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->disk_total += total_bytes * factor;
2767                 found->bytes_used += bytes_used;
2768                 found->disk_used += bytes_used * factor;
2769                 found->full = 0;
2770                 spin_unlock(&found->lock);
2771                 *space_info = found;
2772                 return 0;
2773         }
2774         found = kzalloc(sizeof(*found), GFP_NOFS);
2775         if (!found)
2776                 return -ENOMEM;
2777
2778         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
2779                 INIT_LIST_HEAD(&found->block_groups[i]);
2780         init_rwsem(&found->groups_sem);
2781         spin_lock_init(&found->lock);
2782         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
2783                                 BTRFS_BLOCK_GROUP_SYSTEM |
2784                                 BTRFS_BLOCK_GROUP_METADATA);
2785         found->total_bytes = total_bytes;
2786         found->disk_total = total_bytes * factor;
2787         found->bytes_used = bytes_used;
2788         found->disk_used = bytes_used * factor;
2789         found->bytes_pinned = 0;
2790         found->bytes_reserved = 0;
2791         found->bytes_readonly = 0;
2792         found->bytes_may_use = 0;
2793         found->full = 0;
2794         found->force_alloc = 0;
2795         *space_info = found;
2796         list_add_rcu(&found->list, &info->space_info);
2797         atomic_set(&found->caching_threads, 0);
2798         return 0;
2799 }
2800
2801 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2802 {
2803         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2804                                    BTRFS_BLOCK_GROUP_RAID1 |
2805                                    BTRFS_BLOCK_GROUP_RAID10 |
2806                                    BTRFS_BLOCK_GROUP_DUP);
2807         if (extra_flags) {
2808                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2809                         fs_info->avail_data_alloc_bits |= extra_flags;
2810                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2811                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2812                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2813                         fs_info->avail_system_alloc_bits |= extra_flags;
2814         }
2815 }
2816
2817 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2818 {
2819         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2820
2821         if (num_devices == 1)
2822                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2823         if (num_devices < 4)
2824                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2825
2826         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2827             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2828                       BTRFS_BLOCK_GROUP_RAID10))) {
2829                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2830         }
2831
2832         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2833             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2834                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2835         }
2836
2837         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2838             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2839              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2840              (flags & BTRFS_BLOCK_GROUP_DUP)))
2841                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2842         return flags;
2843 }
2844
2845 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
2846 {
2847         if (flags & BTRFS_BLOCK_GROUP_DATA)
2848                 flags |= root->fs_info->avail_data_alloc_bits &
2849                          root->fs_info->data_alloc_profile;
2850         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2851                 flags |= root->fs_info->avail_system_alloc_bits &
2852                          root->fs_info->system_alloc_profile;
2853         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
2854                 flags |= root->fs_info->avail_metadata_alloc_bits &
2855                          root->fs_info->metadata_alloc_profile;
2856         return btrfs_reduce_alloc_profile(root, flags);
2857 }
2858
2859 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
2860 {
2861         u64 flags;
2862
2863         if (data)
2864                 flags = BTRFS_BLOCK_GROUP_DATA;
2865         else if (root == root->fs_info->chunk_root)
2866                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2867         else
2868                 flags = BTRFS_BLOCK_GROUP_METADATA;
2869
2870         return get_alloc_profile(root, flags);
2871 }
2872
2873 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2874 {
2875         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2876                                                        BTRFS_BLOCK_GROUP_DATA);
2877 }
2878
2879 /*
2880  * This will check the space that the inode allocates from to make sure we have
2881  * enough space for bytes.
2882  */
2883 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
2884 {
2885         struct btrfs_space_info *data_sinfo;
2886         struct btrfs_root *root = BTRFS_I(inode)->root;
2887         u64 used;
2888         int ret = 0, committed = 0;
2889
2890         /* make sure bytes are sectorsize aligned */
2891         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2892
2893         data_sinfo = BTRFS_I(inode)->space_info;
2894         if (!data_sinfo)
2895                 goto alloc;
2896
2897 again:
2898         /* make sure we have enough space to handle the data first */
2899         spin_lock(&data_sinfo->lock);
2900         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
2901                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
2902                 data_sinfo->bytes_may_use;
2903
2904         if (used + bytes > data_sinfo->total_bytes) {
2905                 struct btrfs_trans_handle *trans;
2906
2907                 /*
2908                  * if we don't have enough free bytes in this space then we need
2909                  * to alloc a new chunk.
