent = ref->parent;
1833         else
1834                 ref_root = ref->root;
1835
1836         BUG_ON(node->ref_mod != 1);
1837         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1838                 BUG_ON(!extent_op || !extent_op->update_flags ||
1839                        !extent_op->update_key);
1840                 ret = alloc_reserved_tree_block(trans, root,
1841                                                 parent, ref_root,
1842                                                 extent_op->flags_to_set,
1843                                                 &extent_op->key,
1844                                                 ref->level, &ins);
1845         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1846                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1847                                              node->num_bytes, parent, ref_root,
1848                                              ref->level, 0, 1, extent_op);
1849         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1850                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1851                                           node->num_bytes, parent, ref_root,
1852                                           ref->level, 0, 1, extent_op);
1853         } else {
1854                 BUG();
1855         }
1856         return ret;
1857 }
1858
1859
1860 /* helper function to actually process a single delayed ref entry */
1861 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1862                                struct btrfs_root *root,
1863                                struct btrfs_delayed_ref_node *node,
1864                                struct btrfs_delayed_extent_op *extent_op,
1865                                int insert_reserved)
1866 {
1867         int ret;
1868         if (btrfs_delayed_ref_is_head(node)) {
1869                 struct btrfs_delayed_ref_head *head;
1870                 /*
1871                  * we've hit the end of the chain and we were supposed
1872                  * to insert this extent into the tree.  But, it got
1873                  * deleted before we ever needed to insert it, so all
1874                  * we have to do is clean up the accounting
1875                  */
1876                 BUG_ON(extent_op);
1877                 head = btrfs_delayed_node_to_head(node);
1878                 if (insert_reserved) {
1879                         int mark_free = 0;
1880                         struct extent_buffer *must_clean = NULL;
1881
1882                         ret = pin_down_bytes(trans, root, NULL,
1883                                              node->bytenr, node->num_bytes,
1884                                              head->is_data, 1, &must_clean);
1885                         if (ret > 0)
1886                                 mark_free = 1;
1887
1888                         if (must_clean) {
1889                                 clean_tree_block(NULL, root, must_clean);
1890                                 btrfs_tree_unlock(must_clean);
1891                                 free_extent_buffer(must_clean);
1892                         }
1893                         if (head->is_data) {
1894                                 ret = btrfs_del_csums(trans, root,
1895                                                       node->bytenr,
1896                                                       node->num_bytes);
1897                                 BUG_ON(ret);
1898                         }
1899                         if (mark_free) {
1900                                 ret = btrfs_free_reserved_extent(root,
1901                                                         node->bytenr,
1902                                                         node->num_bytes);
1903                                 BUG_ON(ret);
1904                         }
1905                 }
1906                 mutex_unlock(&head->mutex);
1907                 return 0;
1908         }
1909
1910         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1911             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1912                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1913                                            insert_reserved);
1914         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1915                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1916                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1917                                            insert_reserved);
1918         else
1919                 BUG();
1920         return ret;
1921 }
1922
1923 static noinline struct btrfs_delayed_ref_node *
1924 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1925 {
1926         struct rb_node *node;
1927         struct btrfs_delayed_ref_node *ref;
1928         int action = BTRFS_ADD_DELAYED_REF;
1929 again:
1930         /*
1931          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1932          * this prevents ref count from going down to zero when
1933          * there still are pending delayed ref.
1934          */
1935         node = rb_prev(&head->node.rb_node);
1936         while (1) {
1937                 if (!node)
1938                         break;
1939                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1940                                 rb_node);
1941                 if (ref->bytenr != head->node.bytenr)
1942                         break;
1943                 if (ref->action == action)
1944                         return ref;
1945                 node = rb_prev(node);
1946         }
1947         if (action == BTRFS_ADD_DELAYED_REF) {
1948                 action = BTRFS_DROP_DELAYED_REF;
1949                 goto again;
1950         }
1951         return NULL;
1952 }
1953
1954 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1955                                        struct btrfs_root *root,
1956                                        struct list_head *cluster)
1957 {
1958         struct btrfs_delayed_ref_root *delayed_refs;
1959         struct btrfs_delayed_ref_node *ref;
1960         struct btrfs_delayed_ref_head *locked_ref = NULL;
1961         struct btrfs_delayed_extent_op *extent_op;
1962         int ret;
1963         int count = 0;
1964         int must_insert_reserved = 0;
1965
1966         delayed_refs = &trans->transaction->delayed_refs;
1967         while (1) {
1968                 if (!locked_ref) {
1969                         /* pick a new head ref from the cluster list */
1970                         if (list_empty(cluster))
1971                                 break;
1972
1973                         locked_ref = list_entry(cluster->next,
1974                                      struct btrfs_delayed_ref_head, cluster);
1975
1976                         /* grab the lock that says we are going to process
1977                          * all the refs for this head */
1978                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
1979
1980                         /*
1981                          * we may have dropped the spin lock to get the head
1982                          * mutex lock, and that might have given someone else
1983                          * time to free the head.  If that's true, it has been
1984                          * removed from our list and we can move on.
1985                          */
1986                         if (ret == -EAGAIN) {
1987                                 locked_ref = NULL;
1988                                 count++;
1989                                 continue;
1990                         }
1991                 }
1992
1993                 /*
1994                  * record the must insert reserved flag before we
1995                  * drop the spin lock.
1996                  */
1997                 must_insert_reserved = locked_ref->must_insert_reserved;
1998                 locked_ref->must_insert_reserved = 0;
1999
2000                 extent_op = locked_ref->extent_op;
2001                 locked_ref->extent_op = NULL;
2002
2003                 /*
2004                  * locked_ref is the head node, so we have to go one
2005                  * node back for any delayed ref updates
2006                  */
2007                 ref = select_delayed_ref(locked_ref);
2008                 if (!ref) {
2009                         /* All delayed refs have been processed, Go ahead
2010                          * and send the head node to run_one_delayed_ref,
2011                          * so that any accounting fixes can happen
2012                          */
2013                         ref = &locked_ref->node;
2014
2015                         if (extent_op && must_insert_reserved) {
2016                                 kfree(extent_op);
2017                                 extent_op = NULL;
2018                         }
2019
2020                         if (extent_op) {
2021                                 spin_unlock(&delayed_refs->lock);
2022
2023                                 ret = run_delayed_extent_op(trans, root,
2024                                                             ref, extent_op);
2025                                 BUG_ON(ret);
2026                                 kfree(extent_op);
2027
2028                                 cond_resched();
2029                                 spin_lock(&delayed_refs->lock);
2030                                 continue;
2031                         }
2032
2033                         list_del_init(&locked_ref->cluster);
2034                         locked_ref = NULL;
2035                 }
2036
2037                 ref->in_tree = 0;
2038                 rb_erase(&ref->rb_node, &delayed_refs->root);
2039                 delayed_refs->num_entries--;
2040
2041                 spin_unlock(&delayed_refs->lock);
2042
2043                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2044                                           must_insert_reserved);
2045                 BUG_ON(ret);
2046
2047                 btrfs_put_delayed_ref(ref);
2048                 kfree(extent_op);
2049                 count++;
2050
2051                 cond_resched();
2052                 spin_lock(&delayed_refs->lock);
2053         }
2054         return count;
2055 }
2056
2057 /*
2058  * this starts processing the delayed reference count updates and
2059  * extent insertions we have queued up so far.  count can be
2060  * 0, which means to process everything in the tree at the start
2061  * of the run (but not newly added entries), or it can be some target
2062  * number you'd like to process.
2063  */
2064 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2065                            struct btrfs_root *root, unsigned long count)
2066 {
2067         struct rb_node *node;
2068         struct btrfs_delayed_ref_root *delayed_refs;
2069         struct btrfs_delayed_ref_node *ref;
2070         struct list_head cluster;
2071         int ret;
2072         int run_all = count == (unsigned long)-1;
2073         int run_most = 0;
2074
2075         if (root == root->fs_info->extent_root)
2076                 root = root->fs_info->tree_root;
2077
2078         delayed_refs = &trans->transaction->delayed_refs;
2079         INIT_LIST_HEAD(&cluster);
2080 again:
2081         spin_lock(&delayed_refs->lock);
2082         if (count == 0) {
2083                 count = delayed_refs->num_entries * 2;
2084                 run_most = 1;
2085         }
2086         while (1) {
2087                 if (!(run_all || run_most) &&
2088                     delayed_refs->num_heads_ready < 64)
2089                         break;
2090
2091                 /*
2092                  * go find something we can process in the rbtree.  We start at
2093                  * the beginning of the tree, and then build a cluster
2094                  * of refs to process starting at the first one we are able to
2095                  * lock
2096                  */
2097                 ret = btrfs_find_ref_cluster(trans, &cluster,
2098                                              delayed_refs->run_delayed_start);
2099                 if (ret)
2100                         break;
2101
2102                 ret = run_clustered_refs(trans, root, &cluster);
2103                 BUG_ON(ret < 0);
2104
2105                 count -= min_t(unsigned long, ret, count);
2106
2107                 if (count == 0)
2108                         break;
2109         }
2110
2111         if (run_all) {
2112                 node = rb_first(&delayed_refs->root);
2113                 if (!node)
2114                         goto out;
2115                 count = (unsigned long)-1;
2116
2117                 while (node) {
2118                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2119                                        rb_node);
2120                         if (btrfs_delayed_ref_is_head(ref)) {
2121                                 struct btrfs_delayed_ref_head *head;
2122
2123                                 head = btrfs_delayed_node_to_head(ref);
2124                                 atomic_inc(&ref->refs);
2125
2126                                 spin_unlock(&delayed_refs->lock);
2127                                 mutex_lock(&head->mutex);
2128                                 mutex_unlock(&head->mutex);
2129
2130                                 btrfs_put_delayed_ref(ref);
2131                                 cond_resched();
2132                                 goto again;
2133                         }
2134                         node = rb_next(node);
2135                 }
2136                 spin_unlock(&delayed_refs->lock);
2137                 schedule_timeout(1);
2138                 goto again;
2139         }
2140 out:
2141         spin_unlock(&delayed_refs->lock);
2142         return 0;
2143 }
2144
2145 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2146                                 struct btrfs_root *root,
2147                                 u64 bytenr, u64 num_bytes, u64 flags,
2148                                 int is_data)
2149 {
2150         struct btrfs_delayed_extent_op *extent_op;
2151         int ret;
2152
2153         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2154         if (!extent_op)
2155                 return -ENOMEM;
2156
2157         extent_op->flags_to_set = flags;
2158         extent_op->update_flags = 1;
2159         extent_op->update_key = 0;
2160         extent_op->is_data = is_data ? 1 : 0;
2161
2162         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2163         if (ret)
2164                 kfree(extent_op);
2165         return ret;
2166 }
2167
2168 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2169                                       struct btrfs_root *root,
2170                                       struct btrfs_path *path,
2171                                       u64 objectid, u64 offset, u64 bytenr)
2172 {
2173         struct btrfs_delayed_ref_head *head;
2174         struct btrfs_delayed_ref_node *ref;
2175         struct btrfs_delayed_data_ref *data_ref;
2176         struct btrfs_delayed_ref_root *delayed_refs;
2177         struct rb_node *node;
2178         int ret = 0;
2179
2180         ret = -ENOENT;
2181         delayed_refs = &trans->transaction->delayed_refs;
2182         spin_lock(&delayed_refs->lock);
2183         head = btrfs_find_delayed_ref_head(trans, bytenr);
2184         if (!head)
2185                 goto out;
2186
2187         if (!mutex_trylock(&head->mutex)) {
2188                 atomic_inc(&head->node.refs);
2189                 spin_unlock(&delayed_refs->lock);
2190
2191                 btrfs_release_path(root->fs_info->extent_root, path);
2192
2193                 mutex_lock(&head->mutex);
2194                 mutex_unlock(&head->mutex);
2195                 btrfs_put_delayed_ref(&head->node);
2196                 return -EAGAIN;
2197         }
2198
2199         node = rb_prev(&head->node.rb_node);
2200         if (!node)
2201                 goto out_unlock;
2202
2203         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2204
2205         if (ref->bytenr != bytenr)
2206                 goto out_unlock;
2207
2208         ret = 1;
2209         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2210                 goto out_unlock;
2211
2212         data_ref = btrfs_delayed_node_to_data_ref(ref);
2213
2214         node = rb_prev(node);
2215         if (node) {
2216                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2217                 if (ref->bytenr == bytenr)
2218                         goto out_unlock;
2219         }
2220
2221         if (data_ref->root != root->root_key.objectid ||
2222             data_ref->objectid != objectid || data_ref->offset != offset)
2223                 goto out_unlock;
2224
2225         ret = 0;
2226 out_unlock:
2227         mutex_unlock(&head->mutex);
2228 out:
2229         spin_unlock(&delayed_refs->lock);
2230         return ret;
2231 }
2232
2233 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2234                                         struct btrfs_root *root,
2235                                         struct btrfs_path *path,
2236                                         u64 objectid, u64 offset, u64 bytenr)
2237 {
2238         struct btrfs_root *extent_root = root->fs_info->extent_root;
2239         struct extent_buffer *leaf;
2240         struct btrfs_extent_data_ref *ref;
2241         struct btrfs_extent_inline_ref *iref;
2242         struct btrfs_extent_item *ei;
2243         struct btrfs_key key;
2244         u32 item_size;
2245         int ret;
2246
2247         key.objectid = bytenr;
2248         key.offset = (u64)-1;
2249         key.type = BTRFS_EXTENT_ITEM_KEY;
2250
2251         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2252         if (ret < 0)
2253                 goto out;
2254         BUG_ON(ret == 0);
2255
2256         ret = -ENOENT;
2257         if (path->slots[0] == 0)
2258                 goto out;
2259
2260         path->slots[0]--;
2261         leaf = path->nodes[0];
2262         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2263
2264         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2265                 goto out;
2266
2267         ret = 1;
2268         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2269 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2270         if (item_size < sizeof(*ei)) {
2271                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2272                 goto out;
2273         }
2274 #endif
2275         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2276
2277         if (item_size != sizeof(*ei) +
2278             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2279                 goto out;
2280
2281         if (btrfs_extent_generation(leaf, ei) <=
2282             btrfs_root_last_snapshot(&root->root_item))
2283                 goto out;
2284
2285         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2286         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2287             BTRFS_EXTENT_DATA_REF_KEY)
2288                 goto out;
2289
2290         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2291         if (btrfs_extent_refs(leaf, ei) !=
2292             btrfs_extent_data_ref_count(leaf, ref) ||
2293             btrfs_extent_data_ref_root(leaf, ref) !=
2294             root->root_key.objectid ||
2295             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2296             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2297                 goto out;
2298
2299         ret = 0;
2300 out:
2301         return ret;
2302 }
2303
2304 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2305                           struct btrfs_root *root,
2306                           u64 objectid, u64 offset, u64 bytenr)
2307 {
2308         struct btrfs_path *path;
2309         int ret;
2310         int ret2;
2311
2312         path = btrfs_alloc_path();
2313         if (!path)
2314                 return -ENOENT;
2315
2316         do {
2317                 ret = check_committed_ref(trans, root, path, objectid,
2318                                           offset, bytenr);
2319                 if (ret && ret != -ENOENT)
2320                         goto out;
2321
2322                 ret2 = check_delayed_ref(trans, root, path, objectid,
2323                                          offset, bytenr);
2324         } while (ret2 == -EAGAIN);
2325
2326         if (ret2 && ret2 != -ENOENT) {
2327                 ret = ret2;
2328                 goto out;
2329         }
2330
2331         if (ret != -ENOENT || ret2 != -ENOENT)
2332                 ret = 0;
2333 out:
2334         btrfs_free_path(path);
2335         return ret;
2336 }
2337
2338 #if 0
2339 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2340                     struct extent_buffer *buf, u32 nr_extents)
2341 {
2342         struct btrfs_key key;
2343         struct btrfs_file_extent_item *fi;
2344         u64 root_gen;
2345         u32 nritems;
2346         int i;
2347         int level;
2348         int ret = 0;
2349         int shared = 0;
2350
2351         if (!root->ref_cows)
2352                 return 0;
2353
2354         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2355                 shared = 0;
2356                 root_gen = root->root_key.offset;
2357         } else {
2358                 shared = 1;
2359                 root_gen = trans->transid - 1;
2360         }
2361
2362         level = btrfs_header_level(buf);
2363         nritems = btrfs_header_nritems(buf);
2364
2365         if (level == 0) {
2366                 struct btrfs_leaf_ref *ref;
2367                 struct btrfs_extent_info *info;
2368
2369                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2370                 if (!ref) {
2371                         ret = -ENOMEM;
2372                         goto out;
2373                 }
2374
2375                 ref->root_gen = root_gen;
2376                 ref->bytenr = buf->start;
2377                 ref->owner = btrfs_header_owner(buf);
2378                 ref->generation = btrfs_header_generation(buf);
2379                 ref->nritems = nr_extents;
2380                 info = ref->extents;
2381
2382                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2383                         u64 disk_bytenr;
2384                         btrfs_item_key_to_cpu(buf, &key, i);
2385                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2386                                 continue;
2387                         fi = btrfs_item_ptr(buf, i,
2388                                             struct btrfs_file_extent_item);
2389                         if (btrfs_file_extent_type(buf, fi) ==
2390                             BTRFS_FILE_EXTENT_INLINE)
2391                                 continue;
2392                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2393                         if (disk_bytenr == 0)
2394                                 continue;
2395
2396                         info->bytenr = disk_bytenr;
2397                         info->num_bytes =
2398                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2399                         info->objectid = key.objectid;
2400                         info->offset = key.offset;
2401                         info++;
2402                 }
2403
2404                 ret = btrfs_add_leaf_ref(root, ref, shared);
2405                 if (ret == -EEXIST && shared) {
2406                         struct btrfs_leaf_ref *old;
2407                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2408                         BUG_ON(!old);
2409                         btrfs_remove_leaf_ref(root, old);
2410                         btrfs_free_leaf_ref(root, old);
2411                         ret = btrfs_add_leaf_ref(root, ref, shared);
2412                 }
2413                 WARN_ON(ret);
2414                 btrfs_free_leaf_ref(root, ref);
2415         }
2416 out:
2417         return ret;
2418 }
2419
2420 /* when a block goes through cow, we update the reference counts of
2421  * everything that block points to.  The internal pointers of the block
2422  * can be in just about any order, and it is likely to have clusters of
2423  * things that are close together and clusters of things that are not.
2424  *
2425  * To help reduce the seeks that come with updating all of these reference
2426  * counts, sort them by byte number before actual updates are done.
2427  *
2428  * struct refsort is used to match byte number to slot in the btree block.
2429  * we sort based on the byte number and then use the slot to actually
2430  * find the item.
2431  *
2432  * struct refsort is smaller than strcut btrfs_item and smaller than
2433  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2434  * for a btree block, there's no way for a kmalloc of refsorts for a
2435  * single node to be bigger than a page.
