fs/btrfs/extent-tree.c

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