2910                  */
2911                 if (!data_sinfo->full) {
2912                         u64 alloc_target;
2913
2914                         data_sinfo->force_alloc = 1;
2915                         spin_unlock(&data_sinfo->lock);
2916 alloc:
2917                         alloc_target = btrfs_get_alloc_profile(root, 1);
2918                         trans = btrfs_join_transaction(root, 1);
2919                         if (IS_ERR(trans))
2920                                 return PTR_ERR(trans);
2921
2922                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2923                                              bytes + 2 * 1024 * 1024,
2924                                              alloc_target, 0);
2925                         btrfs_end_transaction(trans, root);
2926                         if (ret < 0)
2927                                 return ret;
2928
2929                         if (!data_sinfo) {
2930                                 btrfs_set_inode_space_info(root, inode);
2931                                 data_sinfo = BTRFS_I(inode)->space_info;
2932                         }
2933                         goto again;
2934                 }
2935                 spin_unlock(&data_sinfo->lock);
2936
2937                 /* commit the current transaction and try again */
2938                 if (!committed && !root->fs_info->open_ioctl_trans) {
2939                         committed = 1;
2940                         trans = btrfs_join_transaction(root, 1);
2941                         if (IS_ERR(trans))
2942                                 return PTR_ERR(trans);
2943                         ret = btrfs_commit_transaction(trans, root);
2944                         if (ret)
2945                                 return ret;
2946                         goto again;
2947                 }
2948
2949 #if 0 /* I hope we never need this code again, just in case */
2950                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
2951                        "%llu bytes_reserved, " "%llu bytes_pinned, "
2952                        "%llu bytes_readonly, %llu may use %llu total\n",
2953                        (unsigned long long)bytes,
2954                        (unsigned long long)data_sinfo->bytes_used,
2955                        (unsigned long long)data_sinfo->bytes_reserved,
2956                        (unsigned long long)data_sinfo->bytes_pinned,
2957                        (unsigned long long)data_sinfo->bytes_readonly,
2958                        (unsigned long long)data_sinfo->bytes_may_use,
2959                        (unsigned long long)data_sinfo->total_bytes);
2960 #endif
2961                 return -ENOSPC;
2962         }
2963         data_sinfo->bytes_may_use += bytes;
2964         BTRFS_I(inode)->reserved_bytes += bytes;
2965         spin_unlock(&data_sinfo->lock);
2966
2967         return 0;
2968 }
2969
2970 /*
2971  * called when we are clearing an delalloc extent from the
2972  * inode's io_tree or there was an error for whatever reason
2973  * after calling btrfs_check_data_free_space
2974  */
2975 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
2976 {
2977         struct btrfs_root *root = BTRFS_I(inode)->root;
2978         struct btrfs_space_info *data_sinfo;
2979
2980         /* make sure bytes are sectorsize aligned */
2981         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2982
2983         data_sinfo = BTRFS_I(inode)->space_info;
2984         spin_lock(&data_sinfo->lock);
2985         data_sinfo->bytes_may_use -= bytes;
2986         BTRFS_I(inode)->reserved_bytes -= bytes;
2987         spin_unlock(&data_sinfo->lock);
2988 }
2989
2990 static void force_metadata_allocation(struct btrfs_fs_info *info)
2991 {
2992         struct list_head *head = &info->space_info;
2993         struct btrfs_space_info *found;
2994
2995         rcu_read_lock();
2996         list_for_each_entry_rcu(found, head, list) {
2997                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2998                         found->force_alloc = 1;
2999         }
3000         rcu_read_unlock();
3001 }
3002
3003 static int should_alloc_chunk(struct btrfs_space_info *sinfo,
3004                               u64 alloc_bytes)
3005 {
3006         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3007