2436  */
2437 struct refsort {
2438         u64 bytenr;
2439         u32 slot;
2440 };
2441
2442 /*
2443  * for passing into sort()
2444  */
2445 static int refsort_cmp(const void *a_void, const void *b_void)
2446 {
2447         const struct refsort *a = a_void;
2448         const struct refsort *b = b_void;
2449
2450         if (a->bytenr < b->bytenr)
2451                 return -1;
2452         if (a->bytenr > b->bytenr)
2453                 return 1;
2454         return 0;
2455 }
2456 #endif
2457
2458 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2459                            struct btrfs_root *root,
2460                            struct extent_buffer *buf,
2461                            int full_backref, int inc)
2462 {
2463         u64 bytenr;
2464         u64 num_bytes;
2465         u64 parent;
2466         u64 ref_root;
2467         u32 nritems;
2468         struct btrfs_key key;
2469         struct btrfs_file_extent_item *fi;
2470         int i;
2471         int level;
2472         int ret = 0;
2473         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2474                             u64, u64, u64, u64, u64, u64);
2475
2476         ref_root = btrfs_header_owner(buf);
2477         nritems = btrfs_header_nritems(buf);
2478         level = btrfs_header_level(buf);
2479
2480         if (!root->ref_cows && level == 0)
2481                 return 0;
2482
2483         if (inc)
2484                 process_func = btrfs_inc_extent_ref;
2485         else
2486                 process_func = btrfs_free_extent;
2487
2488         if (full_backref)
2489                 parent = buf->start;
2490         else
2491                 parent = 0;
2492
2493         for (i = 0; i < nritems; i++) {
2494                 if (level == 0) {
2495                         btrfs_item_key_to_cpu(buf, &key, i);
2496                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2497                                 continue;
2498                         fi = btrfs_item_ptr(buf, i,
2499                                             struct btrfs_file_extent_item);
2500                         if (btrfs_file_extent_type(buf, fi) ==
2501                             BTRFS_FILE_EXTENT_INLINE)
2502                                 continue;
2503                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2504                         if (bytenr == 0)
2505                                 continue;
2506
2507                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2508                         key.offset -= btrfs_file_extent_offset(buf, fi);
2509                         ret = process_func(trans, root, bytenr, num_bytes,
2510                                            parent, ref_root, key.objectid,
2511                                            key.offset);
2512                         if (ret)
2513                                 goto fail;
2514                 } else {
2515                         bytenr = btrfs_node_blockptr(buf, i);
2516                         num_bytes = btrfs_level_size(root, level - 1);
2517                         ret = process_func(trans, root, bytenr, num_bytes,
2518                                            parent, ref_root, level - 1, 0);
2519                         if (ret)
2520                                 goto fail;
2521                 }
2522         }
2523         return 0;
2524 fail:
2525         BUG();
2526         return ret;
2527 }
2528
2529 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2530                   struct extent_buffer *buf, int full_backref)
2531 {
2532         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2533 }
2534
2535 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2536                   struct extent_buffer *buf, int full_backref)
2537 {
2538         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2539 }
2540
2541 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2542                                  struct btrfs_root *root,
2543                                  struct btrfs_path *path,
2544                                  struct btrfs_block_group_cache *cache)
2545 {
2546         int ret;
2547         struct btrfs_root *extent_root = root->fs_info->extent_root;
2548         unsigned long bi;
2549         struct extent_buffer *leaf;
2550
2551         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2552         if (ret < 0)
2553                 goto fail;
2554         BUG_ON(ret);
2555
2556         leaf = path->nodes[0];
2557         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2558         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2559         btrfs_mark_buffer_dirty(leaf);
2560         btrfs_release_path(extent_root, path);
2561 fail:
2562         if (ret)
2563                 return ret;
2564         return 0;
2565
2566 }
2567
2568 static struct btrfs_block_group_cache *
2569 next_block_group(struct btrfs_root *root,
2570                  struct btrfs_block_group_cache *cache)
2571 {
2572         struct rb_node *node;
2573         spin_lock(&root->fs_info->block_group_cache_lock);
2574         node = rb_next(&cache->cache_node);
2575         btrfs_put_block_group(cache);
2576         if (node) {
2577                 cache = rb_entry(node, struct btrfs_block_group_cache,
2578                                  cache_node);
2579                 atomic_inc(&cache->count);
2580         } else
2581                 cache = NULL;
2582         spin_unlock(&root->fs_info->block_group_cache_lock);
2583         return cache;
2584 }
2585
2586 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2587                                    struct btrfs_root *root)
2588 {
2589         struct btrfs_block_group_cache *cache;
2590         int err = 0;
2591         struct btrfs_path *path;
2592         u64 last = 0;
2593
2594         path = btrfs_alloc_path();
2595         if (!path)
2596                 return -ENOMEM;
2597
2598         while (1) {
2599                 if (last == 0) {
2600                         err = btrfs_run_delayed_refs(trans, root,
2601                                                      (unsigned long)-1);
2602                         BUG_ON(err);
2603                 }
2604
2605                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2606                 while (cache) {
2607                         if (cache->dirty)
2608                                 break;
2609                         cache = next_block_group(root, cache);
2610                 }
2611                 if (!cache) {
2612                         if (last == 0)
2613                                 break;
2614                         last = 0;
2615                         continue;
2616                 }
2617
2618                 cache->dirty = 0;
2619                 last = cache->key.objectid + cache->key.offset;
2620
2621                 err = write_one_cache_group(trans, root, path, cache);
2622                 BUG_ON(err);
2623                 btrfs_put_block_group(cache);
2624         }
2625
2626         btrfs_free_path(path);
2627         return 0;
2628 }
2629
2630 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2631 {
2632         struct btrfs_block_group_cache *block_group;
2633         int readonly = 0;
2634
2635         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2636         if (!block_group || block_group->ro)
2637                 readonly = 1;
2638         if (block_group)
2639                 btrfs_put_block_group(block_group);
2640         return readonly;
2641 }
2642
2643 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2644                              u64 total_bytes, u64 bytes_used,
2645                              struct btrfs_space_info **space_info)
2646 {
2647         struct btrfs_space_info *found;
2648
2649         found = __find_space_info(info, flags);
2650         if (found) {
2651                 spin_lock(&found->lock);
2652                 found->total_bytes += total_bytes;
2653                 found->bytes_used += bytes_used;
2654                 found->full = 0;
2655                 spin_unlock(&found->lock);
2656                 *space_info = found;
2657                 return 0;
2658         }
2659         found = kzalloc(sizeof(*found), GFP_NOFS);
2660         if (!found)
2661                 return -ENOMEM;
2662
2663         INIT_LIST_HEAD(&found->block_groups);
2664         init_rwsem(&found->groups_sem);
2665         spin_lock_init(&found->lock);
2666         found->flags = flags;
2667         found->total_bytes = total_bytes;
2668         found->bytes_used = bytes_used;
2669         found->bytes_pinned = 0;
2670         found->bytes_reserved = 0;
2671         found->bytes_readonly = 0;
2672         found->bytes_delalloc = 0;
2673         found->full = 0;
2674         found->force_alloc = 0;
2675         *space_info = found;
2676         list_add_rcu(&found->list, &info->space_info);
2677         atomic_set(&found->caching_threads, 0);
2678         return 0;
2679 }
2680
2681 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2682 {
2683         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2684                                    BTRFS_BLOCK_GROUP_RAID1 |
2685                                    BTRFS_BLOCK_GROUP_RAID10 |
2686                                    BTRFS_BLOCK_GROUP_DUP);
2687         if (extra_flags) {
2688                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2689                         fs_info->avail_data_alloc_bits |= extra_flags;
2690                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2691                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2692                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2693                         fs_info->avail_system_alloc_bits |= extra_flags;
2694         }
2695 }
2696
2697 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2698 {
2699         spin_lock(&cache->space_info->lock);
2700         spin_lock(&cache->lock);
2701         if (!cache->ro) {
2702                 cache->space_info->bytes_readonly += cache->key.offset -
2703                                         btrfs_block_group_used(&cache->item);
2704                 cache->ro = 1;
2705         }
2706         spin_unlock(&cache->lock);
2707         spin_unlock(&cache->space_info->lock);
2708 }
2709
2710 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2711 {
2712         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2713
2714         if (num_devices == 1)
2715                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2716         if (num_devices < 4)
2717                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2718
2719         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2720             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2721                       BTRFS_BLOCK_GROUP_RAID10))) {
2722                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2723         }
2724
2725         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2726             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2727                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2728         }
2729
2730         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2731             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2732              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2733              (flags & BTRFS_BLOCK_GROUP_DUP)))
2734                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2735         return flags;
2736 }
2737
2738 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2739 {
2740         struct btrfs_fs_info *info = root->fs_info;
2741         u64 alloc_profile;
2742
2743         if (data) {
2744                 alloc_profile = info->avail_data_alloc_bits &
2745                         info->data_alloc_profile;
2746                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2747         } else if (root == root->fs_info->chunk_root) {
2748                 alloc_profile = info->avail_system_alloc_bits &
2749                         info->system_alloc_profile;
2750                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2751         } else {
2752                 alloc_profile = info->avail_metadata_alloc_bits &
2753                         info->metadata_alloc_profile;
2754                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2755         }
2756
2757         return btrfs_reduce_alloc_profile(root, data);
2758 }
2759
2760 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2761 {
2762         u64 alloc_target;
2763
2764         alloc_target = btrfs_get_alloc_profile(root, 1);
2765         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2766                                                        alloc_target);
2767 }
2768
2769 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2770 {
2771         u64 num_bytes;
2772         int level;
2773
2774         level = BTRFS_MAX_LEVEL - 2;
2775         /*
2776          * NOTE: these calculations are absolutely the worst possible case.
2777          * This assumes that _every_ item we insert will require a new leaf, and
2778          * that the tree has grown to its maximum level size.
2779          */
2780
2781         /*
2782          * for every item we insert we could insert both an extent item and a
2783          * extent ref item.  Then for ever item we insert, we will need to cow
2784          * both the original leaf, plus the leaf to the left and right of it.
2785          *
2786          * Unless we are talking about the extent root, then we just want the
2787          * number of items * 2, since we just need the extent item plus its ref.
2788          */
2789         if (root == root->fs_info->extent_root)
2790                 num_bytes = num_items * 2;
2791         else
2792                 num_bytes = (num_items + (2 * num_items)) * 3;
2793
2794         /*
2795          * num_bytes is total number of leaves we could need times the leaf
2796          * size, and then for every leaf we could end up cow'ing 2 nodes per
2797          * level, down to the leaf level.
2798          */
2799         num_bytes = (num_bytes * root->leafsize) +
2800                 (num_bytes * (level * 2)) * root->nodesize;
2801
2802         return num_bytes;
2803 }
2804
2805 /*
2806  * Unreserve metadata space for delalloc.  If we have less reserved credits than
2807  * we have extents, this function does nothing.
2808  */
2809 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2810                                           struct inode *inode, int num_items)
2811 {
2812         struct btrfs_fs_info *info = root->fs_info;
2813         struct btrfs_space_info *meta_sinfo;
2814         u64 num_bytes;
2815         u64 alloc_target;
2816         bool bug = false;
2817
2818         /* get the space info for where the metadata will live */
2819         alloc_target = btrfs_get_alloc_profile(root, 0);
2820         meta_sinfo = __find_space_info(info, alloc_target);
2821
2822         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2823                                            num_items);
2824
2825         spin_lock(&meta_sinfo->lock);
2826         if (BTRFS_I(inode)->delalloc_reserved_extents <=
2827             BTRFS_I(inode)->delalloc_extents) {
2828                 spin_unlock(&meta_sinfo->lock);
2829                 return 0;
2830         }
2831
2832         BTRFS_I(inode)->delalloc_reserved_extents--;
2833         BUG_ON(BTRFS_I(inode)->delalloc_reserved_extents < 0);
2834
2835         if (meta_sinfo->bytes_delalloc < num_bytes) {
2836                 bug = true;
2837                 meta_sinfo->bytes_delalloc = 0;
2838         } else {
2839                 meta_sinfo->bytes_delalloc -= num_bytes;
2840         }
2841         spin_unlock(&meta_sinfo->lock);
2842
2843         BUG_ON(bug);
2844
2845         return 0;
2846 }
2847
2848 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2849 {
2850         u64 thresh;
2851
2852         thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2853                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2854                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2855                 meta_sinfo->bytes_may_use;
2856
2857         thresh = meta_sinfo->total_bytes - thresh;
2858         thresh *= 80;
2859         do_div(thresh, 100);
2860         if (thresh <= meta_sinfo->bytes_delalloc)
2861                 meta_sinfo->force_delalloc = 1;
2862         else
2863                 meta_sinfo->force_delalloc = 0;
2864 }
2865
2866 static int maybe_allocate_chunk(struct btrfs_root *root,
2867                                  struct btrfs_space_info *info)
2868 {
2869         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
2870         struct btrfs_trans_handle *trans;
2871         bool wait = false;
2872         int ret = 0;
2873         u64 min_metadata;
2874         u64 free_space;
2875
2876         free_space = btrfs_super_total_bytes(disk_super);
2877         /*
2878          * we allow the metadata to grow to a max of either 5gb or 5% of the
2879          * space in the volume.
2880          */
2881         min_metadata = min((u64)5 * 1024 * 1024 * 1024,
2882                              div64_u64(free_space * 5, 100));
2883         if (info->total_bytes >= min_metadata) {
2884                 spin_unlock(&info->lock);
2885                 return 0;
2886         }
2887
2888         if (info->full) {
2889                 spin_unlock(&info->lock);
2890                 return 0;
2891         }
2892
2893         if (!info->allocating_chunk) {
2894                 info->force_alloc = 1;
2895                 info->allocating_chunk = 1;
2896                 init_waitqueue_head(&info->wait);
2897         } else {
2898                 wait = true;
2899         }
2900
2901         spin_unlock(&info->lock);
2902
2903         if (wait) {
2904                 wait_event(info->wait,
2905                            !info->allocating_chunk);
2906                 return 1;
2907         }
2908
2909         trans = btrfs_start_transaction(root, 1);
2910         if (!trans) {
2911                 ret = -ENOMEM;
2912                 goto out;
2913         }
2914
2915         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2916                              4096 + 2 * 1024 * 1024,
2917                              info->flags, 0);
2918         btrfs_end_transaction(trans, root);
2919         if (ret)
2920                 goto out;
2921 out:
2922         spin_lock(&info->lock);
2923         info->allocating_chunk = 0;
2924         spin_unlock(&info->lock);
2925         wake_up(&info->wait);
2926
2927         if (ret)
2928                 return 0;
2929         return 1;
2930 }
2931
2932 /*
2933  * Reserve metadata space for delalloc.
2934  */
2935 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
2936                                         struct inode *inode, int num_items)
2937 {
2938         struct btrfs_fs_info *info = root->fs_info;
2939         struct btrfs_space_info *meta_sinfo;
2940         u64 num_bytes;
2941         u64 used;
2942         u64 alloc_target;
2943         int flushed = 0;
2944         int force_delalloc;
2945
2946         /* get the space info for where the metadata will live */
2947         alloc_target = btrfs_get_alloc_profile(root, 0);
2948         meta_sinfo = __find_space_info(info, alloc_target);
2949
2950         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2951                                            num_items);
2952 again:
2953         spin_lock(&meta_sinfo->lock);
2954
2955         force_delalloc = meta_sinfo->force_delalloc;
2956
2957         if (unlikely(!meta_sinfo->bytes_root))
2958                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
2959
2960         if (!flushed)
2961                 meta_sinfo->bytes_delalloc += num_bytes;
2962
2963         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2964                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2965                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2966                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
2967
2968         if (used > meta_sinfo->total_bytes) {
2969                 flushed++;
2970
2971                 if (flushed == 1) {
2972                         if (maybe_allocate_chunk(root, meta_sinfo))
2973                                 goto again;
2974                         flushed++;
2975                 } else {
2976                         spin_unlock(&meta_sinfo->lock);
2977                 }
2978
2979                 if (flushed == 2) {
2980                         filemap_flush(inode->i_mapping);
2981                         goto again;
2982                 } else if (flushed == 3) {
2983                         btrfs_start_delalloc_inodes(root);
2984                         btrfs_wait_ordered_extents(root, 0);
2985                         goto again;
2986                 }
2987                 spin_lock(&meta_sinfo->lock);
2988                 meta_sinfo->bytes_delalloc -= num_bytes;
2989                 spin_unlock(&meta_sinfo->lock);
2990                 printk(KERN_ERR "enospc, has %d, reserved %d\n",
2991                        BTRFS_I(inode)->delalloc_extents,
2992                        BTRFS_I(inode)->delalloc_reserved_extents);
2993                 dump_space_info(meta_sinfo, 0, 0);
2994                 return -ENOSPC;
2995         }
2996
2997         BTRFS_I(inode)->delalloc_reserved_extents++;
2998         check_force_delalloc(meta_sinfo);
2999         spin_unlock(&meta_sinfo->lock);
3000
3001         if (!flushed && force_delalloc)
3002                 filemap_flush(inode->i_mapping);
3003
3004         return 0;
3005 }
3006
3007 /*
3008  * unreserve num_items number of items worth of metadata space.  This needs to
3009  * be paired with btrfs_reserve_metadata_space.
3010  *
3011  * NOTE: if you have the option, run this _AFTER_ you do a
3012  * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3013  * oprations which will result in more used metadata, so we want to make sure we
3014  * can do that without issue.
3015  */
3016 int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
3017 {
3018         struct btrfs_fs_info *info = root->fs_info;
3019         struct btrfs_space_info *meta_sinfo;
3020         u64 num_bytes;
3021         u64 alloc_target;
3022         bool bug = false;
3023
3024         /* get the space info for where the metadata will live */
3025         alloc_target = btrfs_get_alloc_profile(root, 0);
3026         meta_sinfo = __find_space_info(info, alloc_target);
3027
3028         num_bytes = calculate_bytes_needed(root, num_items);
3029
3030         spin_lock(&meta_sinfo->lock);
3031         if (meta_sinfo->bytes_may_use < num_bytes) {
3032                 bug = true;
3033                 meta_sinfo->bytes_may_use = 0;
3034         } else {
3035                 meta_sinfo->bytes_may_use -= num_bytes;
3036         }
3037         spin_unlock(&meta_sinfo->lock);
3038
3039         BUG_ON(bug);
3040
3041         return 0;
3042 }
3043
3044 /*
3045  * Reserve some metadata space for use.  We'll calculate the worste case number
3046  * of bytes that would be needed to modify num_items number of items.  If we
3047  * have space, fantastic, if not, you get -ENOSPC.  Please call
3048  * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3049  * items you reserved, since whatever metadata you needed should have already
3050  * been allocated.
3051  *
3052  * This will commit the transaction to make more space if we don't have enough
3053  * metadata space.  THe only time we don't do this is if we're reserving space
3054  * inside of a transaction, then we will just return -ENOSPC and it is the
3055  * callers responsibility to handle it properly.
3056  */
3057 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3058 {
3059         struct btrfs_fs_info *info = root->fs_info;
3060         struct btrfs_space_info *meta_sinfo;
3061         u64 num_bytes;
3062         u64 used;
3063         u64 alloc_target;
3064         int retries = 0;
3065
3066         /* get the space info for where the metadata will live */
3067         alloc_target = btrfs_get_alloc_profile(root, 0);
3068         meta_sinfo = __find_space_info(info, alloc_target);
3069
3070         num_bytes = calculate_bytes_needed(root, num_items);
3071 again:
3072         spin_lock(&meta_sinfo->lock);
3073
3074         if (unlikely(!meta_sinfo->bytes_root))
3075                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3076
3077         if (!retries)
3078                 meta_sinfo->bytes_may_use += num_bytes;
3079
3080         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3081                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3082                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3083                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3084
3085         if (used > meta_sinfo->total_bytes) {
3086                 retries++;
3087                 if (retries == 1) {
3088                         if (maybe_allocate_chunk(root, meta_sinfo))
3089                                 goto again;
3090                         retries++;
3091                 } else {
3092                         spin_unlock(&meta_sinfo->lock);
3093                 }
3094
3095                 if (retries == 2) {
3096                         btrfs_start_delalloc_inodes(root);
3097                         btrfs_wait_ordered_extents(root, 0);
3098                         goto again;
3099                 }
3100                 spin_lock(&meta_sinfo->lock);
3101                 meta_sinfo->bytes_may_use -= num_bytes;
3102                 spin_unlock(&meta_sinfo->lock);
3103
3104                 dump_space_info(meta_sinfo, 0, 0);
3105                 return -ENOSPC;
3106         }
3107
3108         check_force_delalloc(meta_sinfo);
3109         spin_unlock(&meta_sinfo->lock);
3110
3111         return 0;
3112 }
3113
3114 /*
3115  * This will check the space that the inode allocates from to make sure we have
3116  * enough space for bytes.
3117  */
3118 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3119                                 u64 bytes)
3120 {
3121         struct btrfs_space_info *data_sinfo;
3122         int ret = 0, committed = 0;
3123
3124         /* make sure bytes are sectorsize aligned */
3125         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3126
3127         data_sinfo = BTRFS_I(inode)->space_info;
3128         if (!data_sinfo)
3129                 goto alloc;
3130
3131 again:
3132         /* make sure we have enough space to handle the data first */
3133         spin_lock(&data_sinfo->lock);
3134         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
3135             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
3136             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
3137             data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
3138                 struct btrfs_trans_handle *trans;
3139
3140                 /*
3141                  * if we don't have enough free bytes in this space then we need
3142                  * to alloc a new chunk.
3143                  */
3144                 if (!data_sinfo->full) {
3145                         u64 alloc_target;
3146
3147                         data_sinfo->force_alloc = 1;
3148                         spin_unlock(&data_sinfo->lock);
3149 alloc:
3150                         alloc_target = btrfs_get_alloc_profile(root, 1);
3151                         trans = btrfs_start_transaction(root, 1);
3152                         if (!trans)
3153                                 return -ENOMEM;
3154
3155                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3156                                              bytes + 2 * 1024 * 1024,
3157                                              alloc_target, 0);
3158                         btrfs_end_transaction(trans, root);
3159                         if (ret)
3160                                 return ret;
3161
3162                         if (!data_sinfo) {
3163                                 btrfs_set_inode_space_info(root, inode);
3164                                 data_sinfo = BTRFS_I(inode)->space_info;
3165                         }
3166                         goto again;
3167                 }
3168                 spin_unlock(&data_sinfo->lock);
3169
3170                 /* commit the current transaction and try again */
3171                 if (!committed) {
3172                         committed = 1;
3173                         trans = btrfs_join_transaction(root, 1);
3174                         if (!trans)
3175                                 return -ENOMEM;
3176                         ret = btrfs_commit_transaction(trans, root);
3177                         if (ret)
3178                                 return ret;
3179                         goto again;
3180                 }
3181
3182                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3183                        ", %llu bytes_used, %llu bytes_reserved, "
3184                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3185                        "%llu total\n", (unsigned long long)bytes,
3186                        (unsigned long long)data_sinfo->bytes_delalloc,
3187                        (unsigned long long)data_sinfo->bytes_used,
3188                        (unsigned long long)data_sinfo->bytes_reserved,
3189                        (unsigned long long)data_sinfo->bytes_pinned,
3190                        (unsigned long long)data_sinfo->bytes_readonly,
3191                        (unsigned long long)data_sinfo->bytes_may_use,
3192                        (unsigned long long)data_sinfo->total_bytes);
3193                 return -ENOSPC;
3194         }
3195         data_sinfo->bytes_may_use += bytes;
3196         BTRFS_I(inode)->reserved_bytes += bytes;
3197         spin_unlock(&data_sinfo->lock);
3198
3199         return 0;
3200 }
3201
3202 /*
3203  * if there was an error for whatever reason after calling
3204  * btrfs_check_data_free_space, call this so we can cleanup the counters.
3205  */
3206 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3207                                     struct inode *inode, u64 bytes)
3208 {
3209         struct btrfs_space_info *data_sinfo;
3210
3211         /* make sure bytes are sectorsize aligned */
3212         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3213
3214         data_sinfo = BTRFS_I(inode)->space_info;
3215         spin_lock(&data_sinfo->lock);
3216         data_sinfo->bytes_may_use -= bytes;
3217         BTRFS_I(inode)->reserved_bytes -= bytes;
3218         spin_unlock(&data_sinfo->lock);
3219 }
3220
3221 /* called when we are adding a delalloc extent to the inode's io_tree */
3222 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3223                                   u64 bytes)
3224 {
3225         struct btrfs_space_info *data_sinfo;
3226
3227         /* get the space info for where this inode will be storing its data */
3228         data_sinfo = BTRFS_I(inode)->space_info;
3229
3230         /* make sure we have enough space to handle the data first */
3231         spin_lock(&data_sinfo->lock);
3232         data_sinfo->bytes_delalloc += bytes;
3233
3234         /*
3235          * we are adding a delalloc extent without calling
3236          * btrfs_check_data_free_space first.  This happens on a weird
3237          * writepage condition, but shouldn't hurt our accounting
3238          */
3239         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3240                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3241                 BTRFS_I(inode)->reserved_bytes = 0;
3242         } else {
3243                 data_sinfo->bytes_may_use -= bytes;
3244                 BTRFS_I(inode)->reserved_bytes -= bytes;
3245         }
3246
3247         spin_unlock(&data_sinfo->lock);
3248 }
3249
3250 /* called when we are clearing an delalloc extent from the inode's io_tree */
3251 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3252                               u64 bytes)
3253 {
3254         struct btrfs_space_info *info;
3255
3256         info = BTRFS_I(inode)->space_info;
3257
3258         spin_lock(&info->lock);
3259         info->bytes_delalloc -= bytes;
3260         spin_unlock(&info->lock);
3261 }
3262
3263 static void force_metadata_allocation(struct btrfs_fs_info *info)
3264 {
3265         struct list_head *head = &info->space_info;
3266         struct btrfs_space_info *found;
3267
3268         rcu_read_lock();
3269         list_for_each_entry_rcu(found, head, list) {
3270                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3271                         found->force_alloc = 1;
3272         }
3273         rcu_read_unlock();
3274 }
3275
3276 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3277                           struct btrfs_root *extent_root, u64 alloc_bytes,
3278                           u64 flags, int force)
3279 {
3280         struct btrfs_space_info *space_info;
3281         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3282         u64 thresh;
3283         int ret = 0;
3284
3285         mutex_lock(&fs_info->chunk_mutex);
3286
3287         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3288
3289         space_info = __find_space_info(extent_root->fs_info, flags);
3290         if (!space_info) {
3291                 ret = update_space_info(extent_root->fs_info, flags,
3292                                         0, 0, &space_info);
3293                 BUG_ON(ret);
3294         }
3295         BUG_ON(!space_info);
3296
3297         spin_lock(&space_info->lock);
3298         if (space_info->force_alloc)
3299                 force = 1;
3300         if (space_info->full) {
3301                 spin_unlock(&space_info->lock);
3302                 goto out;
3303         }
3304
3305         thresh = space_info->total_bytes - space_info->bytes_readonly;
3306         thresh = div_factor(thresh, 8);
3307         if (!force &&
3308            (space_info->bytes_used + space_info->bytes_pinned +
3309             space_info->bytes_reserved + alloc_bytes) < thresh) {
3310                 spin_unlock(&space_info->lock);
3311                 goto out;
3312         }
3313         spin_unlock(&space_info->lock);
3314
3315         /*
3316          * if we're doing a data chunk, go ahead and make sure that
3317          * we keep a reasonable number of metadata chunks allocated in the
3318          * FS as well.
3319          */
3320         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3321                 fs_info->data_chunk_allocations++;
3322                 if (!(fs_info->data_chunk_allocations %
3323                       fs_info->metadata_ratio))
3324                         force_metadata_allocation(fs_info);
3325         }
3326
3327         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3328         spin_lock(&space_info->lock);
3329         if (ret)
3330                 space_info->full = 1;
3331         space_info->force_alloc = 0;
3332         spin_unlock(&space_info->lock);
3333 out:
3334         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3335         return ret;
3336 }
3337
3338 static int update_block_group(struct btrfs_trans_handle *trans,
3339                               struct btrfs_root *root,
3340                               u64 bytenr, u64 num_bytes, int alloc,
3341                               int mark_free)
3342 {
3343         struct btrfs_block_group_cache *cache;
3344         struct btrfs_fs_info *info = root->fs_info;
3345         u64 total = num_bytes;
3346         u64 old_val;
3347         u64 byte_in_group;
3348
3349         /* block accounting for super block */
3350         spin_lock(&info->delalloc_lock);
3351         old_val = btrfs_super_bytes_used(&info->super_copy);
3352         if (alloc)
3353                 old_val += num_bytes;
3354         else
3355                 old_val -= num_bytes;
3356         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3357
3358         /* block accounting for root item */
3359         old_val = btrfs_root_used(&root->root_item);
3360         if (alloc)
3361                 old_val += num_bytes;
3362         else
3363                 old_val -= num_bytes;
3364         btrfs_set_root_used(&root->root_item, old_val);
3365         spin_unlock(&info->delalloc_lock);
3366
3367         while (total) {
3368                 cache = btrfs_lookup_block_group(info, bytenr);
3369                 if (!cache)
3370                         return -1;
3371                 byte_in_group = bytenr - cache->key.objectid;
3372                 WARN_ON(byte_in_group > cache->key.offset);
3373
3374                 spin_lock(&cache->space_info->lock);
3375                 spin_lock(&cache->lock);
3376                 cache->dirty = 1;
3377                 old_val = btrfs_block_group_used(&cache->item);
3378                 num_bytes = min(total, cache->key.offset - byte_in_group);
3379                 if (alloc) {
3380                         old_val += num_bytes;
3381                         btrfs_set_block_group_used(&cache->item, old_val);
3382                         cache->reserved -= num_bytes;
3383                         cache->space_info->bytes_used += num_bytes;
3384                         cache->space_info->bytes_reserved -= num_bytes;
3385                         if (cache->ro)
3386                                 cache->space_info->bytes_readonly -= num_bytes;
3387                         spin_unlock(&cache->lock);
3388                         spin_unlock(&cache->space_info->lock);
3389                 } else {
3390                         old_val -= num_bytes;
3391                         cache->space_info->bytes_used -= num_bytes;
3392                         if (cache->ro)
3393                                 cache->space_info->bytes_readonly += num_bytes;
3394                         btrfs_set_block_group_used(&cache->item, old_val);
3395                         spin_unlock(&cache->lock);
3396                         spin_unlock(&cache->space_info->lock);
3397                         if (mark_free) {
3398                                 int ret;
3399
3400                                 ret = btrfs_discard_extent(root, bytenr,
3401                                                            num_bytes);
3402                                 WARN_ON(ret);
3403
3404                                 ret = btrfs_add_free_space(cache, bytenr,
3405                                                            num_bytes);
3406                                 WARN_ON(ret);
3407                         }
3408                 }
3409                 btrfs_put_block_group(cache);
3410                 total -= num_bytes;
3411                 bytenr += num_bytes;
3412         }
3413         return 0;
3414 }
3415
3416 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3417 {
3418         struct btrfs_block_group_cache *cache;
3419         u64 bytenr;
3420
3421         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3422         if (!cache)
3423                 return 0;
3424
3425         bytenr = cache->key.objectid;
3426         btrfs_put_block_group(cache);
3427
3428         return bytenr;
3429 }
3430
3431 /*
3432  * this function must be called within transaction
3433  */
3434 int btrfs_pin_extent(struct btrfs_root *root,
3435                      u64 bytenr, u64 num_bytes, int reserved)
3436 {
3437         struct btrfs_fs_info *fs_info = root->fs_info;
3438         struct btrfs_block_group_cache *cache;
3439
3440         cache = btrfs_lookup_block_group(fs_info, bytenr);
3441         BUG_ON(!cache);
3442
3443         spin_lock(&cache->space_info->lock);
3444         spin_lock(&cache->lock);
3445         cache->pinned += num_bytes;
3446         cache->space_info->bytes_pinned += num_bytes;
3447         if (reserved) {
3448                 cache->reserved -= num_bytes;
3449                 cache->space_info->bytes_reserved -= num_bytes;
3450         }
3451         spin_unlock(&cache->lock);
3452         spin_unlock(&cache->space_info->lock);
3453
3454         btrfs_put_block_group(cache);
3455
3456         set_extent_dirty(fs_info->pinned_extents,
3457                          bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3458         return 0;
3459 }
3460
3461 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3462                                    u64 num_bytes, int reserve)
3463 {
3464         spin_lock(&cache->space_info->lock);
3465         spin_lock(&cache->lock);
3466         if (reserve) {
3467                 cache->reserved += num_bytes;
3468                 cache->space_info->bytes_reserved += num_bytes;
3469         } else {
3470                 cache->reserved -= num_bytes;
3471                 cache->space_info->bytes_reserved -= num_bytes;
3472         }
3473         spin_unlock(&cache->lock);
3474         spin_unlock(&cache->space_info->lock);
3475         return 0;
3476 }
3477
3478 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3479                                 struct btrfs_root *root)
3480 {
3481         struct btrfs_fs_info *fs_info = root->fs_info;
3482         struct btrfs_caching_control *next;
3483         struct btrfs_caching_control *caching_ctl;
3484         struct btrfs_block_group_cache *cache;
3485
3486         down_write(&fs_info->extent_commit_sem);
3487
3488         list_for_each_entry_safe(caching_ctl, next,
3489                                  &fs_info->caching_block_groups, list) {
3490                 cache = caching_ctl->block_group;
3491                 if (block_group_cache_done(cache)) {
3492                         cache->last_byte_to_unpin = (u64)-1;
3493                         list_del_init(&caching_ctl->list);
3494                         put_caching_control(caching_ctl);
3495                 } else {
3496                         cache->last_byte_to_unpin = caching_ctl->progress;
3497                 }
3498         }
3499
3500         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3501                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3502         else
3503                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3504
3505         up_write(&fs_info->extent_commit_sem);
3506         return 0;
3507 }
3508
3509 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3510 {
3511         struct btrfs_fs_info *fs_info = root->fs_info;
3512         struct btrfs_block_group_cache *cache = NULL;
3513         u64 len;
3514
3515         while (start <= end) {
3516                 if (!cache ||
3517                     start >= cache->key.objectid + cache->key.offset) {
3518                         if (cache)
3519                                 btrfs_put_block_group(cache);
3520                         cache = btrfs_lookup_block_group(fs_info, start);
3521                         BUG_ON(!cache);
3522                 }
3523
3524                 len = cache->key.objectid + cache->key.offset - start;
3525                 len = min(len, end + 1 - start);
3526
3527                 if (start < cache->last_byte_to_unpin) {
3528                         len = min(len, cache->last_byte_to_unpin - start);
3529                         btrfs_add_free_space(cache, start, len);
3530                 }
3531
3532                 spin_lock(&cache->space_info->lock);
3533                 spin_lock(&cache->lock);
3534                 cache->pinned -= len;
3535                 cache->space_info->bytes_pinned -= len;
3536                 spin_unlock(&cache->lock);
3537                 spin_unlock(&cache->space_info->lock);
3538
3539                 start += len;
3540         }
3541
3542         if (cache)
3543                 btrfs_put_block_group(cache);
3544         return 0;
3545 }
3546
3547 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3548                                struct btrfs_root *root)
3549 {
3550         struct btrfs_fs_info *fs_info = root->fs_info;
3551         struct extent_io_tree *unpin;
3552         u64 start;
3553         u64 end;
3554         int ret;
3555
3556         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3557                 unpin = &fs_info->freed_extents[1];
3558         else
3559                 unpin = &fs_info->freed_extents[0];
3560
3561         while (1) {
3562                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3563                                             EXTENT_DIRTY);
3564                 if (ret)
3565                         break;
3566
3567                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3568
3569                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3570                 unpin_extent_range(root, start, end);
3571                 cond_resched();
3572         }
3573
3574         return ret;
3575 }
3576
3577 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3578                           struct btrfs_root *root,
3579                           struct btrfs_path *path,
3580                           u64 bytenr, u64 num_bytes,
3581                           int is_data, int reserved,
3582                           struct extent_buffer **must_clean)
3583 {
3584         int err = 0;
3585         struct extent_buffer *buf;
3586
3587         if (is_data)
3588                 goto pinit;
3589
3590         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3591         if (!buf)
3592                 goto pinit;
3593
3594         /* we can reuse a block if it hasn't been written
3595          * and it is from this transaction.  We can't
3596          * reuse anything from the tree log root because
3597          * it has tiny sub-transactions.
3598          */
3599         if (btrfs_buffer_uptodate(buf, 0) &&
3600             btrfs_try_tree_lock(buf)) {
3601                 u64 header_owner = btrfs_header_owner(buf);
3602                 u64 header_transid = btrfs_header_generation(buf);
3603                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3604                     header_transid == trans->transid &&
3605                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3606                         *must_clean = buf;
3607                         return 1;
3608                 }
3609                 btrfs_tree_unlock(buf);
3610         }
3611         free_extent_buffer(buf);
3612 pinit:
3613         if (path)
3614                 btrfs_set_path_blocking(path);
3615         /* unlocks the pinned mutex */
3616         btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3617
3618         BUG_ON(err < 0);
3619         return 0;
3620 }
3621
3622 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3623                                 struct btrfs_root *root,
3624                                 u64 bytenr, u64 num_bytes, u64 parent,
3625                                 u64 root_objectid, u64 owner_objectid,
3626                                 u64 owner_offset, int refs_to_drop,
3627                                 struct btrfs_delayed_extent_op *extent_op)
3628 {
3629         struct btrfs_key key;
3630         struct btrfs_path *path;
3631         struct btrfs_fs_info *info = root->fs_info;
3632         struct btrfs_root *extent_root = info->extent_root;
3633         struct extent_buffer *leaf;
3634         struct btrfs_extent_item *ei;
3635         struct btrfs_extent_inline_ref *iref;
3636         int ret;
3637         int is_data;
3638         int extent_slot = 0;
3639         int found_extent = 0;
3640         int num_to_del = 1;
3641         u32 item_size;
3642         u64 refs;
3643
3644         path = btrfs_alloc_path();
3645         if (!path)
3646                 return -ENOMEM;
3647
3648         path->reada = 1;
3649         path->leave_spinning = 1;
3650
3651         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3652         BUG_ON(!is_data && refs_to_drop != 1);
3653
3654         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3655                                     bytenr, num_bytes, parent,
3656                                     root_objectid, owner_objectid,
3657                                     owner_offset);
3658         if (ret == 0) {
3659                 extent_slot = path->slots[0];
3660                 while (extent_slot >= 0) {
3661                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3662                                               extent_slot);
3663                         if (key.objectid != bytenr)
3664                                 break;
3665                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3666                             key.offset == num_bytes) {
3667                                 found_extent = 1;
3668                                 break;
3669                         }
3670                         if (path->slots[0] - extent_slot > 5)
3671                                 break;
3672                         extent_slot--;
3673                 }
3674 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3675                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3676                 if (found_extent && item_size < sizeof(*ei))
3677                         found_extent = 0;
3678 #endif
3679                 if (!found_extent) {
3680                         BUG_ON(iref);
3681                         ret = remove_extent_backref(trans, extent_root, path,
3682                                                     NULL, refs_to_drop,
3683                                                     is_data);
3684                         BUG_ON(ret);
3685                         btrfs_release_path(extent_root, path);
3686                         path->leave_spinning = 1;
3687
3688                         key.objectid = bytenr;
3689                         key.type = BTRFS_EXTENT_ITEM_KEY;
3690                         key.offset = num_bytes;
3691
3692                         ret = btrfs_search_slot(trans, extent_root,
3693                                                 &key, path, -1, 1);
3694                         if (ret) {
3695                                 printk(KERN_ERR "umm, got %d back from search"
3696                                        ", was looking for %llu\n", ret,
3697                                        (unsigned long long)bytenr);
3698                                 btrfs_print_leaf(extent_root, path->nodes[0]);
3699                         }
3700                         BUG_ON(ret);
3701                         extent_slot = path->slots[0];
3702                 }
3703         } else {
3704                 btrfs_print_leaf(extent_root, path->nodes[0]);
3705                 WARN_ON(1);
3706                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3707                        "parent %llu root %llu  owner %llu offset %llu\n",
3708                        (unsigned long long)bytenr,
3709                        (unsigned long long)parent,
3710                        (unsigned long long)root_objectid,
3711                        (unsigned long long)owner_objectid,
3712                        (unsigned long long)owner_offset);
3713         }
3714
3715         leaf = path->nodes[0];
3716         item_size = btrfs_item_size_nr(leaf, extent_slot);
3717 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3718         if (item_size < sizeof(*ei)) {
3719                 BUG_ON(found_extent || extent_slot != path->slots[0]);
3720                 ret = convert_extent_item_v0(trans, extent_root, path,
3721                                              owner_objectid, 0);
3722                 BUG_ON(ret < 0);
3723
3724                 btrfs_release_path(extent_root, path);
3725                 path->leave_spinning = 1;
3726
3727                 key.objectid = bytenr;
3728                 key.type = BTRFS_EXTENT_ITEM_KEY;
3729                 key.offset = num_bytes;
3730
3731                 ret = btrfs_search_slot(trans, extent_root, &key, path,
3732                                         -1, 1);
3733                 if (ret) {
3734                         printk(KERN_ERR "umm, got %d back from search"
3735                                ", was looking for %llu\n", ret,
3736                                (unsigned long long)bytenr);
3737                         btrfs_print_leaf(extent_root, path->nodes[0]);
3738                 }
3739                 BUG_ON(ret);
3740                 extent_slot = path->slots[0];
3741                 leaf = path->nodes[0];
3742                 item_size = btrfs_item_size_nr(leaf, extent_slot);
3743         }
3744 #endif
3745         BUG_ON(item_size < sizeof(*ei));
3746         ei = btrfs_item_ptr(leaf, extent_slot,
3747                             struct btrfs_extent_item);
3748         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3749                 struct btrfs_tree_block_info *bi;
3750                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3751                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3752                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3753         }
3754
3755         refs = btrfs_extent_refs(leaf, ei);
3756         BUG_ON(refs < refs_to_drop);
3757         refs -= refs_to_drop;
3758
3759         if (refs > 0) {
3760                 if (extent_op)
3761                         __run_delayed_extent_op(extent_op, leaf, ei);
3762                 /*
3763                  * In the case of inline back ref, reference count will
3764                  * be updated by remove_extent_backref
3765                  */
3766                 if (iref) {
3767                         BUG_ON(!found_extent);
3768                 } else {
3769                         btrfs_set_extent_refs(leaf, ei, refs);
3770                         btrfs_mark_buffer_dirty(leaf);
3771                 }
3772                 if (found_extent) {
3773                         ret = remove_extent_backref(trans, extent_root, path,
3774                                                     iref, refs_to_drop,
3775                                                     is_data);
3776                         BUG_ON(ret);
3777                 }
3778         } else {
3779                 int mark_free = 0;
3780                 struct extent_buffer *must_clean = NULL;
3781
3782                 if (found_extent) {
3783                         BUG_ON(is_data && refs_to_drop !=
3784                                extent_data_ref_count(root, path, iref));
3785                         if (iref) {
3786                                 BUG_ON(path->slots[0] != extent_slot);
3787                         } else {
3788                                 BUG_ON(path->slots[0] != extent_slot + 1);
3789                                 path->slots[0] = extent_slot;
3790                                 num_to_del = 2;
3791                         }
3792                 }
3793
3794                 ret = pin_down_bytes(trans, root, path, bytenr,
3795                                      num_bytes, is_data, 0, &must_clean);
3796                 if (ret > 0)
3797                         mark_free = 1;
3798                 BUG_ON(ret < 0);
3799                 /*
3800                  * it is going to be very rare for someone to be waiting
3801                  * on the block we're freeing.  del_items might need to
3802                  * schedule, so rather than get fancy, just force it
3803                  * to blocking here
3804                  */
3805                 if (must_clean)
3806                         btrfs_set_lock_blocking(must_clean);
3807
3808                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3809                                       num_to_del);
3810                 BUG_ON(ret);
3811                 btrfs_release_path(extent_root, path);
3812
3813                 if (must_clean) {
3814                         clean_tree_block(NULL, root, must_clean);
3815                         btrfs_tree_unlock(must_clean);
3816                         free_extent_buffer(must_clean);
3817                 }
3818
3819                 if (is_data) {
3820                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3821                         BUG_ON(ret);
3822                 } else {
3823                         invalidate_mapping_pages(info->btree_inode->i_mapping,
3824                              bytenr >> PAGE_CACHE_SHIFT,
3825                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3826                 }
3827
3828                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3829                                          mark_free);
3830                 BUG_ON(ret);
3831         }
3832         btrfs_free_path(path);
3833         return ret;
3834 }
3835
3836 /*
3837  * when we free an extent, it is possible (and likely) that we free the last
3838  * delayed ref for that extent as well.  This searches the delayed ref tree for
3839  * a given extent, and if there are no other delayed refs to be processed, it
3840  * removes it from the tree.
3841  */
3842 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3843                                       struct btrfs_root *root, u64 bytenr)
3844 {
3845         struct btrfs_delayed_ref_head *head;
3846         struct btrfs_delayed_ref_root *delayed_refs;
3847         struct btrfs_delayed_ref_node *ref;
3848         struct rb_node *node;
3849         int ret;
3850
3851         delayed_refs = &trans->transaction->delayed_refs;
3852         spin_lock(&delayed_refs->lock);
3853         head = btrfs_find_delayed_ref_head(trans, bytenr);
3854         if (!head)
3855                 goto out;
3856
3857         node = rb_prev(&head->node.rb_node);
3858         if (!node)
3859                 goto out;
3860
3861         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3862
3863         /* there are still entries for this ref, we can't drop it */
3864         if (ref->bytenr == bytenr)
3865                 goto out;
3866
3867         if (head->extent_op) {
3868                 if (!head->must_insert_reserved)
3869                         goto out;
3870                 kfree(head->extent_op);
3871                 head->extent_op = NULL;
3872         }
3873
3874         /*
3875          * waiting for the lock here would deadlock.  If someone else has it
3876          * locked they are already in the process of dropping it anyway
3877          */
3878         if (!mutex_trylock(&head->mutex))
3879                 goto out;
3880
3881         /*
3882          * at this point we have a head with no other entries.  Go
3883          * ahead and process it.
3884          */
3885         head->node.in_tree = 0;
3886         rb_erase(&head->node.rb_node, &delayed_refs->root);
3887
3888         delayed_refs->num_entries--;
3889
3890         /*
3891          * we don't take a ref on the node because we're removing it from the
3892          * tree, so we just steal the ref the tree was holding.
3893          */
3894         delayed_refs->num_heads--;
3895         if (list_empty(&head->cluster))
3896                 delayed_refs->num_heads_ready--;
3897
3898         list_del_init(&head->cluster);
3899         spin_unlock(&delayed_refs->lock);
3900
3901         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
3902                                   &head->node, head->extent_op,
3903                                   head->must_insert_reserved);
3904         BUG_ON(ret);
3905         btrfs_put_delayed_ref(&head->node);
3906         return 0;
3907 out:
3908         spin_unlock(&delayed_refs->lock);
3909         return 0;
3910 }
3911
3912 int btrfs_free_extent(struct btrfs_trans_handle *trans,
3913                       struct btrfs_root *root,
3914                       u64 bytenr, u64 num_bytes, u64 parent,
3915                       u64 root_objectid, u64 owner, u64 offset)
3916 {
3917         int ret;
3918
3919         /*
3920          * tree log blocks never actually go into the extent allocation
3921          * tree, just update pinning info and exit early.
3922          */
3923         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
3924                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
3925                 /* unlocks the pinned mutex */
3926                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
3927                 ret = 0;
3928         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
3929                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
3930                                         parent, root_objectid, (int)owner,
3931                                         BTRFS_DROP_DELAYED_REF, NULL);
3932                 BUG_ON(ret);
3933                 ret = check_ref_cleanup(trans, root, bytenr);
3934                 BUG_ON(ret);
3935         } else {
3936                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
3937                                         parent, root_objectid, owner,
3938                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
3939                 BUG_ON(ret);
3940         }
3941         return ret;
3942 }
3943
3944 static u64 stripe_align(struct btrfs_root *root, u64 val)
3945 {
3946         u64 mask = ((u64)root->stripesize - 1);
3947         u64 ret = (val + mask) & ~mask;
3948         return ret;
3949 }
3950
3951 /*
3952  * when we wait for progress in the block group caching, its because
3953  * our allocation attempt failed at least once.  So, we must sleep
3954  * and let some progress happen before we try again.