3008         if (sinfo->bytes_used + sinfo->bytes_reserved +
3009             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3010                 return 0;
3011
3012         if (sinfo->bytes_used + sinfo->bytes_reserved +
3013             alloc_bytes < div_factor(num_bytes, 8))
3014                 return 0;
3015
3016         return 1;
3017 }
3018
3019 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3020                           struct btrfs_root *extent_root, u64 alloc_bytes,
3021                           u64 flags, int force)
3022 {
3023         struct btrfs_space_info *space_info;
3024         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3025         int ret = 0;
3026
3027         mutex_lock(&fs_info->chunk_mutex);
3028
3029         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3030
3031         space_info = __find_space_info(extent_root->fs_info, flags);
3032         if (!space_info) {
3033                 ret = update_space_info(extent_root->fs_info, flags,
3034                                         0, 0, &space_info);
3035                 BUG_ON(ret);
3036         }
3037         BUG_ON(!space_info);
3038
3039         spin_lock(&space_info->lock);
3040         if (space_info->force_alloc)
3041                 force = 1;
3042         if (space_info->full) {
3043                 spin_unlock(&space_info->lock);
3044                 goto out;
3045         }
3046
3047         if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3048                 spin_unlock(&space_info->lock);
3049                 goto out;
3050         }
3051         spin_unlock(&space_info->lock);
3052
3053         /*
3054          * if we're doing a data chunk, go ahead and make sure that
3055          * we keep a reasonable number of metadata chunks allocated in the
3056          * FS as well.
3057          */
3058         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3059                 fs_info->data_chunk_allocations++;
3060                 if (!(fs_info->data_chunk_allocations %
3061                       fs_info->metadata_ratio))
3062                         force_metadata_allocation(fs_info);
3063         }
3064
3065         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3066         spin_lock(&space_info->lock);
3067         if (ret)
3068                 space_info->full = 1;
3069         else
3070                 ret = 1;
3071         space_info->force_alloc = 0;
3072         spin_unlock(&space_info->lock);
3073 out:
3074         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3075         return ret;
3076 }
3077
3078 static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
3079                                 struct btrfs_root *root,
3080                                 struct btrfs_space_info *sinfo, u64 num_bytes)
3081 {
3082         int ret;
3083         int end_trans = 0;
3084
3085         if (sinfo->full)
3086                 return 0;
3087
3088         spin_lock(&sinfo->lock);
3089         ret = should_alloc_chunk(sinfo, num_bytes + 2 * 1024 * 1024);
3090         spin_unlock(&sinfo->lock);
3091         if (!ret)
3092                 return 0;
3093
3094         if (!trans) {
3095                 trans = btrfs_join_transaction(root, 1);
3096                 BUG_ON(IS_ERR(trans));
3097                 end_trans = 1;
3098         }
3099
3100         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3101                              num_bytes + 2 * 1024 * 1024,
3102                              get_alloc_profile(root, sinfo->flags), 0);
3103
3104         if (end_trans)
3105                 btrfs_end_transaction(trans, root);
3106
3107         return ret == 1 ? 1 : 0;
3108 }
3109
3110 /*
3111  * shrink metadata reservation for delalloc
3112  */
3113 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3114                            struct btrfs_root *root, u64 to_reclaim)
3115 {
3116         struct btrfs_block_rsv *block_rsv;
3117         u64 reserved;
3118         u64 max_reclaim;
3119         u64 reclaimed = 0;
3120         int no_reclaim = 0;
3121         int pause = 1;
3122         int ret;
3123
3124         block_rsv = &root->fs_info->delalloc_block_rsv;
3125         spin_lock(&block_rsv->lock);
3126         reserved = block_rsv->reserved;