3955  *
3956  * This function will sleep at least once waiting for new free space to
3957  * show up, and then it will check the block group free space numbers
3958  * for our min num_bytes.  Another option is to have it go ahead
3959  * and look in the rbtree for a free extent of a given size, but this
3960  * is a good start.
3961  */
3962 static noinline int
3963 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
3964                                 u64 num_bytes)
3965 {
3966         struct btrfs_caching_control *caching_ctl;
3967         DEFINE_WAIT(wait);
3968
3969         caching_ctl = get_caching_control(cache);
3970         if (!caching_ctl)
3971                 return 0;
3972
3973         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
3974                    (cache->free_space >= num_bytes));
3975
3976         put_caching_control(caching_ctl);
3977         return 0;
3978 }
3979
3980 static noinline int
3981 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
3982 {
3983         struct btrfs_caching_control *caching_ctl;
3984         DEFINE_WAIT(wait);
3985
3986         caching_ctl = get_caching_control(cache);
3987         if (!caching_ctl)
3988                 return 0;
3989
3990         wait_event(caching_ctl->wait, block_group_cache_done(cache));
3991
3992         put_caching_control(caching_ctl);
3993         return 0;
3994 }
3995
3996 enum btrfs_loop_type {
3997         LOOP_CACHED_ONLY = 0,
3998         LOOP_CACHING_NOWAIT = 1,
3999         LOOP_CACHING_WAIT = 2,
4000         LOOP_ALLOC_CHUNK = 3,
4001         LOOP_NO_EMPTY_SIZE = 4,
4002 };
4003
4004 /*
4005  * walks the btree of allocated extents and find a hole of a given size.
4006  * The key ins is changed to record the hole:
4007  * ins->objectid == block start
4008  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4009  * ins->offset == number of blocks
4010  * Any available blocks before search_start are skipped.
4011  */
4012 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4013                                      struct btrfs_root *orig_root,
4014                                      u64 num_bytes, u64 empty_size,
4015                                      u64 search_start, u64 search_end,
4016                                      u64 hint_byte, struct btrfs_key *ins,
4017                                      u64 exclude_start, u64 exclude_nr,
4018                                      int data)
4019 {
4020         int ret = 0;
4021         struct btrfs_root *root = orig_root->fs_info->extent_root;
4022         struct btrfs_free_cluster *last_ptr = NULL;
4023         struct btrfs_block_group_cache *block_group = NULL;
4024         int empty_cluster = 2 * 1024 * 1024;
4025         int allowed_chunk_alloc = 0;
4026         struct btrfs_space_info *space_info;
4027         int last_ptr_loop = 0;
4028         int loop = 0;
4029         bool found_uncached_bg = false;
4030         bool failed_cluster_refill = false;
4031
4032         WARN_ON(num_bytes < root->sectorsize);
4033         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4034         ins->objectid = 0;
4035         ins->offset = 0;
4036
4037         space_info = __find_space_info(root->fs_info, data);
4038
4039         if (orig_root->ref_cows || empty_size)
4040                 allowed_chunk_alloc = 1;
4041
4042         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4043                 last_ptr = &root->fs_info->meta_alloc_cluster;
4044                 if (!btrfs_test_opt(root, SSD))
4045                         empty_cluster = 64 * 1024;
4046         }
4047
4048         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4049                 last_ptr = &root->fs_info->data_alloc_cluster;
4050         }
4051
4052         if (last_ptr) {
4053                 spin_lock(&last_ptr->lock);
4054                 if (last_ptr->block_group)
4055                         hint_byte = last_ptr->window_start;
4056                 spin_unlock(&last_ptr->lock);
4057         }
4058
4059         search_start = max(search_start, first_logical_byte(root, 0));
4060         search_start = max(search_start, hint_byte);
4061
4062         if (!last_ptr)
4063                 empty_cluster = 0;
4064
4065         if (search_start == hint_byte) {
4066                 block_group = btrfs_lookup_block_group(root->fs_info,
4067                                                        search_start);
4068                 /*
4069                  * we don't want to use the block group if it doesn't match our
4070                  * allocation bits, or if its not cached.
4071                  */
4072                 if (block_group && block_group_bits(block_group, data) &&
4073                     block_group_cache_done(block_group)) {
4074                         down_read(&space_info->groups_sem);
4075                         if (list_empty(&block_group->list) ||
4076                             block_group->ro) {
4077                                 /*
4078                                  * someone is removing this block group,
4079                                  * we can't jump into the have_block_group
4080                                  * target because our list pointers are not
4081                                  * valid
4082                                  */
4083                                 btrfs_put_block_group(block_group);
4084                                 up_read(&space_info->groups_sem);
4085                         } else
4086                                 goto have_block_group;
4087                 } else if (block_group) {
4088                         btrfs_put_block_group(block_group);
4089                 }
4090         }
4091
4092 search:
4093         down_read(&space_info->groups_sem);
4094         list_for_each_entry(block_group, &space_info->block_groups, list) {
4095                 u64 offset;
4096                 int cached;
4097
4098                 atomic_inc(&block_group->count);
4099                 search_start = block_group->key.objectid;
4100
4101 have_block_group:
4102                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4103                         /*
4104                          * we want to start caching kthreads, but not too many
4105                          * right off the bat so we don't overwhelm the system,
4106                          * so only start them if there are less than 2 and we're
4107                          * in the initial allocation phase.
4108                          */
4109                         if (loop > LOOP_CACHING_NOWAIT ||
4110                             atomic_read(&space_info->caching_threads) < 2) {
4111                                 ret = cache_block_group(block_group);
4112                                 BUG_ON(ret);
4113                         }
4114                 }
4115
4116                 cached = block_group_cache_done(block_group);
4117                 if (unlikely(!cached)) {
4118                         found_uncached_bg = true;
4119
4120                         /* if we only want cached bgs, loop */
4121                         if (loop == LOOP_CACHED_ONLY)
4122                                 goto loop;
4123                 }
4124
4125                 if (unlikely(block_group->ro))
4126                         goto loop;
4127
4128                 /*
4129                  * Ok we want to try and use the cluster allocator, so lets look
4130                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4131                  * have tried the cluster allocator plenty of times at this
4132                  * point and not have found anything, so we are likely way too
4133                  * fragmented for the clustering stuff to find anything, so lets
4134                  * just skip it and let the allocator find whatever block it can
4135                  * find
4136                  */
4137                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4138                         /*
4139                          * the refill lock keeps out other
4140                          * people trying to start a new cluster
4141                          */
4142                         spin_lock(&last_ptr->refill_lock);
4143                         if (last_ptr->block_group &&
4144                             (last_ptr->block_group->ro ||
4145                             !block_group_bits(last_ptr->block_group, data))) {
4146                                 offset = 0;
4147                                 goto refill_cluster;
4148                         }
4149
4150                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4151                                                  num_bytes, search_start);
4152                         if (offset) {
4153                                 /* we have a block, we're done */
4154                                 spin_unlock(&last_ptr->refill_lock);
4155                                 goto checks;
4156                         }
4157
4158                         spin_lock(&last_ptr->lock);
4159                         /*
4160                          * whoops, this cluster doesn't actually point to
4161                          * this block group.  Get a ref on the block
4162                          * group is does point to and try again
4163                          */
4164                         if (!last_ptr_loop && last_ptr->block_group &&
4165                             last_ptr->block_group != block_group) {
4166
4167                                 btrfs_put_block_group(block_group);
4168                                 block_group = last_ptr->block_group;
4169                                 atomic_inc(&block_group->count);
4170                                 spin_unlock(&last_ptr->lock);
4171                                 spin_unlock(&last_ptr->refill_lock);
4172
4173                                 last_ptr_loop = 1;
4174                                 search_start = block_group->key.objectid;
4175                                 /*
4176                                  * we know this block group is properly
4177                                  * in the list because
4178                                  * btrfs_remove_block_group, drops the
4179                                  * cluster before it removes the block
4180                                  * group from the list
4181                                  */
4182                                 goto have_block_group;
4183                         }
4184                         spin_unlock(&last_ptr->lock);
4185 refill_cluster:
4186                         /*
4187                          * this cluster didn't work out, free it and
4188                          * start over
4189                          */
4190                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4191
4192                         last_ptr_loop = 0;
4193
4194                         /* allocate a cluster in this block group */
4195                         ret = btrfs_find_space_cluster(trans, root,
4196                                                block_group, last_ptr,
4197                                                offset, num_bytes,
4198                                                empty_cluster + empty_size);
4199                         if (ret == 0) {
4200                                 /*
4201                                  * now pull our allocation out of this
4202                                  * cluster
4203                                  */
4204                                 offset = btrfs_alloc_from_cluster(block_group,
4205                                                   last_ptr, num_bytes,
4206                                                   search_start);
4207                                 if (offset) {
4208                                         /* we found one, proceed */
4209                                         spin_unlock(&last_ptr->refill_lock);
4210                                         goto checks;
4211                                 }
4212                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4213                                    && !failed_cluster_refill) {
4214                                 spin_unlock(&last_ptr->refill_lock);
4215
4216                                 failed_cluster_refill = true;
4217                                 wait_block_group_cache_progress(block_group,
4218                                        num_bytes + empty_cluster + empty_size);
4219                                 goto have_block_group;
4220                         }
4221
4222                         /*
4223                          * at this point we either didn't find a cluster
4224                          * or we weren't able to allocate a block from our
4225                          * cluster.  Free the cluster we've been trying
4226                          * to use, and go to the next block group
4227                          */
4228                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4229                         spin_unlock(&last_ptr->refill_lock);
4230                         goto loop;
4231                 }
4232
4233                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4234                                                     num_bytes, empty_size);
4235                 if (!offset && (cached || (!cached &&
4236                                            loop == LOOP_CACHING_NOWAIT))) {
4237                         goto loop;
4238                 } else if (!offset && (!cached &&
4239                                        loop > LOOP_CACHING_NOWAIT)) {
4240                         wait_block_group_cache_progress(block_group,
4241                                         num_bytes + empty_size);
4242                         goto have_block_group;
4243                 }
4244 checks:
4245                 search_start = stripe_align(root, offset);
4246                 /* move on to the next group */
4247                 if (search_start + num_bytes >= search_end) {
4248                         btrfs_add_free_space(block_group, offset, num_bytes);
4249                         goto loop;
4250                 }
4251
4252                 /* move on to the next group */
4253                 if (search_start + num_bytes >
4254                     block_group->key.objectid + block_group->key.offset) {
4255                         btrfs_add_free_space(block_group, offset, num_bytes);
4256                         goto loop;
4257                 }
4258
4259                 if (exclude_nr > 0 &&
4260                     (search_start + num_bytes > exclude_start &&
4261                      search_start < exclude_start + exclude_nr)) {
4262                         search_start = exclude_start + exclude_nr;
4263
4264                         btrfs_add_free_space(block_group, offset, num_bytes);
4265                         /*
4266                          * if search_start is still in this block group
4267                          * then we just re-search this block group
4268                          */
4269                         if (search_start >= block_group->key.objectid &&
4270                             search_start < (block_group->key.objectid +
4271                                             block_group->key.offset))
4272                                 goto have_block_group;
4273                         goto loop;
4274                 }
4275
4276                 ins->objectid = search_start;
4277                 ins->offset = num_bytes;
4278
4279                 if (offset < search_start)
4280                         btrfs_add_free_space(block_group, offset,
4281                                              search_start - offset);
4282                 BUG_ON(offset > search_start);
4283
4284                 update_reserved_extents(block_group, num_bytes, 1);
4285
4286                 /* we are all good, lets return */
4287                 break;
4288 loop:
4289                 failed_cluster_refill = false;
4290                 btrfs_put_block_group(block_group);
4291         }
4292         up_read(&space_info->groups_sem);
4293
4294         /* LOOP_CACHED_ONLY, only search fully cached block groups
4295          * LOOP_CACHING_NOWAIT, search partially cached block groups, but
4296          *                      dont wait foR them to finish caching
4297          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4298          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4299          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4300          *                      again
4301          */
4302         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4303             (found_uncached_bg || empty_size || empty_cluster ||
4304              allowed_chunk_alloc)) {
4305                 if (found_uncached_bg) {
4306                         found_uncached_bg = false;
4307                         if (loop < LOOP_CACHING_WAIT) {
4308                                 loop++;
4309                                 goto search;
4310                         }
4311                 }
4312
4313                 if (loop == LOOP_ALLOC_CHUNK) {
4314                         empty_size = 0;
4315                         empty_cluster = 0;
4316                 }
4317
4318                 if (allowed_chunk_alloc) {
4319                         ret = do_chunk_alloc(trans, root, num_bytes +
4320                                              2 * 1024 * 1024, data, 1);
4321                         allowed_chunk_alloc = 0;
4322                 } else {
4323                         space_info->force_alloc = 1;
4324                 }
4325
4326                 if (loop < LOOP_NO_EMPTY_SIZE) {
4327                         loop++;
4328                         goto search;
4329                 }
4330                 ret = -ENOSPC;
4331         } else if (!ins->objectid) {
4332                 ret = -ENOSPC;
4333         }
4334
4335         /* we found what we needed */
4336         if (ins->objectid) {
4337                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4338                         trans->block_group = block_group->key.objectid;
4339
4340                 btrfs_put_block_group(block_group);
4341                 ret = 0;
4342         }
4343
4344         return ret;
4345 }
4346
4347 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4348                             int dump_block_groups)
4349 {
4350         struct btrfs_block_group_cache *cache;
4351
4352         spin_lock(&info->lock);
4353         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4354                (unsigned long long)(info->total_bytes - info->bytes_used -
4355                                     info->bytes_pinned - info->bytes_reserved -
4356                                     info->bytes_super),
4357                (info->full) ? "" : "not ");
4358         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4359                " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4360                "\n",
4361                (unsigned long long)info->total_bytes,
4362                (unsigned long long)info->bytes_pinned,
4363                (unsigned long long)info->bytes_delalloc,
4364                (unsigned long long)info->bytes_may_use,
4365                (unsigned long long)info->bytes_used,
4366                (unsigned long long)info->bytes_root,
4367                (unsigned long long)info->bytes_super,
4368                (unsigned long long)info->bytes_reserved);
4369         spin_unlock(&info->lock);
4370
4371         if (!dump_block_groups)
4372                 return;
4373
4374         down_read(&info->groups_sem);
4375         list_for_each_entry(cache, &info->block_groups, list) {
4376                 spin_lock(&cache->lock);
4377                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4378                        "%llu pinned %llu reserved\n",
4379                        (unsigned long long)cache->key.objectid,
4380                        (unsigned long long)cache->key.offset,
4381                        (unsigned long long)btrfs_block_group_used(&cache->item),
4382                        (unsigned long long)cache->pinned,
4383                        (unsigned long long)cache->reserved);
4384                 btrfs_dump_free_space(cache, bytes);
4385                 spin_unlock(&cache->lock);
4386         }
4387         up_read(&info->groups_sem);
4388 }
4389
4390 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4391                          struct btrfs_root *root,
4392                          u64 num_bytes, u64 min_alloc_size,
4393                          u64 empty_size, u64 hint_byte,
4394                          u64 search_end, struct btrfs_key *ins,
4395                          u64 data)
4396 {
4397         int ret;
4398         u64 search_start = 0;
4399         struct btrfs_fs_info *info = root->fs_info;
4400
4401         data = btrfs_get_alloc_profile(root, data);
4402 again:
4403         /*
4404          * the only place that sets empty_size is btrfs_realloc_node, which
4405          * is not called recursively on allocations
4406          */
4407         if (empty_size || root->ref_cows) {
4408                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
4409                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4410                                      2 * 1024 * 1024,
4411                                      BTRFS_BLOCK_GROUP_METADATA |
4412                                      (info->metadata_alloc_profile &
4413                                       info->avail_metadata_alloc_bits), 0);
4414                 }
4415                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4416                                      num_bytes + 2 * 1024 * 1024, data, 0);
4417         }
4418
4419         WARN_ON(num_bytes < root->sectorsize);
4420         ret = find_free_extent(trans, root, num_bytes, empty_size,
4421                                search_start, search_end, hint_byte, ins,
4422                                trans->alloc_exclude_start,
4423                                trans->alloc_exclude_nr, data);
4424
4425         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4426                 num_bytes = num_bytes >> 1;
4427                 num_bytes = num_bytes & ~(root->sectorsize - 1);
4428                 num_bytes = max(num_bytes, min_alloc_size);
4429                 do_chunk_alloc(trans, root->fs_info->extent_root,
4430                                num_bytes, data, 1);
4431                 goto again;
4432         }
4433         if (ret == -ENOSPC) {
4434                 struct btrfs_space_info *sinfo;
4435
4436                 sinfo = __find_space_info(root->fs_info, data);
4437                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4438                        "wanted %llu\n", (unsigned long long)data,
4439                        (unsigned long long)num_bytes);
4440                 dump_space_info(sinfo, num_bytes, 1);
4441         }
4442
4443         return ret;
4444 }
4445
4446 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4447 {
4448         struct btrfs_block_group_cache *cache;
4449         int ret = 0;
4450
4451         cache = btrfs_lookup_block_group(root->fs_info, start);
4452         if (!cache) {
4453                 printk(KERN_ERR "Unable to find block group for %llu\n",
4454                        (unsigned long long)start);
4455                 return -ENOSPC;
4456         }
4457
4458         ret = btrfs_discard_extent(root, start, len);
4459
4460         btrfs_add_free_space(cache, start, len);
4461         update_reserved_extents(cache, len, 0);
4462         btrfs_put_block_group(cache);
4463
4464         return ret;
4465 }
4466
4467 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4468                                       struct btrfs_root *root,
4469                                       u64 parent, u64 root_objectid,
4470                                       u64 flags, u64 owner, u64 offset,
4471                                       struct btrfs_key *ins, int ref_mod)
4472 {
4473         int ret;
4474         struct btrfs_fs_info *fs_info = root->fs_info;
4475         struct btrfs_extent_item *extent_item;
4476         struct btrfs_extent_inline_ref *iref;
4477         struct btrfs_path *path;
4478         struct extent_buffer *leaf;
4479         int type;
4480         u32 size;
4481
4482         if (parent > 0)
4483                 type = BTRFS_SHARED_DATA_REF_KEY;
4484         else
4485                 type = BTRFS_EXTENT_DATA_REF_KEY;
4486
4487         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4488
4489         path = btrfs_alloc_path();
4490         BUG_ON(!path);
4491
4492         path->leave_spinning = 1;
4493         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4494                                       ins, size);
4495         BUG_ON(ret);
4496
4497         leaf = path->nodes[0];
4498         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4499                                      struct btrfs_extent_item);
4500         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4501         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4502         btrfs_set_extent_flags(leaf, extent_item,
4503                                flags | BTRFS_EXTENT_FLAG_DATA);
4504
4505         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4506         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4507         if (parent > 0) {
4508                 struct btrfs_shared_data_ref *ref;
4509                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4510                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4511                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4512         } else {
4513                 struct btrfs_extent_data_ref *ref;
4514                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4515                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4516                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4517                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4518                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4519         }
4520
4521         btrfs_mark_buffer_dirty(path->nodes[0]);
4522         btrfs_free_path(path);
4523
4524         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4525                                  1, 0);
4526         if (ret) {
4527                 printk(KERN_ERR "btrfs update block group failed for %llu "
4528                        "%llu\n", (unsigned long long)ins->objectid,
4529                        (unsigned long long)ins->offset);
4530                 BUG();
4531         }
4532         return ret;
4533 }
4534
4535 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4536                                      struct btrfs_root *root,
4537                                      u64 parent, u64 root_objectid,
4538                                      u64 flags, struct btrfs_disk_key *key,
4539                                      int level, struct btrfs_key *ins)
4540 {
4541         int ret;
4542         struct btrfs_fs_info *fs_info = root->fs_info;
4543         struct btrfs_extent_item *extent_item;
4544         struct btrfs_tree_block_info *block_info;
4545         struct btrfs_extent_inline_ref *iref;
4546         struct btrfs_path *path;
4547         struct extent_buffer *leaf;
4548         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4549
4550         path = btrfs_alloc_path();
4551         BUG_ON(!path);
4552
4553         path->leave_spinning = 1;
4554         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4555                                       ins, size);
4556         BUG_ON(ret);
4557
4558         leaf = path->nodes[0];
4559         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4560                                      struct btrfs_extent_item);
4561         btrfs_set_extent_refs(leaf, extent_item, 1);
4562         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4563         btrfs_set_extent_flags(leaf, extent_item,
4564                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4565         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4566
4567         btrfs_set_tree_block_key(leaf, block_info, key);
4568         btrfs_set_tree_block_level(leaf, block_info, level);
4569
4570         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4571         if (parent > 0) {
4572                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4573                 btrfs_set_extent_inline_ref_type(leaf, iref,
4574                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4575                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4576         } else {
4577                 btrfs_set_extent_inline_ref_type(leaf, iref,
4578                                                  BTRFS_TREE_BLOCK_REF_KEY);
4579                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4580         }
4581
4582         btrfs_mark_buffer_dirty(leaf);
4583         btrfs_free_path(path);
4584
4585         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4586                                  1, 0);
4587         if (ret) {
4588                 printk(KERN_ERR "btrfs update block group failed for %llu "
4589                        "%llu\n", (unsigned long long)ins->objectid,
4590                        (unsigned long long)ins->offset);
4591                 BUG();
4592         }
4593         return ret;
4594 }
4595
4596 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4597                                      struct btrfs_root *root,
4598                                      u64 root_objectid, u64 owner,
4599                                      u64 offset, struct btrfs_key *ins)
4600 {
4601         int ret;
4602
4603         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4604
4605         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4606                                          0, root_objectid, owner, offset,
4607                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4608         return ret;
4609 }
4610
4611 /*
4612  * this is used by the tree logging recovery code.  It records that
4613  * an extent has been allocated and makes sure to clear the free
4614  * space cache bits as well
4615  */
4616 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4617                                    struct btrfs_root *root,
4618                                    u64 root_objectid, u64 owner, u64 offset,
4619                                    struct btrfs_key *ins)
4620 {
4621         int ret;
4622         struct btrfs_block_group_cache *block_group;
4623         struct btrfs_caching_control *caching_ctl;
4624         u64 start = ins->objectid;
4625         u64 num_bytes = ins->offset;
4626
4627         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4628         cache_block_group(block_group);
4629         caching_ctl = get_caching_control(block_group);
4630
4631         if (!caching_ctl) {
4632                 BUG_ON(!block_group_cache_done(block_group));
4633                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4634                 BUG_ON(ret);
4635         } else {
4636                 mutex_lock(&caching_ctl->mutex);
4637
4638                 if (start >= caching_ctl->progress) {
4639                         ret = add_excluded_extent(root, start, num_bytes);
4640                         BUG_ON(ret);
4641                 } else if (start + num_bytes <= caching_ctl->progress) {
4642                         ret = btrfs_remove_free_space(block_group,
4643                                                       start, num_bytes);
4644                         BUG_ON(ret);
4645                 } else {
4646                         num_bytes = caching_ctl->progress - start;
4647                         ret = btrfs_remove_free_space(block_group,
4648                                                       start, num_bytes);
4649                         BUG_ON(ret);
4650
4651                         start = caching_ctl->progress;
4652                         num_bytes = ins->objectid + ins->offset -
4653                                     caching_ctl->progress;
4654                         ret = add_excluded_extent(root, start, num_bytes);
4655                         BUG_ON(ret);
4656                 }
4657
4658                 mutex_unlock(&caching_ctl->mutex);
4659                 put_caching_control(caching_ctl);
4660         }
4661
4662         update_reserved_extents(block_group, ins->offset, 1);
4663         btrfs_put_block_group(block_group);
4664         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4665                                          0, owner, offset, ins, 1);
4666         return ret;
4667 }
4668
4669 /*
4670  * finds a free extent and does all the dirty work required for allocation
4671  * returns the key for the extent through ins, and a tree buffer for
4672  * the first block of the extent through buf.