3127         spin_unlock(&block_rsv->lock);
3128
3129         if (reserved == 0)
3130                 return 0;
3131
3132         max_reclaim = min(reserved, to_reclaim);
3133
3134         while (1) {
3135                 ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0);
3136                 if (!ret) {
3137                         if (no_reclaim > 2)
3138                                 break;
3139                         no_reclaim++;
3140                         __set_current_state(TASK_INTERRUPTIBLE);
3141                         schedule_timeout(pause);
3142                         pause <<= 1;
3143                         if (pause > HZ / 10)
3144                                 pause = HZ / 10;
3145                 } else {
3146                         no_reclaim = 0;
3147                         pause = 1;
3148                 }
3149
3150                 spin_lock(&block_rsv->lock);
3151                 if (reserved > block_rsv->reserved)
3152                         reclaimed = reserved - block_rsv->reserved;
3153                 reserved = block_rsv->reserved;
3154                 spin_unlock(&block_rsv->lock);
3155
3156                 if (reserved == 0 || reclaimed >= max_reclaim)
3157                         break;
3158
3159                 if (trans && trans->transaction->blocked)
3160                         return -EAGAIN;
3161         }
3162         return reclaimed >= to_reclaim;
3163 }
3164
3165 static int should_retry_reserve(struct btrfs_trans_handle *trans,
3166                                 struct btrfs_root *root,
3167                                 struct btrfs_block_rsv *block_rsv,
3168                                 u64 num_bytes, int *retries)
3169 {
3170         struct btrfs_space_info *space_info = block_rsv->space_info;
3171         int ret;
3172
3173         if ((*retries) > 2)
3174                 return -ENOSPC;
3175
3176         ret = maybe_allocate_chunk(trans, root, space_info, num_bytes);
3177         if (ret)
3178                 return 1;
3179
3180         if (trans && trans->transaction->in_commit)
3181                 return -ENOSPC;
3182
3183         ret = shrink_delalloc(trans, root, num_bytes);
3184         if (ret)
3185                 return ret;
3186
3187         spin_lock(&space_info->lock);
3188         if (space_info->bytes_pinned < num_bytes)
3189                 ret = 1;
3190         spin_unlock(&space_info->lock);
3191         if (ret)
3192                 return -ENOSPC;
3193
3194         (*retries)++;
3195
3196         if (trans)
3197                 return -EAGAIN;
3198
3199         trans = btrfs_join_transaction(root, 1);
3200         BUG_ON(IS_ERR(trans));
3201         ret = btrfs_commit_transaction(trans, root);
3202         BUG_ON(ret);
3203
3204         return 1;
3205 }
3206
3207 static int reserve_metadata_bytes(struct btrfs_block_rsv *block_rsv,
3208                                   u64 num_bytes)
3209 {
3210         struct btrfs_space_info *space_info = block_rsv->space_info;
3211         u64 unused;
3212         int ret = -ENOSPC;
3213
3214         spin_lock(&space_info->lock);
3215         unused = space_info->bytes_used + space_info->bytes_reserved +
3216                  space_info->bytes_pinned + space_info->bytes_readonly +
3217                  space_info->bytes_may_use;
3218
3219         if (unused < space_info->total_bytes)
3220                 unused = space_info->total_bytes - unused;
3221         else
3222                 unused = 0;
3223
3224         if (unused >= num_bytes) {
3225                 if (block_rsv->priority >= 10) {
3226                         space_info->bytes_reserved += num_bytes;
3227                         ret = 0;
3228                 } else {
3229                         if ((unused + block_rsv->reserved) *
3230                             block_rsv->priority >=
3231                             (num_bytes + block_rsv->reserved) * 10) {
3232                                 space_info->bytes_reserved += num_bytes;
3233                                 ret = 0;
3234                         }
3235                 }
3236         }
3237         spin_unlock(&space_info->lock);
3238
3239         return ret;
3240 }
3241
3242 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3243                                              struct btrfs_root *root)