4673  *
4674  * returns 0 if everything worked, non-zero otherwise.
4675  */
4676 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4677                             struct btrfs_root *root,
4678                             u64 num_bytes, u64 parent, u64 root_objectid,
4679                             struct btrfs_disk_key *key, int level,
4680                             u64 empty_size, u64 hint_byte, u64 search_end,
4681                             struct btrfs_key *ins)
4682 {
4683         int ret;
4684         u64 flags = 0;
4685
4686         ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4687                                    empty_size, hint_byte, search_end,
4688                                    ins, 0);
4689         if (ret)
4690                 return ret;
4691
4692         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4693                 if (parent == 0)
4694                         parent = ins->objectid;
4695                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4696         } else
4697                 BUG_ON(parent > 0);
4698
4699         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4700                 struct btrfs_delayed_extent_op *extent_op;
4701                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4702                 BUG_ON(!extent_op);
4703                 if (key)
4704                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4705                 else
4706                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4707                 extent_op->flags_to_set = flags;
4708                 extent_op->update_key = 1;
4709                 extent_op->update_flags = 1;
4710                 extent_op->is_data = 0;
4711
4712                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4713                                         ins->offset, parent, root_objectid,
4714                                         level, BTRFS_ADD_DELAYED_EXTENT,
4715                                         extent_op);
4716                 BUG_ON(ret);
4717         }
4718         return ret;
4719 }
4720
4721 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4722                                             struct btrfs_root *root,
4723                                             u64 bytenr, u32 blocksize,
4724                                             int level)
4725 {
4726         struct extent_buffer *buf;
4727
4728         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4729         if (!buf)
4730                 return ERR_PTR(-ENOMEM);
4731         btrfs_set_header_generation(buf, trans->transid);
4732         btrfs_set_buffer_lockdep_class(buf, level);
4733         btrfs_tree_lock(buf);
4734         clean_tree_block(trans, root, buf);
4735
4736         btrfs_set_lock_blocking(buf);
4737         btrfs_set_buffer_uptodate(buf);
4738
4739         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4740                 set_extent_dirty(&root->dirty_log_pages, buf->start,
4741                          buf->start + buf->len - 1, GFP_NOFS);
4742         } else {
4743                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4744                          buf->start + buf->len - 1, GFP_NOFS);
4745         }
4746         trans->blocks_used++;
4747         /* this returns a buffer locked for blocking */
4748         return buf;
4749 }
4750
4751 /*
4752  * helper function to allocate a block for a given tree
4753  * returns the tree buffer or NULL.
4754  */
4755 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4756                                         struct btrfs_root *root, u32 blocksize,
4757                                         u64 parent, u64 root_objectid,
4758                                         struct btrfs_disk_key *key, int level,
4759                                         u64 hint, u64 empty_size)
4760 {
4761         struct btrfs_key ins;
4762         int ret;
4763         struct extent_buffer *buf;
4764
4765         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4766                                key, level, empty_size, hint, (u64)-1, &ins);
4767         if (ret) {
4768                 BUG_ON(ret > 0);
4769                 return ERR_PTR(ret);
4770         }
4771
4772         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4773                                     blocksize, level);
4774         return buf;
4775 }
4776
4777 struct walk_control {
4778         u64 refs[BTRFS_MAX_LEVEL];
4779         u64 flags[BTRFS_MAX_LEVEL];
4780         struct btrfs_key update_progress;
4781         int stage;
4782         int level;
4783         int shared_level;
4784         int update_ref;
4785         int keep_locks;
4786         int reada_slot;
4787         int reada_count;
4788 };
4789
4790 #define DROP_REFERENCE  1
4791 #define UPDATE_BACKREF  2
4792
4793 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4794                                      struct btrfs_root *root,
4795                                      struct walk_control *wc,
4796                                      struct btrfs_path *path)
4797 {
4798         u64 bytenr;
4799         u64 generation;
4800         u64 refs;
4801         u64 last = 0;
4802         u32 nritems;
4803         u32 blocksize;
4804         struct btrfs_key key;
4805         struct extent_buffer *eb;
4806         int ret;
4807         int slot;
4808         int nread = 0;
4809
4810         if (path->slots[wc->level] < wc->reada_slot) {
4811                 wc->reada_count = wc->reada_count * 2 / 3;
4812                 wc->reada_count = max(wc->reada_count, 2);
4813         } else {
4814                 wc->reada_count = wc->reada_count * 3 / 2;
4815                 wc->reada_count = min_t(int, wc->reada_count,
4816                                         BTRFS_NODEPTRS_PER_BLOCK(root));
4817         }
4818
4819         eb = path->nodes[wc->level];
4820         nritems = btrfs_header_nritems(eb);
4821         blocksize = btrfs_level_size(root, wc->level - 1);
4822
4823         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4824                 if (nread >= wc->reada_count)
4825                         break;
4826
4827                 cond_resched();
4828                 bytenr = btrfs_node_blockptr(eb, slot);
4829                 generation = btrfs_node_ptr_generation(eb, slot);
4830
4831                 if (slot == path->slots[wc->level])
4832                         goto reada;
4833
4834                 if (wc->stage == UPDATE_BACKREF &&
4835                     generation <= root->root_key.offset)
4836                         continue;
4837
4838                 if (wc->stage == DROP_REFERENCE) {
4839                         ret = btrfs_lookup_extent_info(trans, root,
4840                                                 bytenr, blocksize,
4841                                                 &refs, NULL);
4842                         BUG_ON(ret);
4843                         BUG_ON(refs == 0);
4844                         if (refs == 1)
4845                                 goto reada;
4846
4847                         if (!wc->update_ref ||
4848                             generation <= root->root_key.offset)
4849                                 continue;
4850                         btrfs_node_key_to_cpu(eb, &key, slot);
4851                         ret = btrfs_comp_cpu_keys(&key,
4852                                                   &wc->update_progress);
4853                         if (ret < 0)
4854                                 continue;
4855                 }
4856 reada:
4857                 ret = readahead_tree_block(root, bytenr, blocksize,
4858                                            generation);
4859                 if (ret)
4860                         break;
4861                 last = bytenr + blocksize;
4862                 nread++;
4863         }
4864         wc->reada_slot = slot;
4865 }
4866
4867 /*
4868  * hepler to process tree block while walking down the tree.
4869  *
4870  * when wc->stage == UPDATE_BACKREF, this function updates
4871  * back refs for pointers in the block.
4872  *
4873  * NOTE: return value 1 means we should stop walking down.
4874  */
4875 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4876                                    struct btrfs_root *root,
4877                                    struct btrfs_path *path,
4878                                    struct walk_control *wc)
4879 {
4880         int level = wc->level;
4881         struct extent_buffer *eb = path->nodes[level];
4882         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4883         int ret;
4884
4885         if (wc->stage == UPDATE_BACKREF &&
4886             btrfs_header_owner(eb) != root->root_key.objectid)
4887                 return 1;
4888
4889         /*
4890          * when reference count of tree block is 1, it won't increase
4891          * again. once full backref flag is set, we never clear it.
4892          */
4893         if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4894             (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
4895                 BUG_ON(!path->locks[level]);
4896                 ret = btrfs_lookup_extent_info(trans, root,
4897                                                eb->start, eb->len,
4898                                                &wc->refs[level],
4899                                                &wc->flags[level]);
4900                 BUG_ON(ret);
4901                 BUG_ON(wc->refs[level] == 0);
4902         }
4903
4904         if (wc->stage == DROP_REFERENCE) {
4905                 if (wc->refs[level] > 1)
4906                         return 1;
4907
4908                 if (path->locks[level] && !wc->keep_locks) {
4909                         btrfs_tree_unlock(eb);
4910                         path->locks[level] = 0;
4911                 }
4912                 return 0;
4913         }
4914
4915         /* wc->stage == UPDATE_BACKREF */
4916         if (!(wc->flags[level] & flag)) {
4917                 BUG_ON(!path->locks[level]);
4918                 ret = btrfs_inc_ref(trans, root, eb, 1);
4919                 BUG_ON(ret);
4920                 ret = btrfs_dec_ref(trans, root, eb, 0);
4921                 BUG_ON(ret);
4922                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
4923                                                   eb->len, flag, 0);
4924                 BUG_ON(ret);
4925                 wc->flags[level] |= flag;
4926         }
4927
4928         /*
4929          * the block is shared by multiple trees, so it's not good to
4930          * keep the tree lock
4931          */
4932         if (path->locks[level] && level > 0) {
4933                 btrfs_tree_unlock(eb);
4934                 path->locks[level] = 0;
4935         }
4936         return 0;
4937 }
4938
4939 /*
4940  * hepler to process tree block pointer.
4941  *
4942  * when wc->stage == DROP_REFERENCE, this function checks
4943  * reference count of the block pointed to. if the block
4944  * is shared and we need update back refs for the subtree
4945  * rooted at the block, this function changes wc->stage to
4946  * UPDATE_BACKREF. if the block is shared and there is no
4947  * need to update back, this function drops the reference
4948  * to the block.
4949  *
4950  * NOTE: return value 1 means we should stop walking down.
4951  */
4952 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4953                                  struct btrfs_root *root,
4954                                  struct btrfs_path *path,
4955                                  struct walk_control *wc)
4956 {
4957         u64 bytenr;
4958         u64 generation;
4959         u64 parent;
4960         u32 blocksize;
4961         struct btrfs_key key;
4962         struct extent_buffer *next;
4963         int level = wc->level;
4964         int reada = 0;
4965         int ret = 0;
4966
4967         generation = btrfs_node_ptr_generation(path->nodes[level],
4968                                                path->slots[level]);
4969         /*
4970          * if the lower level block was created before the snapshot
4971          * was created, we know there is no need to update back refs
4972          * for the subtree
4973          */
4974         if (wc->stage == UPDATE_BACKREF &&
4975             generation <= root->root_key.offset)
4976                 return 1;
4977
4978         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
4979         blocksize = btrfs_level_size(root, level - 1);
4980
4981         next = btrfs_find_tree_block(root, bytenr, blocksize);
4982         if (!next) {
4983                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
4984                 reada = 1;
4985         }
4986         btrfs_tree_lock(next);
4987         btrfs_set_lock_blocking(next);
4988
4989         if (wc->stage == DROP_REFERENCE) {
4990                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
4991                                                &wc->refs[level - 1],
4992                                                &wc->flags[level - 1]);
4993                 BUG_ON(ret);
4994                 BUG_ON(wc->refs[level - 1] == 0);
4995
4996                 if (wc->refs[level - 1] > 1) {
4997                         if (!wc->update_ref ||
4998                             generation <= root->root_key.offset)
4999                                 goto skip;
5000
5001                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5002                                               path->slots[level]);
5003                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5004                         if (ret < 0)
5005                                 goto skip;
5006
5007                         wc->stage = UPDATE_BACKREF;
5008                         wc->shared_level = level - 1;
5009                 }
5010         }
5011
5012         if (!btrfs_buffer_uptodate(next, generation)) {
5013                 btrfs_tree_unlock(next);
5014                 free_extent_buffer(next);
5015                 next = NULL;
5016         }
5017
5018         if (!next) {
5019                 if (reada && level == 1)
5020                         reada_walk_down(trans, root, wc, path);
5021                 next = read_tree_block(root, bytenr, blocksize, generation);
5022                 btrfs_tree_lock(next);
5023                 btrfs_set_lock_blocking(next);
5024         }
5025
5026         level--;
5027         BUG_ON(level != btrfs_header_level(next));
5028         path->nodes[level] = next;
5029         path->slots[level] = 0;
5030         path->locks[level] = 1;
5031         wc->level = level;
5032         if (wc->level == 1)
5033                 wc->reada_slot = 0;
5034         return 0;
5035 skip:
5036         wc->refs[level - 1] = 0;
5037         wc->flags[level - 1] = 0;
5038
5039         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5040                 parent = path->nodes[level]->start;
5041         } else {
5042                 BUG_ON(root->root_key.objectid !=
5043                        btrfs_header_owner(path->nodes[level]));
5044                 parent = 0;
5045         }
5046
5047         ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5048                                 root->root_key.objectid, level - 1, 0);
5049         BUG_ON(ret);
5050
5051         btrfs_tree_unlock(next);
5052         free_extent_buffer(next);
5053         return 1;
5054 }
5055
5056 /*
5057  * hepler to process tree block while walking up the tree.
5058  *
5059  * when wc->stage == DROP_REFERENCE, this function drops
5060  * reference count on the block.
5061  *
5062  * when wc->stage == UPDATE_BACKREF, this function changes
5063  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5064  * to UPDATE_BACKREF previously while processing the block.
5065  *
5066  * NOTE: return value 1 means we should stop walking up.
5067  */
5068 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5069                                  struct btrfs_root *root,
5070                                  struct btrfs_path *path,
5071                                  struct walk_control *wc)
5072 {
5073         int ret = 0;
5074         int level = wc->level;
5075         struct extent_buffer *eb = path->nodes[level];
5076         u64 parent = 0;
5077
5078         if (wc->stage == UPDATE_BACKREF) {
5079                 BUG_ON(wc->shared_level < level);
5080                 if (level < wc->shared_level)
5081                         goto out;
5082
5083                 ret = find_next_key(path, level + 1, &wc->update_progress);
5084                 if (ret > 0)
5085                         wc->update_ref = 0;
5086
5087                 wc->stage = DROP_REFERENCE;
5088                 wc->shared_level = -1;
5089                 path->slots[level] = 0;
5090
5091                 /*
5092                  * check reference count again if the block isn't locked.
5093                  * we should start walking down the tree again if reference
5094                  * count is one.
5095                  */
5096                 if (!path->locks[level]) {
5097                         BUG_ON(level == 0);
5098                         btrfs_tree_lock(eb);
5099                         btrfs_set_lock_blocking(eb);
5100                         path->locks[level] = 1;
5101
5102                         ret = btrfs_lookup_extent_info(trans, root,
5103                                                        eb->start, eb->len,
5104                                                        &wc->refs[level],
5105                                                        &wc->flags[level]);
5106                         BUG_ON(ret);
5107                         BUG_ON(wc->refs[level] == 0);
5108                         if (wc->refs[level] == 1) {
5109                                 btrfs_tree_unlock(eb);
5110                                 path->locks[level] = 0;
5111                                 return 1;
5112                         }
5113                 }
5114         }
5115
5116         /* wc->stage == DROP_REFERENCE */
5117         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5118
5119         if (wc->refs[level] == 1) {
5120                 if (level == 0) {
5121                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5122                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5123                         else
5124                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5125                         BUG_ON(ret);
5126                 }
5127                 /* make block locked assertion in clean_tree_block happy */
5128                 if (!path->locks[level] &&
5129                     btrfs_header_generation(eb) == trans->transid) {
5130                         btrfs_tree_lock(eb);
5131                         btrfs_set_lock_blocking(eb);
5132                         path->locks[level] = 1;
5133                 }
5134                 clean_tree_block(trans, root, eb);
5135         }
5136
5137         if (eb == root->node) {
5138                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5139                         parent = eb->start;
5140                 else
5141                         BUG_ON(root->root_key.objectid !=
5142                                btrfs_header_owner(eb));
5143         } else {
5144                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5145                         parent = path->nodes[level + 1]->start;
5146                 else
5147                         BUG_ON(root->root_key.objectid !=
5148                                btrfs_header_owner(path->nodes[level + 1]));
5149         }
5150
5151         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5152                                 root->root_key.objectid, level, 0);
5153         BUG_ON(ret);
5154 out:
5155         wc->refs[level] = 0;
5156         wc->flags[level] = 0;
5157         return ret;
5158 }
5159
5160 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5161                                    struct btrfs_root *root,
5162                                    struct btrfs_path *path,
5163                                    struct walk_control *wc)
5164 {
5165         int level = wc->level;
5166         int ret;
5167
5168         while (level >= 0) {
5169                 if (path->slots[level] >=
5170                     btrfs_header_nritems(path->nodes[level]))
5171                         break;
5172
5173                 ret = walk_down_proc(trans, root, path, wc);
5174                 if (ret > 0)
5175                         break;
5176
5177                 if (level == 0)
5178                         break;
5179
5180                 ret = do_walk_down(trans, root, path, wc);
5181                 if (ret > 0) {
5182                         path->slots[level]++;
5183                         continue;
5184                 }
5185                 level = wc->level;
5186         }
5187         return 0;
5188 }
5189
5190 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5191                                  struct btrfs_root *root,
5192                                  struct btrfs_path *path,
5193                                  struct walk_control *wc, int max_level)
5194 {
5195         int level = wc->level;
5196         int ret;
5197
5198         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5199         while (level < max_level && path->nodes[level]) {
5200                 wc->level = level;
5201                 if (path->slots[level] + 1 <
5202                     btrfs_header_nritems(path->nodes[level])) {
5203                         path->slots[level]++;
5204                         return 0;
5205                 } else {
5206                         ret = walk_up_proc(trans, root, path, wc);
5207                         if (ret > 0)
5208                                 return 0;
5209
5210                         if (path->locks[level]) {
5211                                 btrfs_tree_unlock(path->nodes[level]);
5212                                 path->locks[level] = 0;
5213                         }
5214                         free_extent_buffer(path->nodes[level]);
5215                         path->nodes[level] = NULL;
5216                         level++;
5217                 }
5218         }
5219         return 1;
5220 }
5221
5222 /*
5223  * drop a subvolume tree.
5224  *
5225  * this function traverses the tree freeing any blocks that only
5226  * referenced by the tree.
5227  *
5228  * when a shared tree block is found. this function decreases its
5229  * reference count by one. if update_ref is true, this function
5230  * also make sure backrefs for the shared block and all lower level
5231  * blocks are properly updated.
5232  */
5233 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5234 {
5235         struct btrfs_path *path;
5236         struct btrfs_trans_handle *trans;
5237         struct btrfs_root *tree_root = root->fs_info->tree_root;
5238         struct btrfs_root_item *root_item = &root->root_item;
5239         struct walk_control *wc;
5240         struct btrfs_key key;
5241         int err = 0;
5242         int ret;
5243         int level;
5244
5245         path = btrfs_alloc_path();
5246         BUG_ON(!path);
5247
5248         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5249         BUG_ON(!wc);
5250
5251         trans = btrfs_start_transaction(tree_root, 1);
5252
5253         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5254                 level = btrfs_header_level(root->node);
5255                 path->nodes[level] = btrfs_lock_root_node(root);
5256                 btrfs_set_lock_blocking(path->nodes[level]);
5257                 path->slots[level] = 0;
5258                 path->locks[level] = 1;
5259                 memset(&wc->update_progress, 0,
5260                        sizeof(wc->update_progress));
5261         } else {
5262                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5263                 memcpy(&wc->update_progress, &key,
5264                        sizeof(wc->update_progress));
5265
5266                 level = root_item->drop_level;
5267                 BUG_ON(level == 0);
5268                 path->lowest_level = level;
5269                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5270                 path->lowest_level = 0;
5271                 if (ret < 0) {
5272                         err = ret;
5273                         goto out;
5274                 }
5275                 WARN_ON(ret > 0);
5276
5277                 /*
5278                  * unlock our path, this is safe because only this
5279                  * function is allowed to delete this snapshot
5280                  */
5281                 btrfs_unlock_up_safe(path, 0);
5282
5283                 level = btrfs_header_level(root->node);
5284                 while (1) {
5285                         btrfs_tree_lock(path->nodes[level]);
5286                         btrfs_set_lock_blocking(path->nodes[level]);
5287
5288                         ret = btrfs_lookup_extent_info(trans, root,
5289                                                 path->nodes[level]->start,
5290                                                 path->nodes[level]->len,
5291                                                 &wc->refs[level],
5292                                                 &wc->flags[level]);
5293                         BUG_ON(ret);
5294                         BUG_ON(wc->refs[level] == 0);
5295
5296                         if (level == root_item->drop_level)
5297                                 break;
5298
5299                         btrfs_tree_unlock(path->nodes[level]);
5300                         WARN_ON(wc->refs[level] != 1);
5301                         level--;
5302                 }
5303         }
5304
5305         wc->level = level;
5306         wc->shared_level = -1;
5307         wc->stage = DROP_REFERENCE;
5308         wc->update_ref = update_ref;
5309         wc->keep_locks = 0;
5310         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5311
5312         while (1) {
5313                 ret = walk_down_tree(trans, root, path, wc);
5314                 if (ret < 0) {
5315                         err = ret;
5316                         break;
5317                 }
5318
5319                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5320                 if (ret < 0) {
5321                         err = ret;
5322                         break;
5323                 }
5324
5325                 if (ret > 0) {
5326                         BUG_ON(wc->stage != DROP_REFERENCE);
5327                         break;
5328                 }
5329
5330                 if (wc->stage == DROP_REFERENCE) {
5331                         level = wc->level;
5332                         btrfs_node_key(path->nodes[level],
5333                                        &root_item->drop_progress,
5334                                        path->slots[level]);
5335                         root_item->drop_level = level;
5336                 }
5337
5338                 BUG_ON(wc->level == 0);
5339                 if (trans->transaction->in_commit ||
5340                     trans->transaction->delayed_refs.flushing) {
5341                         ret = btrfs_update_root(trans, tree_root,
5342                                                 &root->root_key,
5343                                                 root_item);
5344                         BUG_ON(ret);
5345
5346                         btrfs_end_transaction(trans, tree_root);
5347                         trans = btrfs_start_transaction(tree_root, 1);
5348                 } else {
5349                         unsigned long update;
5350                         update = trans->delayed_ref_updates;
5351                         trans->delayed_ref_updates = 0;
5352                         if (update)
5353                                 btrfs_run_delayed_refs(trans, tree_root,
5354                                                        update);
5355                 }
5356         }
5357         btrfs_release_path(root, path);
5358         BUG_ON(err);
5359
5360         ret = btrfs_del_root(trans, tree_root, &root->root_key);
5361         BUG_ON(ret);
5362
5363         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5364                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5365                                            NULL, NULL);
5366                 BUG_ON(ret < 0);
5367                 if (ret > 0) {
5368                         ret = btrfs_del_orphan_item(trans, tree_root,
5369                                                     root->root_key.objectid);
5370                         BUG_ON(ret);
5371                 }
5372         }
5373
5374         if (root->in_radix) {
5375                 btrfs_free_fs_root(tree_root->fs_info, root);
5376         } else {
5377                 free_extent_buffer(root->node);
5378                 free_extent_buffer(root->commit_root);
5379                 kfree(root);
5380         }
5381 out:
5382         btrfs_end_transaction(trans, tree_root);
5383         kfree(wc);
5384         btrfs_free_path(path);
5385         return err;
5386 }
5387
5388 /*
5389  * drop subtree rooted at tree block 'node'.
5390  *
5391  * NOTE: this function will unlock and release tree block 'node'
5392  */
5393 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5394                         struct btrfs_root *root,
5395                         struct extent_buffer *node,
5396                         struct extent_buffer *parent)
5397 {
5398         struct btrfs_path *path;
5399         struct walk_control *wc;
5400         int level;
5401         int parent_level;
5402         int ret = 0;
5403         int wret;
5404
5405         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5406
5407         path = btrfs_alloc_path();
5408         BUG_ON(!path);
5409
5410         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5411         BUG_ON(!wc);
5412
5413         btrfs_assert_tree_locked(parent);
5414         parent_level = btrfs_header_level(parent);
5415         extent_buffer_get(parent);
5416         path->nodes[parent_level] = parent;
5417         path->slots[parent_level] = btrfs_header_nritems(parent);
5418
5419         btrfs_assert_tree_locked(node);
5420         level = btrfs_header_level(node);
5421         path->nodes[level] = node;
5422         path->slots[level] = 0;
5423         path->locks[level] = 1;
5424
5425         wc->refs[parent_level] = 1;
5426         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5427         wc->level = level;
5428         wc->shared_level = -1;
5429         wc->stage = DROP_REFERENCE;
5430         wc->update_ref = 0;
5431         wc->keep_locks = 1;
5432         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5433
5434         while (1) {
5435                 wret = walk_down_tree(trans, root, path, wc);
5436                 if (wret < 0) {
5437                         ret = wret;
5438                         break;
5439                 }
5440
5441                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5442                 if (wret < 0)
5443                         ret = wret;
5444                 if (wret != 0)
5445                         break;
5446         }
5447
5448         kfree(wc);
5449         btrfs_free_path(path);
5450         return ret;
5451 }
5452
5453 #if 0
5454 static unsigned long calc_ra(unsigned long start, unsigned long last,
5455                              unsigned long nr)
5456 {
5457         return min(last, start + nr - 1);
5458 }
5459
5460 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5461                                          u64 len)
5462 {
5463         u64 page_start;
5464         u64 page_end;
5465         unsigned long first_index;
5466         unsigned long last_index;
5467         unsigned long i;
5468         struct page *page;
5469         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5470         struct file_ra_state *ra;
5471         struct btrfs_ordered_extent *ordered;
5472         unsigned int total_read = 0;
5473         unsigned int total_dirty = 0;
5474         int ret = 0;
5475
5476         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5477
5478         mutex_lock(&inode->i_mutex);
5479         first_index = start >> PAGE_CACHE_SHIFT;
5480         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5481
5482         /* make sure the dirty trick played by the caller work */
5483         ret = invalidate_inode_pages2_range(inode->i_mapping,
5484                                             first_index, last_index);
5485         if (ret)
5486                 goto out_unlock;
5487
5488         file_ra_state_init(ra, inode->i_mapping);
5489
5490         for (i = first_index ; i <= last_index; i++) {
5491                 if (total_read % ra->ra_pages == 0) {
5492                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5493                                        calc_ra(i, last_index, ra->ra_pages));
5494                 }
5495                 total_read++;
5496 again:
5497                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5498                         BUG_ON(1);
5499                 page = grab_cache_page(inode->i_mapping, i);
5500                 if (!page) {
5501                         ret = -ENOMEM;
5502                         goto out_unlock;
5503                 }
5504                 if (!PageUptodate(page)) {
5505                         btrfs_readpage(NULL, page);
5506                         lock_page(page);
5507                         if (!PageUptodate(page)) {
5508                                 unlock_page(page);
5509                                 page_cache_release(page);
5510                                 ret = -EIO;
5511                                 goto out_unlock;
5512                         }
5513                 }
5514                 wait_on_page_writeback(page);
5515
5516                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5517                 page_end = page_start + PAGE_CACHE_SIZE - 1;
5518                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5519
5520                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5521                 if (ordered) {
5522                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5523                         unlock_page(page);
5524                         page_cache_release(page);
5525                         btrfs_start_ordered_extent(inode, ordered, 1);
5526                         btrfs_put_ordered_extent(ordered);
5527                         goto again;
5528                 }
5529                 set_page_extent_mapped(page);
5530
5531                 if (i == first_index)
5532                         set_extent_bits(io_tree, page_start, page_end,
5533                                         EXTENT_BOUNDARY, GFP_NOFS);
5534                 btrfs_set_extent_delalloc(inode, page_start, page_end);
5535
5536                 set_page_dirty(page);
5537                 total_dirty++;
5538
5539                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5540                 unlock_page(page);
5541                 page_cache_release(page);
5542         }
5543
5544 out_unlock:
5545         kfree(ra);
5546         mutex_unlock(&inode->i_mutex);
5547         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5548         return ret;
5549 }
5550
5551 static noinline int relocate_data_extent(struct inode *reloc_inode,
5552                                          struct btrfs_key *extent_key,
5553                                          u64 offset)
5554 {
5555         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5556         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5557         struct extent_map *em;
5558         u64 start = extent_key->objectid - offset;
5559         u64 end = start + extent_key->offset - 1;
5560
5561         em = alloc_extent_map(GFP_NOFS);
5562         BUG_ON(!em || IS_ERR(em));
5563
5564         em->start = start;
5565         em->len = extent_key->offset;
5566         em->block_len = extent_key->offset;
5567         em->block_start = extent_key->objectid;
5568         em->bdev = root->fs_info->fs_devices->latest_bdev;
5569         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5570
5571         /* setup extent map to cheat btrfs_readpage */
5572         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5573         while (1) {
5574                 int ret;
5575                 write_lock(&em_tree->lock);
5576                 ret = add_extent_mapping(em_tree, em);
5577                 write_unlock(&em_tree->lock);
5578                 if (ret != -EEXIST) {
5579                         free_extent_map(em);
5580                         break;
5581                 }
5582                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5583         }
5584         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5585
5586         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5587 }
5588
5589 struct btrfs_ref_path {
5590         u64 extent_start;
5591         u64 nodes[BTRFS_MAX_LEVEL];
5592         u64 root_objectid;
5593         u64 root_generation;
5594         u64 owner_objectid;
5595         u32 num_refs;
5596         int lowest_level;
5597         int current_level;
5598         int shared_level;
5599
5600         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5601         u64 new_nodes[BTRFS_MAX_LEVEL];
5602 };
5603
5604 struct disk_extent {
5605         u64 ram_bytes;
5606         u64 disk_bytenr;
5607         u64 disk_num_bytes;
5608         u64 offset;
5609         u64 num_bytes;
5610         u8 compression;
5611         u8 encryption;
5612         u16 other_encoding;
5613 };
5614
5615 static int is_cowonly_root(u64 root_objectid)
5616 {
5617         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5618             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5619             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5620             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5621             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5622             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5623                 return 1;
5624         return 0;
5625 }
5626
5627 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5628                                     struct btrfs_root *extent_root,
5629                                     struct btrfs_ref_path *ref_path,
5630                                     int first_time)
5631 {
5632         struct extent_buffer *leaf;
5633         struct btrfs_path *path;
5634         struct btrfs_extent_ref *ref;
5635         struct btrfs_key key;
5636         struct btrfs_key found_key;
5637         u64 bytenr;
5638         u32 nritems;
5639         int level;
5640         int ret = 1;
5641
5642         path = btrfs_alloc_path();
5643         if (!path)
5644                 return -ENOMEM;
5645
5646         if (first_time) {
5647                 ref_path->lowest_level = -1;
5648                 ref_path->current_level = -1;
5649                 ref_path->shared_level = -1;
5650                 goto walk_up;
5651         }
5652 walk_down:
5653         level = ref_path->current_level - 1;
5654         while (level >= -1) {
5655                 u64 parent;
5656                 if (level < ref_path->lowest_level)
5657                         break;
5658
5659                 if (level >= 0)
5660                         bytenr = ref_path->nodes[level];
5661                 else
5662                         bytenr = ref_path->extent_start;
5663                 BUG_ON(bytenr == 0);
5664
5665                 parent = ref_path->nodes[level + 1];
5666                 ref_path->nodes[level + 1] = 0;
5667                 ref_path->current_level = level;
5668                 BUG_ON(parent == 0);
5669
5670                 key.objectid = bytenr;
5671                 key.offset = parent + 1;
5672                 key.type = BTRFS_EXTENT_REF_KEY;
5673
5674                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5675                 if (ret < 0)
5676                         goto out;
5677                 BUG_ON(ret == 0);
5678
5679                 leaf = path->nodes[0];
5680                 nritems = btrfs_header_nritems(leaf);
5681                 if (path->slots[0] >= nritems) {
5682                         ret = btrfs_next_leaf(extent_root, path);
5683                         if (ret < 0)
5684                                 goto out;
5685                         if (ret > 0)
5686                                 goto next;
5687                         leaf = path->nodes[0];
5688                 }
5689
5690                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5691                 if (found_key.objectid == bytenr &&
5692                     found_key.type == BTRFS_EXTENT_REF_KEY) {
5693                         if (level < ref_path->shared_level)
5694                                 ref_path->shared_level = level;
5695                         goto found;
5696                 }
5697 next:
5698                 level--;
5699                 btrfs_release_path(extent_root, path);
5700                 cond_resched();
5701         }
5702         /* reached lowest level */
5703         ret = 1;
5704         goto out;
5705 walk_up:
5706         level = ref_path->current_level;
5707         while (level < BTRFS_MAX_LEVEL - 1) {
5708                 u64 ref_objectid;
5709
5710                 if (level >= 0)
5711                         bytenr = ref_path->nodes[level];
5712                 else
5713                         bytenr = ref_path->extent_start;
5714
5715                 BUG_ON(bytenr == 0);
5716
5717                 key.objectid = bytenr;
5718                 key.offset = 0;
5719                 key.type = BTRFS_EXTENT_REF_KEY;
5720
5721                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5722                 if (ret < 0)
5723                         goto out;
5724
5725                 leaf = path->nodes[0];
5726                 nritems = btrfs_header_nritems(leaf);
5727                 if (path->slots[0] >= nritems) {
5728                         ret = btrfs_next_leaf(extent_root, path);
5729                         if (ret < 0)
5730                                 goto out;
5731                         if (ret > 0) {
5732                                 /* the extent was freed by someone */
5733                                 if (ref_path->lowest_level == level)
5734                                         goto out;
5735                                 btrfs_release_path(extent_root, path);
5736                                 goto walk_down;
5737                         }
5738                         leaf = path->nodes[0];
5739                 }
5740
5741                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5742                 if (found_key.objectid != bytenr ||
5743                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
5744                         /* the extent was freed by someone */
5745                         if (ref_path->lowest_level == level) {
5746                                 ret = 1;
5747                                 goto out;
5748                         }
5749                         btrfs_release_path(extent_root, path);
5750                         goto walk_down;
5751                 }
5752 found:
5753                 ref = btrfs_item_ptr(leaf, path->slots[0],
5754                                 struct btrfs_extent_ref);
5755                 ref_objectid = btrfs_ref_objectid(leaf, ref);
5756                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5757                         if (first_time) {
5758                                 level = (int)ref_objectid;
5759                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
5760                                 ref_path->lowest_level = level;
5761                                 ref_path->current_level = level;
5762                                 ref_path->nodes[level] = bytenr;
5763                         } else {
5764                                 WARN_ON(ref_objectid != level);
5765                         }
5766                 } else {
5767                         WARN_ON(level != -1);
5768                 }
5769                 first_time = 0;
5770
5771                 if (ref_path->lowest_level == level) {
5772                         ref_path->owner_objectid = ref_objectid;
5773                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5774                 }
5775
5776                 /*
5777                  * the block is tree root or the block isn't in reference
5778                  * counted tree.
5779                  */
5780                 if (found_key.objectid == found_key.offset ||
5781                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5782                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5783                         ref_path->root_generation =
5784                                 btrfs_ref_generation(leaf, ref);
5785                         if (level < 0) {
5786                                 /* special reference from the tree log */
5787                                 ref_path->nodes[0] = found_key.offset;
5788                                 ref_path->current_level = 0;
5789                         }
5790                         ret = 0;
5791                         goto out;
5792                 }
5793
5794                 level++;
5795                 BUG_ON(ref_path->nodes[level] != 0);
5796                 ref_path->nodes[level] = found_key.offset;
5797                 ref_path->current_level = level;
5798
5799                 /*
5800                  * the reference was created in the running transaction,
5801                  * no need to continue walking up.
5802                  */
5803                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
5804                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5805                         ref_path->root_generation =
5806                                 btrfs_ref_generation(leaf, ref);
5807                         ret = 0;
5808                         goto out;
5809                 }
5810
5811                 btrfs_release_path(extent_root, path);
5812                 cond_resched();
5813         }
5814         /* reached max tree level, but no tree root found. */
5815         BUG();
5816 out:
5817         btrfs_free_path(path);
5818         return ret;
5819 }
5820
5821 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
5822                                 struct btrfs_root *extent_root,
5823                                 struct btrfs_ref_path *ref_path,
5824                                 u64 extent_start)
5825 {
5826         memset(ref_path, 0, sizeof(*ref_path));
5827         ref_path->extent_start = extent_start;
5828
5829         return __next_ref_path(trans, extent_root, ref_path, 1);
5830 }
5831
5832 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
5833                                struct btrfs_root *extent_root,
5834                                struct btrfs_ref_path *ref_path)
5835 {
5836         return __next_ref_path(trans, extent_root, ref_path, 0);
5837 }
5838
5839 static noinline int get_new_locations(struct inode *reloc_inode,
5840                                       struct btrfs_key *extent_key,
5841                                       u64 offset, int no_fragment,
5842                                       struct disk_extent **extents,
5843                                       int *nr_extents)
5844 {
5845         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5846         struct btrfs_path *path;
5847         struct btrfs_file_extent_item *fi;
5848         struct extent_buffer *leaf;
5849         struct disk_extent *exts = *extents;
5850         struct btrfs_key found_key;
5851         u64 cur_pos;
5852         u64 last_byte;
5853         u32 nritems;
5854         int nr = 0;
5855         int max = *nr_extents;
5856         int ret;
5857
5858         WARN_ON(!no_fragment && *extents);
5859         if (!exts) {
5860                 max = 1;
5861                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
5862                 if (!exts)
5863                         return -ENOMEM;
5864         }
5865
5866         path = btrfs_alloc_path();
5867         BUG_ON(!path);
5868
5869         cur_pos = extent_key->objectid - offset;
5870         last_byte = extent_key->objectid + extent_key->offset;
5871         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
5872                                        cur_pos, 0);
5873         if (ret < 0)
5874                 goto out;
5875         if (ret > 0) {
5876                 ret = -ENOENT;
5877                 goto out;
5878         }
5879
5880         while (1) {
5881                 leaf = path->nodes[0];
5882                 nritems = btrfs_header_nritems(leaf);
5883                 if (path->slots[0] >= nritems) {
5884                         ret = btrfs_next_leaf(root, path);
5885                         if (ret < 0)
5886                                 goto out;
5887                         if (ret > 0)
5888                                 break;
5889                         leaf = path->nodes[0];
5890                 }
5891
5892                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5893                 if (found_key.offset != cur_pos ||
5894                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
5895                     found_key.objectid != reloc_inode->i_ino)
5896                         break;
5897
5898                 fi = btrfs_item_ptr(leaf, path->slots[0],
5899                                     struct btrfs_file_extent_item);
5900                 if (btrfs_file_extent_type(leaf, fi) !=
5901                     BTRFS_FILE_EXTENT_REG ||
5902                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
5903                         break;
5904
5905                 if (nr == max) {
5906                         struct disk_extent *old = exts;
5907                         max *= 2;
5908                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
5909                         memcpy(exts, old, sizeof(*exts) * nr);
5910                         if (old != *extents)
5911                                 kfree(old);
5912                 }
5913
5914                 exts[nr].disk_bytenr =
5915                         btrfs_file_extent_disk_bytenr(leaf, fi);
5916                 exts[nr].disk_num_bytes =
5917                         btrfs_file_extent_disk_num_bytes(leaf, fi);
5918                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
5919                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
5920                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
5921                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
5922                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
5923                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
5924                                                                            fi);
5925                 BUG_ON(exts[nr].offset > 0);
5926                 BUG_ON(exts[nr].compression || exts[nr].encryption);
5927                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
5928
5929                 cur_pos += exts[nr].num_bytes;
5930                 nr++;
5931
5932                 if (cur_pos + offset >= last_byte)
5933                         break;
5934
5935                 if (no_fragment) {
5936                         ret = 1;
5937                         goto out;
5938                 }
5939                 path->slots[0]++;
5940         }
5941
5942         BUG_ON(cur_pos + offset > last_byte);
5943         if (cur_pos + offset < last_byte) {
5944                 ret = -ENOENT;
5945                 goto out;
5946         }
5947         ret = 0;
5948 out:
5949         btrfs_free_path(path);
5950         if (ret) {
5951                 if (exts != *extents)
5952                         kfree(exts);
5953         } else {
5954                 *extents = exts;
5955                 *nr_extents = nr;
5956         }
5957         return ret;
5958 }
5959
5960 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
5961                                         struct btrfs_root *root,
5962                                         struct btrfs_path *path,
5963                                         struct btrfs_key *extent_key,
5964                                         struct btrfs_key *leaf_key,
5965                                         struct btrfs_ref_path *ref_path,
5966                                         struct disk_extent *new_extents,
5967                                         int nr_extents)
5968 {
5969         struct extent_buffer *leaf;
5970         struct btrfs_file_extent_item *fi;
5971         struct inode *inode = NULL;
5972         struct btrfs_key key;
5973         u64 lock_start = 0;
5974         u64 lock_end = 0;
5975         u64 num_bytes;
5976         u64 ext_offset;
5977         u64 search_end = (u64)-1;
5978         u32 nritems;
5979         int nr_scaned = 0;
5980         int extent_locked = 0;
5981         int extent_type;
5982         int ret;
5983
5984         memcpy(&key, leaf_key, sizeof(key));
5985         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
5986                 if (key.objectid < ref_path->owner_objectid ||
5987                     (key.objectid == ref_path->owner_objectid &&
5988                      key.type < BTRFS_EXTENT_DATA_KEY)) {
5989                         key.objectid = ref_path->owner_objectid;
5990                         key.type = BTRFS_EXTENT_DATA_KEY;
5991                         key.offset = 0;
5992                 }
5993         }
5994
5995         while (1) {
5996                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5997                 if (ret < 0)
5998                         goto out;
5999
6000                 leaf = path->nodes[0];
6001                 nritems = btrfs_header_nritems(leaf);
6002 next:
6003                 if (extent_locked && ret > 0) {
6004                         /*
6005                          * the file extent item was modified by someone
6006                          * before the extent got locked.
6007                          */
6008                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6009                                       lock_end, GFP_NOFS);
6010                         extent_locked = 0;
6011                 }
6012
6013                 if (path->slots[0] >= nritems) {
6014                         if (++nr_scaned > 2)
6015                                 break;
6016
6017                         BUG_ON(extent_locked);
6018                         ret = btrfs_next_leaf(root, path);
6019                         if (ret < 0)
6020                                 goto out;
6021                         if (ret > 0)
6022                                 break;
6023                         leaf = path->nodes[0];
6024                         nritems = btrfs_header_nritems(leaf);
6025                 }
6026
6027                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6028
6029                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6030                         if ((key.objectid > ref_path->owner_objectid) ||
6031                             (key.objectid == ref_path->owner_objectid &&
6032                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6033                             key.offset >= search_end)
6034                                 break;
6035                 }
6036
6037                 if (inode && key.objectid != inode->i_ino) {
6038                         BUG_ON(extent_locked);
6039                         btrfs_release_path(root, path);
6040                         mutex_unlock(&inode->i_mutex);
6041                         iput(inode);
6042                         inode = NULL;
6043                         continue;
6044                 }
6045
6046                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6047                         path->slots[0]++;
6048                         ret = 1;
6049                         goto next;
6050                 }
6051                 fi = btrfs_item_ptr(leaf, path->slots[0],
6052                                     struct btrfs_file_extent_item);
6053                 extent_type = btrfs_file_extent_type(leaf, fi);
6054                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6055                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6056                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6057                      extent_key->objectid)) {
6058                         path->slots[0]++;
6059                         ret = 1;
6060                         goto next;
6061                 }
6062
6063                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6064                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6065
6066                 if (search_end == (u64)-1) {
6067                         search_end = key.offset - ext_offset +
6068                                 btrfs_file_extent_ram_bytes(leaf, fi);
6069                 }
6070
6071                 if (!extent_locked) {
6072                         lock_start = key.offset;
6073                         lock_end = lock_start + num_bytes - 1;
6074                 } else {
6075                         if (lock_start > key.offset ||
6076                             lock_end + 1 < key.offset + num_bytes) {
6077                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6078                                               lock_start, lock_end, GFP_NOFS);
6079                                 extent_locked = 0;
6080                         }
6081                 }
6082
6083                 if (!inode) {
6084                         btrfs_release_path(root, path);
6085
6086                         inode = btrfs_iget_locked(root->fs_info->sb,
6087                                                   key.objectid, root);
6088                         if (inode->i_state & I_NEW) {
6089                                 BTRFS_I(inode)->root = root;
6090                                 BTRFS_I(inode)->location.objectid =
6091                                         key.objectid;
6092                                 BTRFS_I(inode)->location.type =
6093                                         BTRFS_INODE_ITEM_KEY;
6094                                 BTRFS_I(inode)->location.offset = 0;
6095                                 btrfs_read_locked_inode(inode);
6096                                 unlock_new_inode(inode);
6097                         }
6098                         /*
6099                          * some code call btrfs_commit_transaction while
6100                          * holding the i_mutex, so we can't use mutex_lock
6101                          * here.