3244 {
3245         struct btrfs_block_rsv *block_rsv;
3246         if (root->ref_cows)
3247                 block_rsv = trans->block_rsv;
3248         else
3249                 block_rsv = root->block_rsv;
3250
3251         if (!block_rsv)
3252                 block_rsv = &root->fs_info->empty_block_rsv;
3253
3254         return block_rsv;
3255 }
3256
3257 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3258                                u64 num_bytes)
3259 {
3260         int ret = -ENOSPC;
3261         spin_lock(&block_rsv->lock);
3262         if (block_rsv->reserved >= num_bytes) {
3263                 block_rsv->reserved -= num_bytes;
3264                 if (block_rsv->reserved < block_rsv->size)
3265                         block_rsv->full = 0;
3266                 ret = 0;
3267         }
3268         spin_unlock(&block_rsv->lock);
3269         return ret;
3270 }
3271
3272 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3273                                 u64 num_bytes, int update_size)
3274 {
3275         spin_lock(&block_rsv->lock);
3276         block_rsv->reserved += num_bytes;
3277         if (update_size)
3278                 block_rsv->size += num_bytes;
3279         else if (block_rsv->reserved >= block_rsv->size)
3280                 block_rsv->full = 1;
3281         spin_unlock(&block_rsv->lock);
3282 }
3283
3284 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3285                              struct btrfs_block_rsv *dest, u64 num_bytes)
3286 {
3287         struct btrfs_space_info *space_info = block_rsv->space_info;
3288
3289         spin_lock(&block_rsv->lock);
3290         if (num_bytes == (u64)-1)
3291                 num_bytes = block_rsv->size;
3292         block_rsv->size -= num_bytes;
3293         if (block_rsv->reserved >= block_rsv->size) {
3294                 num_bytes = block_rsv->reserved - block_rsv->size;
3295                 block_rsv->reserved = block_rsv->size;
3296                 block_rsv->full = 1;
3297         } else {
3298                 num_bytes = 0;
3299         }
3300         spin_unlock(&block_rsv->lock);
3301
3302         if (num_bytes > 0) {
3303                 if (dest) {
3304                         block_rsv_add_bytes(dest, num_bytes, 0);
3305                 } else {
3306                         spin_lock(&space_info->lock);
3307                         space_info->bytes_reserved -= num_bytes;
3308                         spin_unlock(&space_info->lock);
3309                 }
3310         }
3311 }
3312
3313 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3314                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3315 {
3316         int ret;
3317
3318         ret = block_rsv_use_bytes(src, num_bytes);
3319         if (ret)
3320                 return ret;
3321
3322         block_rsv_add_bytes(dst, num_bytes, 1);
3323         return 0;
3324 }
3325
3326 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3327 {
3328         memset(rsv, 0, sizeof(*rsv));
3329         spin_lock_init(&rsv->lock);
3330         atomic_set(&rsv->usage, 1);
3331         rsv->priority = 6;
3332         INIT_LIST_HEAD(&rsv->list);
3333 }
3334
3335 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3336 {
3337         struct btrfs_block_rsv *block_rsv;
3338         struct btrfs_fs_info *fs_info = root->fs_info;
3339         u64 alloc_target;
3340
3341         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3342         if (!block_rsv)
3343                 return NULL;
3344
3345         btrfs_init_block_rsv(block_rsv);
3346
3347         alloc_target = btrfs_get_alloc_profile(root, 0);
3348         block_rsv->space_info = __find_space_info(fs_info,
3349                                                   BTRFS_BLOCK_GROUP_METADATA);
3350
3351         return block_rsv;
3352 }
3353
3354 void btrfs_free_block_rsv(struct btrfs_root *root,
3355                           struct btrfs_block_rsv *rsv)
3356 {
3357         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3358                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3359                 if (!rsv->durable)
3360                         kfree(rsv);
3361         }
3362 }
3363
3364 /*
3365  * make the block_rsv struct be able to capture freed space.