6102                          */
6103                         if (is_bad_inode(inode) ||
6104                             !mutex_trylock(&inode->i_mutex)) {
6105                                 iput(inode);
6106                                 inode = NULL;
6107                                 key.offset = (u64)-1;
6108                                 goto skip;
6109                         }
6110                 }
6111
6112                 if (!extent_locked) {
6113                         struct btrfs_ordered_extent *ordered;
6114
6115                         btrfs_release_path(root, path);
6116
6117                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6118                                     lock_end, GFP_NOFS);
6119                         ordered = btrfs_lookup_first_ordered_extent(inode,
6120                                                                     lock_end);
6121                         if (ordered &&
6122                             ordered->file_offset <= lock_end &&
6123                             ordered->file_offset + ordered->len > lock_start) {
6124                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6125                                               lock_start, lock_end, GFP_NOFS);
6126                                 btrfs_start_ordered_extent(inode, ordered, 1);
6127                                 btrfs_put_ordered_extent(ordered);
6128                                 key.offset += num_bytes;
6129                                 goto skip;
6130                         }
6131                         if (ordered)
6132                                 btrfs_put_ordered_extent(ordered);
6133
6134                         extent_locked = 1;
6135                         continue;
6136                 }
6137
6138                 if (nr_extents == 1) {
6139                         /* update extent pointer in place */
6140                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6141                                                 new_extents[0].disk_bytenr);
6142                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6143                                                 new_extents[0].disk_num_bytes);
6144                         btrfs_mark_buffer_dirty(leaf);
6145
6146                         btrfs_drop_extent_cache(inode, key.offset,
6147                                                 key.offset + num_bytes - 1, 0);
6148
6149                         ret = btrfs_inc_extent_ref(trans, root,
6150                                                 new_extents[0].disk_bytenr,
6151                                                 new_extents[0].disk_num_bytes,
6152                                                 leaf->start,
6153                                                 root->root_key.objectid,
6154                                                 trans->transid,
6155                                                 key.objectid);
6156                         BUG_ON(ret);
6157
6158                         ret = btrfs_free_extent(trans, root,
6159                                                 extent_key->objectid,
6160                                                 extent_key->offset,
6161                                                 leaf->start,
6162                                                 btrfs_header_owner(leaf),
6163                                                 btrfs_header_generation(leaf),
6164                                                 key.objectid, 0);
6165                         BUG_ON(ret);
6166
6167                         btrfs_release_path(root, path);
6168                         key.offset += num_bytes;
6169                 } else {
6170                         BUG_ON(1);
6171 #if 0
6172                         u64 alloc_hint;
6173                         u64 extent_len;
6174                         int i;
6175                         /*
6176                          * drop old extent pointer at first, then insert the
6177                          * new pointers one bye one
6178                          */
6179                         btrfs_release_path(root, path);
6180                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
6181                                                  key.offset + num_bytes,
6182                                                  key.offset, &alloc_hint);
6183                         BUG_ON(ret);
6184
6185                         for (i = 0; i < nr_extents; i++) {
6186                                 if (ext_offset >= new_extents[i].num_bytes) {
6187                                         ext_offset -= new_extents[i].num_bytes;
6188                                         continue;
6189                                 }
6190                                 extent_len = min(new_extents[i].num_bytes -
6191                                                  ext_offset, num_bytes);
6192
6193                                 ret = btrfs_insert_empty_item(trans, root,
6194                                                               path, &key,
6195                                                               sizeof(*fi));
6196                                 BUG_ON(ret);
6197
6198                                 leaf = path->nodes[0];
6199                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6200                                                 struct btrfs_file_extent_item);
6201                                 btrfs_set_file_extent_generation(leaf, fi,
6202                                                         trans->transid);
6203                                 btrfs_set_file_extent_type(leaf, fi,
6204                                                         BTRFS_FILE_EXTENT_REG);
6205                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6206                                                 new_extents[i].disk_bytenr);
6207                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6208                                                 new_extents[i].disk_num_bytes);
6209                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6210                                                 new_extents[i].ram_bytes);
6211
6212                                 btrfs_set_file_extent_compression(leaf, fi,
6213                                                 new_extents[i].compression);
6214                                 btrfs_set_file_extent_encryption(leaf, fi,
6215                                                 new_extents[i].encryption);
6216                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6217                                                 new_extents[i].other_encoding);
6218
6219                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6220                                                         extent_len);
6221                                 ext_offset += new_extents[i].offset;
6222                                 btrfs_set_file_extent_offset(leaf, fi,
6223                                                         ext_offset);
6224                                 btrfs_mark_buffer_dirty(leaf);
6225
6226                                 btrfs_drop_extent_cache(inode, key.offset,
6227                                                 key.offset + extent_len - 1, 0);
6228
6229                                 ret = btrfs_inc_extent_ref(trans, root,
6230                                                 new_extents[i].disk_bytenr,
6231                                                 new_extents[i].disk_num_bytes,
6232                                                 leaf->start,
6233                                                 root->root_key.objectid,
6234                                                 trans->transid, key.objectid);
6235                                 BUG_ON(ret);
6236                                 btrfs_release_path(root, path);
6237
6238                                 inode_add_bytes(inode, extent_len);
6239
6240                                 ext_offset = 0;
6241                                 num_bytes -= extent_len;
6242                                 key.offset += extent_len;
6243
6244                                 if (num_bytes == 0)
6245                                         break;
6246                         }
6247                         BUG_ON(i >= nr_extents);
6248 #endif
6249                 }
6250
6251                 if (extent_locked) {
6252                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6253                                       lock_end, GFP_NOFS);
6254                         extent_locked = 0;
6255                 }
6256 skip:
6257                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6258                     key.offset >= search_end)
6259                         break;
6260
6261                 cond_resched();
6262         }
6263         ret = 0;
6264 out:
6265         btrfs_release_path(root, path);
6266         if (inode) {
6267                 mutex_unlock(&inode->i_mutex);
6268                 if (extent_locked) {
6269                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6270                                       lock_end, GFP_NOFS);
6271                 }
6272                 iput(inode);
6273         }
6274         return ret;
6275 }
6276
6277 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6278                                struct btrfs_root *root,
6279                                struct extent_buffer *buf, u64 orig_start)
6280 {
6281         int level;
6282         int ret;
6283
6284         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6285         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6286
6287         level = btrfs_header_level(buf);
6288         if (level == 0) {
6289                 struct btrfs_leaf_ref *ref;
6290                 struct btrfs_leaf_ref *orig_ref;
6291
6292                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6293                 if (!orig_ref)
6294                         return -ENOENT;
6295
6296                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6297                 if (!ref) {
6298                         btrfs_free_leaf_ref(root, orig_ref);
6299                         return -ENOMEM;
6300                 }
6301
6302                 ref->nritems = orig_ref->nritems;
6303                 memcpy(ref->extents, orig_ref->extents,
6304                         sizeof(ref->extents[0]) * ref->nritems);
6305
6306                 btrfs_free_leaf_ref(root, orig_ref);
6307
6308                 ref->root_gen = trans->transid;
6309                 ref->bytenr = buf->start;
6310                 ref->owner = btrfs_header_owner(buf);
6311                 ref->generation = btrfs_header_generation(buf);
6312
6313                 ret = btrfs_add_leaf_ref(root, ref, 0);
6314                 WARN_ON(ret);
6315                 btrfs_free_leaf_ref(root, ref);
6316         }
6317         return 0;
6318 }
6319
6320 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6321                                         struct extent_buffer *leaf,
6322                                         struct btrfs_block_group_cache *group,
6323                                         struct btrfs_root *target_root)
6324 {
6325         struct btrfs_key key;
6326         struct inode *inode = NULL;
6327         struct btrfs_file_extent_item *fi;
6328         u64 num_bytes;
6329         u64 skip_objectid = 0;
6330         u32 nritems;
6331         u32 i;
6332
6333         nritems = btrfs_header_nritems(leaf);
6334         for (i = 0; i < nritems; i++) {
6335                 btrfs_item_key_to_cpu(leaf, &key, i);
6336                 if (key.objectid == skip_objectid ||
6337                     key.type != BTRFS_EXTENT_DATA_KEY)
6338                         continue;
6339                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6340                 if (btrfs_file_extent_type(leaf, fi) ==
6341                     BTRFS_FILE_EXTENT_INLINE)
6342                         continue;
6343                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6344                         continue;
6345                 if (!inode || inode->i_ino != key.objectid) {
6346                         iput(inode);
6347                         inode = btrfs_ilookup(target_root->fs_info->sb,
6348                                               key.objectid, target_root, 1);
6349                 }
6350                 if (!inode) {
6351                         skip_objectid = key.objectid;
6352                         continue;
6353                 }
6354                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6355
6356                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6357                             key.offset + num_bytes - 1, GFP_NOFS);
6358                 btrfs_drop_extent_cache(inode, key.offset,
6359                                         key.offset + num_bytes - 1, 1);
6360                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6361                               key.offset + num_bytes - 1, GFP_NOFS);
6362                 cond_resched();
6363         }
6364         iput(inode);
6365         return 0;
6366 }
6367
6368 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6369                                         struct btrfs_root *root,
6370                                         struct extent_buffer *leaf,
6371                                         struct btrfs_block_group_cache *group,
6372                                         struct inode *reloc_inode)
6373 {
6374         struct btrfs_key key;
6375         struct btrfs_key extent_key;
6376         struct btrfs_file_extent_item *fi;
6377         struct btrfs_leaf_ref *ref;
6378         struct disk_extent *new_extent;
6379         u64 bytenr;
6380         u64 num_bytes;
6381         u32 nritems;
6382         u32 i;
6383         int ext_index;
6384         int nr_extent;
6385         int ret;
6386
6387         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6388         BUG_ON(!new_extent);
6389
6390         ref = btrfs_lookup_leaf_ref(root, leaf->start);
6391         BUG_ON(!ref);
6392
6393         ext_index = -1;
6394         nritems = btrfs_header_nritems(leaf);
6395         for (i = 0; i < nritems; i++) {
6396                 btrfs_item_key_to_cpu(leaf, &key, i);
6397                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6398                         continue;
6399                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6400                 if (btrfs_file_extent_type(leaf, fi) ==
6401                     BTRFS_FILE_EXTENT_INLINE)
6402                         continue;
6403                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6404                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6405                 if (bytenr == 0)
6406                         continue;
6407
6408                 ext_index++;
6409                 if (bytenr >= group->key.objectid + group->key.offset ||
6410                     bytenr + num_bytes <= group->key.objectid)
6411                         continue;
6412
6413                 extent_key.objectid = bytenr;
6414                 extent_key.offset = num_bytes;
6415                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6416                 nr_extent = 1;
6417                 ret = get_new_locations(reloc_inode, &extent_key,
6418                                         group->key.objectid, 1,
6419                                         &new_extent, &nr_extent);
6420                 if (ret > 0)
6421                         continue;
6422                 BUG_ON(ret < 0);
6423
6424                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6425                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6426                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6427                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6428
6429                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6430                                                 new_extent->disk_bytenr);
6431                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6432                                                 new_extent->disk_num_bytes);
6433                 btrfs_mark_buffer_dirty(leaf);
6434
6435                 ret = btrfs_inc_extent_ref(trans, root,
6436                                         new_extent->disk_bytenr,
6437                                         new_extent->disk_num_bytes,
6438                                         leaf->start,
6439                                         root->root_key.objectid,
6440                                         trans->transid, key.objectid);
6441                 BUG_ON(ret);
6442
6443                 ret = btrfs_free_extent(trans, root,
6444                                         bytenr, num_bytes, leaf->start,
6445                                         btrfs_header_owner(leaf),
6446                                         btrfs_header_generation(leaf),
6447                                         key.objectid, 0);
6448                 BUG_ON(ret);
6449                 cond_resched();
6450         }
6451         kfree(new_extent);
6452         BUG_ON(ext_index + 1 != ref->nritems);
6453         btrfs_free_leaf_ref(root, ref);
6454         return 0;
6455 }
6456
6457 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6458                           struct btrfs_root *root)
6459 {
6460         struct btrfs_root *reloc_root;
6461         int ret;
6462
6463         if (root->reloc_root) {
6464                 reloc_root = root->reloc_root;
6465                 root->reloc_root = NULL;
6466                 list_add(&reloc_root->dead_list,
6467                          &root->fs_info->dead_reloc_roots);
6468
6469                 btrfs_set_root_bytenr(&reloc_root->root_item,
6470                                       reloc_root->node->start);
6471                 btrfs_set_root_level(&root->root_item,
6472                                      btrfs_header_level(reloc_root->node));
6473                 memset(&reloc_root->root_item.drop_progress, 0,
6474                         sizeof(struct btrfs_disk_key));
6475                 reloc_root->root_item.drop_level = 0;
6476
6477                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6478                                         &reloc_root->root_key,
6479                                         &reloc_root->root_item);
6480                 BUG_ON(ret);
6481         }
6482         return 0;
6483 }
6484
6485 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6486 {
6487         struct btrfs_trans_handle *trans;
6488         struct btrfs_root *reloc_root;
6489         struct btrfs_root *prev_root = NULL;
6490         struct list_head dead_roots;
6491         int ret;
6492         unsigned long nr;
6493
6494         INIT_LIST_HEAD(&dead_roots);
6495         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6496
6497         while (!list_empty(&dead_roots)) {
6498                 reloc_root = list_entry(dead_roots.prev,
6499                                         struct btrfs_root, dead_list);
6500                 list_del_init(&reloc_root->dead_list);
6501
6502                 BUG_ON(reloc_root->commit_root != NULL);
6503                 while (1) {
6504                         trans = btrfs_join_transaction(root, 1);
6505                         BUG_ON(!trans);
6506
6507                         mutex_lock(&root->fs_info->drop_mutex);
6508                         ret = btrfs_drop_snapshot(trans, reloc_root);
6509                         if (ret != -EAGAIN)
6510                                 break;
6511                         mutex_unlock(&root->fs_info->drop_mutex);
6512
6513                         nr = trans->blocks_used;
6514                         ret = btrfs_end_transaction(trans, root);
6515                         BUG_ON(ret);
6516                         btrfs_btree_balance_dirty(root, nr);
6517                 }
6518
6519                 free_extent_buffer(reloc_root->node);
6520
6521                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6522                                      &reloc_root->root_key);
6523                 BUG_ON(ret);
6524                 mutex_unlock(&root->fs_info->drop_mutex);
6525
6526                 nr = trans->blocks_used;
6527                 ret = btrfs_end_transaction(trans, root);
6528                 BUG_ON(ret);
6529                 btrfs_btree_balance_dirty(root, nr);
6530
6531                 kfree(prev_root);
6532                 prev_root = reloc_root;
6533         }
6534         if (prev_root) {
6535                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6536                 kfree(prev_root);
6537         }
6538         return 0;
6539 }
6540
6541 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6542 {
6543         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6544         return 0;
6545 }
6546
6547 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6548 {
6549         struct btrfs_root *reloc_root;
6550         struct btrfs_trans_handle *trans;
6551         struct btrfs_key location;
6552         int found;
6553         int ret;
6554
6555         mutex_lock(&root->fs_info->tree_reloc_mutex);
6556         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6557         BUG_ON(ret);
6558         found = !list_empty(&root->fs_info->dead_reloc_roots);
6559         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6560
6561         if (found) {
6562                 trans = btrfs_start_transaction(root, 1);
6563                 BUG_ON(!trans);
6564                 ret = btrfs_commit_transaction(trans, root);
6565                 BUG_ON(ret);
6566         }
6567
6568         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6569         location.offset = (u64)-1;
6570         location.type = BTRFS_ROOT_ITEM_KEY;
6571
6572         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6573         BUG_ON(!reloc_root);
6574         btrfs_orphan_cleanup(reloc_root);
6575         return 0;
6576 }
6577
6578 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6579                                     struct btrfs_root *root)
6580 {
6581         struct btrfs_root *reloc_root;
6582         struct extent_buffer *eb;
6583         struct btrfs_root_item *root_item;
6584         struct btrfs_key root_key;
6585         int ret;
6586
6587         BUG_ON(!root->ref_cows);
6588         if (root->reloc_root)
6589                 return 0;
6590
6591         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6592         BUG_ON(!root_item);
6593
6594         ret = btrfs_copy_root(trans, root, root->commit_root,
6595                               &eb, BTRFS_TREE_RELOC_OBJECTID);
6596         BUG_ON(ret);
6597
6598         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6599         root_key.offset = root->root_key.objectid;
6600         root_key.type = BTRFS_ROOT_ITEM_KEY;
6601
6602         memcpy(root_item, &root->root_item, sizeof(root_item));
6603         btrfs_set_root_refs(root_item, 0);
6604         btrfs_set_root_bytenr(root_item, eb->start);
6605         btrfs_set_root_level(root_item, btrfs_header_level(eb));
6606         btrfs_set_root_generation(root_item, trans->transid);
6607
6608         btrfs_tree_unlock(eb);
6609         free_extent_buffer(eb);
6610
6611         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6612                                 &root_key, root_item);
6613         BUG_ON(ret);
6614         kfree(root_item);
6615
6616         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6617                                                  &root_key);
6618         BUG_ON(!reloc_root);
6619         reloc_root->last_trans = trans->transid;
6620         reloc_root->commit_root = NULL;
6621         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6622
6623         root->reloc_root = reloc_root;
6624         return 0;
6625 }
6626
6627 /*
6628  * Core function of space balance.
6629  *
6630  * The idea is using reloc trees to relocate tree blocks in reference
6631  * counted roots. There is one reloc tree for each subvol, and all
6632  * reloc trees share same root key objectid. Reloc trees are snapshots
6633  * of the latest committed roots of subvols (root->commit_root).
6634  *
6635  * To relocate a tree block referenced by a subvol, there are two steps.
6636  * COW the block through subvol's reloc tree, then update block pointer
6637  * in the subvol to point to the new block. Since all reloc trees share
6638  * same root key objectid, doing special handing for tree blocks owned
6639  * by them is easy. Once a tree block has been COWed in one reloc tree,
6640  * we can use the resulting new block directly when the same block is
6641  * required to COW again through other reloc trees. By this way, relocated
6642  * tree blocks are shared between reloc trees, so they are also shared
6643  * between subvols.
6644  */
6645 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6646                                       struct btrfs_root *root,
6647                                       struct btrfs_path *path,
6648                                       struct btrfs_key *first_key,
6649                                       struct btrfs_ref_path *ref_path,
6650                                       struct btrfs_block_group_cache *group,
6651                                       struct inode *reloc_inode)
6652 {
6653         struct btrfs_root *reloc_root;
6654         struct extent_buffer *eb = NULL;
6655         struct btrfs_key *keys;
6656         u64 *nodes;
6657         int level;
6658         int shared_level;
6659         int lowest_level = 0;
6660         int ret;
6661
6662         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6663                 lowest_level = ref_path->owner_objectid;
6664
6665         if (!root->ref_cows) {
6666                 path->lowest_level = lowest_level;
6667                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6668                 BUG_ON(ret < 0);
6669                 path->lowest_level = 0;
6670                 btrfs_release_path(root, path);
6671                 return 0;
6672         }
6673
6674         mutex_lock(&root->fs_info->tree_reloc_mutex);
6675         ret = init_reloc_tree(trans, root);
6676         BUG_ON(ret);
6677         reloc_root = root->reloc_root;
6678
6679         shared_level = ref_path->shared_level;
6680         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6681
6682         keys = ref_path->node_keys;
6683         nodes = ref_path->new_nodes;
6684         memset(&keys[shared_level + 1], 0,
6685                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6686         memset(&nodes[shared_level + 1], 0,
6687                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6688
6689         if (nodes[lowest_level] == 0) {
6690                 path->lowest_level = lowest_level;
6691                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6692                                         0, 1);
6693                 BUG_ON(ret);
6694                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6695                         eb = path->nodes[level];
6696                         if (!eb || eb == reloc_root->node)
6697                                 break;
6698                         nodes[level] = eb->start;
6699                         if (level == 0)
6700                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6701                         else
6702                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6703                 }
6704                 if (nodes[0] &&
6705                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6706                         eb = path->nodes[0];
6707                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
6708                                                       group, reloc_inode);
6709                         BUG_ON(ret);
6710                 }
6711                 btrfs_release_path(reloc_root, path);
6712         } else {
6713                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6714                                        lowest_level);
6715                 BUG_ON(ret);
6716         }
6717
6718         /*
6719          * replace tree blocks in the fs tree with tree blocks in
6720          * the reloc tree.
6721          */
6722         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6723         BUG_ON(ret < 0);
6724
6725         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6726                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6727                                         0, 0);
6728                 BUG_ON(ret);
6729                 extent_buffer_get(path->nodes[0]);
6730                 eb = path->nodes[0];
6731                 btrfs_release_path(reloc_root, path);
6732                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6733                 BUG_ON(ret);
6734                 free_extent_buffer(eb);
6735         }
6736
6737         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6738         path->lowest_level = 0;
6739         return 0;
6740 }
6741
6742 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6743                                         struct btrfs_root *root,
6744                                         struct btrfs_path *path,
6745                                         struct btrfs_key *first_key,
6746                                         struct btrfs_ref_path *ref_path)
6747 {
6748         int ret;
6749
6750         ret = relocate_one_path(trans, root, path, first_key,
6751                                 ref_path, NULL, NULL);
6752         BUG_ON(ret);
6753
6754         return 0;
6755 }
6756
6757 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6758                                     struct btrfs_root *extent_root,
6759                                     struct btrfs_path *path,
6760                                     struct btrfs_key *extent_key)
6761 {
6762         int ret;
6763
6764         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6765         if (ret)
6766                 goto out;
6767         ret = btrfs_del_item(trans, extent_root, path);
6768 out:
6769         btrfs_release_path(extent_root, path);
6770         return ret;
6771 }
6772
6773 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6774                                                 struct btrfs_ref_path *ref_path)
6775 {
6776         struct btrfs_key root_key;
6777
6778         root_key.objectid = ref_path->root_objectid;
6779         root_key.type = BTRFS_ROOT_ITEM_KEY;
6780         if (is_cowonly_root(ref_path->root_objectid))
6781                 root_key.offset = 0;
6782         else
6783                 root_key.offset = (u64)-1;
6784
6785         return btrfs_read_fs_root_no_name(fs_info, &root_key);
6786 }
6787
6788 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6789                                         struct btrfs_path *path,
6790                                         struct btrfs_key *extent_key,
6791                                         struct btrfs_block_group_cache *group,
6792                                         struct inode *reloc_inode, int pass)
6793 {
6794         struct btrfs_trans_handle *trans;
6795         struct btrfs_root *found_root;
6796         struct btrfs_ref_path *ref_path = NULL;
6797         struct disk_extent *new_extents = NULL;
6798         int nr_extents = 0;
6799         int loops;
6800         int ret;
6801         int level;
6802         struct btrfs_key first_key;
6803         u64 prev_block = 0;
6804
6805
6806         trans = btrfs_start_transaction(extent_root, 1);
6807         BUG_ON(!trans);
6808
6809         if (extent_key->objectid == 0) {
6810                 ret = del_extent_zero(trans, extent_root, path, extent_key);
6811                 goto out;
6812         }
6813
6814         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
6815         if (!ref_path) {
6816                 ret = -ENOMEM;
6817                 goto out;
6818         }
6819
6820         for (loops = 0; ; loops++) {
6821                 if (loops == 0) {
6822                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
6823                                                    extent_key->objectid);
6824                 } else {
6825                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
6826                 }
6827                 if (ret < 0)
6828                         goto out;
6829                 if (ret > 0)
6830                         break;
6831
6832                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6833                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
6834                         continue;
6835
6836                 found_root = read_ref_root(extent_root->fs_info, ref_path);
6837                 BUG_ON(!found_root);
6838                 /*
6839                  * for reference counted tree, only process reference paths
6840                  * rooted at the latest committed root.
6841                  */
6842                 if (found_root->ref_cows &&
6843                     ref_path->root_generation != found_root->root_key.offset)
6844                         continue;
6845
6846                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6847                         if (pass == 0) {
6848                                 /*
6849                                  * copy data extents to new locations
6850                                  */
6851                                 u64 group_start = group->key.objectid;
6852                                 ret = relocate_data_extent(reloc_inode,
6853                                                            extent_key,
6854                                                            group_start);
6855                                 if (ret < 0)
6856                                         goto out;
6857                                 break;
6858                         }
6859                         level = 0;
6860                 } else {
6861                         level = ref_path->owner_objectid;
6862                 }
6863
6864                 if (prev_block != ref_path->nodes[level]) {
6865                         struct extent_buffer *eb;
6866                         u64 block_start = ref_path->nodes[level];
6867                         u64 block_size = btrfs_level_size(found_root, level);
6868
6869                         eb = read_tree_block(found_root, block_start,
6870                                              block_size, 0);
6871                         btrfs_tree_lock(eb);
6872                         BUG_ON(level != btrfs_header_level(eb));
6873
6874                         if (level == 0)
6875                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
6876                         else
6877                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
6878
6879                         btrfs_tree_unlock(eb);
6880                         free_extent_buffer(eb);
6881                         prev_block = block_start;
6882                 }
6883
6884                 mutex_lock(&extent_root->fs_info->trans_mutex);
6885                 btrfs_record_root_in_trans(found_root);
6886                 mutex_unlock(&extent_root->fs_info->trans_mutex);
6887                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6888                         /*
6889                          * try to update data extent references while
6890                          * keeping metadata shared between snapshots.
6891                          */
6892                         if (pass == 1) {
6893                                 ret = relocate_one_path(trans, found_root,
6894                                                 path, &first_key, ref_path,
6895                                                 group, reloc_inode);
6896                                 if (ret < 0)
6897                                         goto out;
6898                                 continue;
6899                         }
6900                         /*
6901                          * use fallback method to process the remaining
6902                          * references.
6903                          */
6904                         if (!new_extents) {
6905                                 u64 group_start = group->key.objectid;
6906                                 new_extents = kmalloc(sizeof(*new_extents),
6907                                                       GFP_NOFS);
6908                                 nr_extents = 1;
6909                                 ret = get_new_locations(reloc_inode,
6910                                                         extent_key,
6911                                                         group_start, 1,
6912                                                         &new_extents,
6913                                                         &nr_extents);
6914                                 if (ret)
6915                                         goto out;
6916                         }
6917                         ret = replace_one_extent(trans, found_root,
6918                                                 path, extent_key,
6919                                                 &first_key, ref_path,
6920                                                 new_extents, nr_extents);
6921                 } else {
6922                         ret = relocate_tree_block(trans, found_root, path,
6923                                                   &first_key, ref_path);
6924                 }
6925                 if (ret < 0)
6926                         goto out;
6927         }
6928         ret = 0;
6929 out:
6930         btrfs_end_transaction(trans, extent_root);
6931         kfree(new_extents);
6932         kfree(ref_path);
6933         return ret;
6934 }
6935 #endif
6936
6937 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
6938 {
6939         u64 num_devices;
6940         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
6941                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
6942
6943         num_devices = root->fs_info->fs_devices->rw_devices;
6944         if (num_devices == 1) {
6945                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6946                 stripped = flags & ~stripped;
6947
6948                 /* turn raid0 into single device chunks */
6949                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
6950                         return stripped;
6951
6952                 /* turn mirroring into duplication */
6953                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
6954                              BTRFS_BLOCK_GROUP_RAID10))
6955                         return stripped | BTRFS_BLOCK_GROUP_DUP;
6956                 return flags;
6957         } else {
6958                 /* they already had raid on here, just return */
6959                 if (flags & stripped)
6960                         return flags;
6961
6962                 stripped |= BTRFS_BLOCK_GROUP_DUP;
6963                 stripped = flags & ~stripped;
6964
6965                 /* switch duplicated blocks with raid1 */
6966                 if (flags & BTRFS_BLOCK_GROUP_DUP)
6967                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
6968
6969                 /* turn single device chunks into raid0 */
6970                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
6971         }
6972         return flags;
6973 }
6974
6975 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
6976                      struct btrfs_block_group_cache *shrink_block_group,
6977                      int force)
6978 {
6979         struct btrfs_trans_handle *trans;
6980         u64 new_alloc_flags;
6981         u64 calc;
6982
6983         spin_lock(&shrink_block_group->lock);
6984         if (btrfs_block_group_used(&shrink_block_group->item) +
6985             shrink_block_group->reserved > 0) {
6986                 spin_unlock(&shrink_block_group->lock);
6987
6988                 trans = btrfs_start_transaction(root, 1);
6989                 spin_lock(&shrink_block_group->lock);
6990
6991                 new_alloc_flags = update_block_group_flags(root,
6992                                                    shrink_block_group->flags);
6993                 if (new_alloc_flags != shrink_block_group->flags) {
6994                         calc =
6995                              btrfs_block_group_used(&shrink_block_group->item);
6996                 } else {
6997                         calc = shrink_block_group->key.offset;
6998                 }
6999                 spin_unlock(&shrink_block_group->lock);
7000
7001                 do_chunk_alloc(trans, root->fs_info->extent_root,
7002                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
7003
7004                 btrfs_end_transaction(trans, root);
7005         } else
7006                 spin_unlock(&shrink_block_group->lock);
7007         return 0;
7008 }
7009
7010
7011 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
7012                                          struct btrfs_block_group_cache *group)
7013
7014 {
7015         __alloc_chunk_for_shrink(root, group, 1);
7016         set_block_group_readonly(group);
7017         return 0;
7018 }
7019
7020 /*
7021  * checks to see if its even possible to relocate this block group.
7022  *
7023  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7024  * ok to go ahead and try.
7025  */
7026 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7027 {
7028         struct btrfs_block_group_cache *block_group;
7029         struct btrfs_space_info *space_info;
7030         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7031         struct btrfs_device *device;
7032         int full = 0;
7033         int ret = 0;
7034
7035         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7036
7037         /* odd, couldn't find the block group, leave it alone */
7038         if (!block_group)
7039                 return -1;
7040
7041         /* no bytes used, we're good */
7042         if (!btrfs_block_group_used(&block_group->item))
7043                 goto out;
7044
7045         space_info = block_group->space_info;
7046         spin_lock(&space_info->lock);
7047
7048         full = space_info->full;
7049
7050         /*
7051          * if this is the last block group we have in this space, we can't
7052          * relocate it unless we're able to allocate a new chunk below.
7053          *
7054          * Otherwise, we need to make sure we have room in the space to handle
7055          * all of the extents from this block group.  If we can, we're good
7056          */
7057         if ((space_info->total_bytes != block_group->key.offset) &&
7058            (space_info->bytes_used + space_info->bytes_reserved +
7059             space_info->bytes_pinned + space_info->bytes_readonly +
7060             btrfs_block_group_used(&block_group->item) <
7061             space_info->total_bytes)) {
7062                 spin_unlock(&space_info->lock);
7063                 goto out;
7064         }
7065         spin_unlock(&space_info->lock);
7066
7067         /*
7068          * ok we don't have enough space, but maybe we have free space on our
7069          * devices to allocate new chunks for relocation, so loop through our
7070          * alloc devices and guess if we have enough space.  However, if we
7071          * were marked as full, then we know there aren't enough chunks, and we
7072          * can just return.
7073          */
7074         ret = -1;
7075         if (full)
7076                 goto out;
7077
7078         mutex_lock(&root->fs_info->chunk_mutex);
7079         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7080                 u64 min_free = btrfs_block_group_used(&block_group->item);
7081                 u64 dev_offset, max_avail;
7082
7083                 /*
7084                  * check to make sure we can actually find a chunk with enough
7085                  * space to fit our block group in.
7086                  */
7087                 if (device->total_bytes > device->bytes_used + min_free) {
7088                         ret = find_free_dev_extent(NULL, device, min_free,
7089                                                    &dev_offset, &max_avail);
7090                         if (!ret)
7091                                 break;
7092                         ret = -1;
7093                 }
7094         }
7095         mutex_unlock(&root->fs_info->chunk_mutex);
7096 out:
7097         btrfs_put_block_group(block_group);
7098         return ret;
7099 }
7100
7101 static int find_first_block_group(struct btrfs_root *root,
7102                 struct btrfs_path *path, struct btrfs_key *key)
7103 {
7104         int ret = 0;
7105         struct btrfs_key found_key;
7106         struct extent_buffer *leaf;
7107         int slot;
7108
7109         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7110         if (ret < 0)
7111                 goto out;
7112
7113         while (1) {
7114                 slot = path->slots[0];
7115                 leaf = path->nodes[0];
7116                 if (slot >= btrfs_header_nritems(leaf)) {
7117                         ret = btrfs_next_leaf(root, path);
7118                         if (ret == 0)
7119                                 continue;
7120                         if (ret < 0)
7121                                 goto out;
7122                         break;
7123                 }
7124                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7125
7126                 if (found_key.objectid >= key->objectid &&
7127                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7128                         ret = 0;
7129                         goto out;
7130                 }
7131                 path->slots[0]++;
7132         }
7133         ret = -ENOENT;
7134 out:
7135         return ret;
7136 }
7137
7138 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7139 {
7140         struct btrfs_block_group_cache *block_group;
7141         struct btrfs_space_info *space_info;
7142         struct btrfs_caching_control *caching_ctl;
7143         struct rb_node *n;
7144
7145         down_write(&info->extent_commit_sem);
7146         while (!list_empty(&info->caching_block_groups)) {
7147                 caching_ctl = list_entry(info->caching_block_groups.next,
7148                                          struct btrfs_caching_control, list);
7149                 list_del(&caching_ctl->list);
7150                 put_caching_control(caching_ctl);
7151         }
7152         up_write(&info->extent_commit_sem);
7153
7154         spin_lock(&info->block_group_cache_lock);
7155         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7156                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7157                                        cache_node);
7158                 rb_erase(&block_group->cache_node,
7159                          &info->block_group_cache_tree);
7160                 spin_unlock(&info->block_group_cache_lock);
7161
7162                 down_write(&block_group->space_info->groups_sem);
7163                 list_del(&block_group->list);
7164                 up_write(&block_group->space_info->groups_sem);
7165
7166                 if (block_group->cached == BTRFS_CACHE_STARTED)
7167                         wait_block_group_cache_done(block_group);
7168
7169                 btrfs_remove_free_space_cache(block_group);
7170
7171                 WARN_ON(atomic_read(&block_group->count) != 1);
7172                 kfree(block_group);
7173
7174                 spin_lock(&info->block_group_cache_lock);
7175         }
7176         spin_unlock(&info->block_group_cache_lock);
7177
7178         /* now that all the block groups are freed, go through and
7179          * free all the space_info structs.  This is only called during
7180          * the final stages of unmount, and so we know nobody is
7181          * using them.  We call synchronize_rcu() once before we start,
7182          * just to be on the safe side.
7183          */
7184         synchronize_rcu();
7185
7186         while(!list_empty(&info->space_info)) {
7187                 space_info = list_entry(info->space_info.next,
7188                                         struct btrfs_space_info,
7189                                         list);
7190
7191                 list_del(&space_info->list);
7192                 kfree(space_info);
7193         }
7194         return 0;
7195 }
7196
7197 int btrfs_read_block_groups(struct btrfs_root *root)
7198 {
7199         struct btrfs_path *path;
7200         int ret;
7201         struct btrfs_block_group_cache *cache;
7202         struct btrfs_fs_info *info = root->fs_info;
7203         struct btrfs_space_info *space_info;
7204         struct btrfs_key key;
7205         struct btrfs_key found_key;
7206         struct extent_buffer *leaf;
7207
7208         root = info->extent_root;
7209         key.objectid = 0;
7210         key.offset = 0;
7211         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7212         path = btrfs_alloc_path();
7213         if (!path)
7214                 return -ENOMEM;
7215
7216         while (1) {
7217                 ret = find_first_block_group(root, path, &key);
7218                 if (ret > 0) {
7219                         ret = 0;
7220                         goto error;
7221                 }
7222                 if (ret != 0)
7223                         goto error;
7224
7225                 leaf = path->nodes[0];
7226                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7227                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7228                 if (!cache) {
7229                         ret = -ENOMEM;
7230                         break;
7231                 }
7232
7233                 atomic_set(&cache->count, 1);
7234                 spin_lock_init(&cache->lock);
7235                 spin_lock_init(&cache->tree_lock);
7236                 cache->fs_info = info;
7237                 INIT_LIST_HEAD(&cache->list);
7238                 INIT_LIST_HEAD(&cache->cluster_list);
7239
7240                 /*
7241                  * we only want to have 32k of ram per block group for keeping
7242                  * track of free space, and if we pass 1/2 of that we want to
7243                  * start converting things over to using bitmaps
7244                  */
7245                 cache->extents_thresh = ((1024 * 32) / 2) /
7246                         sizeof(struct btrfs_free_space);
7247
7248                 read_extent_buffer(leaf, &cache->item,
7249                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7250                                    sizeof(cache->item));
7251                 memcpy(&cache->key, &found_key, sizeof(found_key));
7252
7253                 key.objectid = found_key.objectid + found_key.offset;
7254                 btrfs_release_path(root, path);
7255                 cache->flags = btrfs_block_group_flags(&cache->item);
7256                 cache->sectorsize = root->sectorsize;
7257
7258                 /*
7259                  * check for two cases, either we are full, and therefore
7260                  * don't need to bother with the caching work since we won't
7261                  * find any space, or we are empty, and we can just add all
7262                  * the space in and be done with it.  This saves us _alot_ of
7263                  * time, particularly in the full case.
7264                  */
7265                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7266                         exclude_super_stripes(root, cache);
7267                         cache->last_byte_to_unpin = (u64)-1;
7268                         cache->cached = BTRFS_CACHE_FINISHED;
7269                         free_excluded_extents(root, cache);
7270                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7271                         exclude_super_stripes(root, cache);
7272                         cache->last_byte_to_unpin = (u64)-1;
7273                         cache->cached = BTRFS_CACHE_FINISHED;
7274                         add_new_free_space(cache, root->fs_info,
7275                                            found_key.objectid,
7276                                            found_key.objectid +
7277                                            found_key.offset);
7278                         free_excluded_extents(root, cache);
7279                 }
7280
7281                 ret = update_space_info(info, cache->flags, found_key.offset,
7282                                         btrfs_block_group_used(&cache->item),
7283                                         &space_info);
7284                 BUG_ON(ret);
7285                 cache->space_info = space_info;
7286                 spin_lock(&cache->space_info->lock);
7287                 cache->space_info->bytes_super += cache->bytes_super;
7288                 spin_unlock(&cache->space_info->lock);
7289
7290                 down_write(&space_info->groups_sem);
7291                 list_add_tail(&cache->list, &space_info->block_groups);
7292                 up_write(&space_info->groups_sem);
7293
7294                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7295                 BUG_ON(ret);
7296
7297                 set_avail_alloc_bits(root->fs_info, cache->flags);
7298                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7299                         set_block_group_readonly(cache);
7300         }
7301         ret = 0;
7302 error:
7303         btrfs_free_path(path);
7304         return ret;
7305 }
7306
7307 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7308                            struct btrfs_root *root, u64 bytes_used,
7309                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7310                            u64 size)
7311 {
7312         int ret;
7313         struct btrfs_root *extent_root;
7314         struct btrfs_block_group_cache *cache;
7315
7316         extent_root = root->fs_info->extent_root;
7317
7318         root->fs_info->last_trans_log_full_commit = trans->transid;
7319
7320         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7321         if (!cache)
7322                 return -ENOMEM;
7323
7324         cache->key.objectid = chunk_offset;
7325         cache->key.offset = size;
7326         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7327         cache->sectorsize = root->sectorsize;
7328
7329         /*
7330          * we only want to have 32k of ram per block group for keeping track
7331          * of free space, and if we pass 1/2 of that we want to start
7332          * converting things over to using bitmaps
7333          */
7334         cache->extents_thresh = ((1024 * 32) / 2) /
7335                 sizeof(struct btrfs_free_space);
7336         atomic_set(&cache->count, 1);
7337         spin_lock_init(&cache->lock);
7338         spin_lock_init(&cache->tree_lock);
7339         INIT_LIST_HEAD(&cache->list);
7340         INIT_LIST_HEAD(&cache->cluster_list);
7341
7342         btrfs_set_block_group_used(&cache->item, bytes_used);
7343         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7344         cache->flags = type;
7345         btrfs_set_block_group_flags(&cache->item, type);
7346
7347         cache->last_byte_to_unpin = (u64)-1;
7348         cache->cached = BTRFS_CACHE_FINISHED;
7349         exclude_super_stripes(root, cache);
7350
7351         add_new_free_space(cache, root->fs_info, chunk_offset,
7352                            chunk_offset + size);
7353
7354         free_excluded_extents(root, cache);
7355
7356         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7357                                 &cache->space_info);
7358         BUG_ON(ret);
7359
7360         spin_lock(&cache->space_info->lock);
7361         cache->space_info->bytes_super += cache->bytes_super;
7362         spin_unlock(&cache->space_info->lock);
7363
7364         down_write(&cache->space_info->groups_sem);
7365         list_add_tail(&cache->list, &cache->space_info->block_groups);
7366         up_write(&cache->space_info->groups_sem);
7367
7368         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7369         BUG_ON(ret);
7370
7371         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7372                                 sizeof(cache->item));
7373         BUG_ON(ret);
7374
7375         set_avail_alloc_bits(extent_root->fs_info, type);
7376
7377         return 0;
7378 }
7379
7380 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7381                              struct btrfs_root *root, u64 group_start)
7382 {
7383         struct btrfs_path *path;
7384         struct btrfs_block_group_cache *block_group;
7385         struct btrfs_free_cluster *cluster;
7386         struct btrfs_key key;
7387         int ret;
7388
7389         root = root->fs_info->extent_root;
7390
7391         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7392         BUG_ON(!block_group);
7393         BUG_ON(!block_group->ro);
7394
7395         memcpy(&key, &block_group->key, sizeof(key));
7396
7397         /* make sure this block group isn't part of an allocation cluster */
7398         cluster = &root->fs_info->data_alloc_cluster;
7399         spin_lock(&cluster->refill_lock);
7400         btrfs_return_cluster_to_free_space(block_group, cluster);
7401         spin_unlock(&cluster->refill_lock);
7402
7403         /*
7404          * make sure this block group isn't part of a metadata
7405          * allocation cluster
7406          */
7407         cluster = &root->fs_info->meta_alloc_cluster;
7408         spin_lock(&cluster->refill_lock);
7409         btrfs_return_cluster_to_free_space(block_group, cluster);
7410         spin_unlock(&cluster->refill_lock);
7411
7412         path = btrfs_alloc_path();
7413         BUG_ON(!path);
7414
7415         spin_lock(&root->fs_info->block_group_cache_lock);
7416         rb_erase(&block_group->cache_node,
7417                  &root->fs_info->block_group_cache_tree);
7418         spin_unlock(&root->fs_info->block_group_cache_lock);
7419
7420         down_write(&block_group->space_info->groups_sem);
7421         /*
7422          * we must use list_del_init so people can check to see if they
7423          * are still on the list after taking the semaphore
7424          */
7425         list_del_init(&block_group->list);
7426         up_write(&block_group->space_info->groups_sem);
7427
7428         if (block_group->cached == BTRFS_CACHE_STARTED)
7429                 wait_block_group_cache_done(block_group);
7430
7431         btrfs_remove_free_space_cache(block_group);
7432
7433         spin_lock(&block_group->space_info->lock);
7434         block_group->space_info->total_bytes -= block_group->key.offset;
7435         block_group->space_info->bytes_readonly -= block_group->key.offset;
7436         spin_unlock(&block_group->space_info->lock);
7437
7438         btrfs_clear_space_info_full(root->fs_info);
7439
7440         btrfs_put_block_group(block_group);
7441         btrfs_put_block_group(block_group);
7442
7443         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7444         if (ret > 0)
7445                 ret = -EIO;
7446         if (ret < 0)
7447                 goto out;
7448
7449         ret = btrfs_del_item(trans, root, path);
7450 out:
7451         btrfs_free_path(path);
7452         return ret;
7453 }