3366  * the captured space will re-add to the the block_rsv struct
3367  * after transaction commit
3368  */
3369 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3370                                  struct btrfs_block_rsv *block_rsv)
3371 {
3372         block_rsv->durable = 1;
3373         mutex_lock(&fs_info->durable_block_rsv_mutex);
3374         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3375         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3376 }
3377
3378 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3379                         struct btrfs_root *root,
3380                         struct btrfs_block_rsv *block_rsv,
3381                         u64 num_bytes, int *retries)
3382 {
3383         int ret;
3384
3385         if (num_bytes == 0)
3386                 return 0;
3387 again:
3388         ret = reserve_metadata_bytes(block_rsv, num_bytes);
3389         if (!ret) {
3390                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3391                 return 0;
3392         }
3393
3394         ret = should_retry_reserve(trans, root, block_rsv, num_bytes, retries);
3395         if (ret > 0)
3396                 goto again;
3397
3398         return ret;
3399 }
3400
3401 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3402                           struct btrfs_root *root,
3403                           struct btrfs_block_rsv *block_rsv,
3404                           u64 min_reserved, int min_factor)
3405 {
3406         u64 num_bytes = 0;
3407         int commit_trans = 0;
3408         int ret = -ENOSPC;
3409
3410         if (!block_rsv)
3411                 return 0;
3412
3413         spin_lock(&block_rsv->lock);
3414         if (min_factor > 0)
3415                 num_bytes = div_factor(block_rsv->size, min_factor);
3416         if (min_reserved > num_bytes)
3417                 num_bytes = min_reserved;
3418
3419         if (block_rsv->reserved >= num_bytes) {
3420                 ret = 0;
3421         } else {
3422                 num_bytes -= block_rsv->reserved;
3423                 if (block_rsv->durable &&
3424                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3425                         commit_trans = 1;
3426         }
3427         spin_unlock(&block_rsv->lock);
3428         if (!ret)
3429                 return 0;
3430
3431         if (block_rsv->refill_used) {
3432                 ret = reserve_metadata_bytes(block_rsv, num_bytes);
3433                 if (!ret) {
3434                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3435                         return 0;
3436                 }
3437         }
3438
3439         if (commit_trans) {
3440                 if (trans)
3441                         return -EAGAIN;
3442
3443                 trans = btrfs_join_transaction(root, 1);
3444                 BUG_ON(IS_ERR(trans));
3445                 ret = btrfs_commit_transaction(trans, root);
3446                 return 0;
3447         }
3448
3449         WARN_ON(1);
3450         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3451                 block_rsv->size, block_rsv->reserved,
3452                 block_rsv->freed[0], block_rsv->freed[1]);
3453
3454         return -ENOSPC;
3455 }
3456
3457 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3458                             struct btrfs_block_rsv *dst_rsv,
3459                             u64 num_bytes)
3460 {
3461         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3462 }
3463
3464 void btrfs_block_rsv_release(struct btrfs_root *root,
3465                              struct btrfs_block_rsv *block_rsv,
3466                              u64 num_bytes)
3467 {
3468         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3469         if (global_rsv->full || global_rsv == block_rsv ||
3470             block_rsv->space_info != global_rsv->space_info)
3471                 global_rsv = NULL;
3472         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3473 }
3474
3475 /*
3476  * helper to calculate size of global block reservation.
3477  * the desired value is sum of space used by extent tree,
3478  * checksum tree and root tree
3479  */
3480 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3481 {
3482         struct btrfs_space_info *sinfo;
3483         u64 num_bytes;
3484         u64 meta_used;
3485         u64 data_used;
3486         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3487 #if 0
3488         /*
3489          * per tree used space accounting can be inaccuracy, so we
3490          * can't rely on it.
3491          */
3492         spin_lock(&fs_info->extent_root->accounting_lock);
3493         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3494         spin_unlock(&fs_info->extent_root->accounting_lock);
3495
3496         spin_lock(&fs_info->csum_root->accounting_lock);
3497         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3498         spin_unlock(&fs_info->csum_root->accounting_lock);
3499
3500         spin_lock(&fs_info->tree_root->accounting_lock);
3501         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3502         spin_unlock(&fs_info->tree_root->accounting_lock);
3503 #endif
3504         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3505         spin_lock(&sinfo->lock);
3506         data_used = sinfo->bytes_used;
3507         spin_unlock(&sinfo->lock);
3508
3509         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3510         spin_lock(&sinfo->lock);
3511         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3512                 data_used = 0;
3513         meta_used = sinfo->bytes_used;
3514         spin_unlock(&sinfo->lock);
3515
3516         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3517                     csum_size * 2;
3518         num_bytes += div64_u64(data_used + meta_used, 50);
3519
3520         if (num_bytes * 3 > meta_used)
3521                 num_bytes = div64_u64(meta_used, 3);
3522
3523         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3524 }
3525
3526 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3527 {
3528         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3529         struct btrfs_space_info *sinfo = block_rsv->space_info;
3530         u64 num_bytes;
3531
3532         num_bytes = calc_global_metadata_size(fs_info);
3533
3534         spin_lock(&block_rsv->lock);
3535         spin_lock(&sinfo->lock);
3536
3537         block_rsv->size = num_bytes;
3538
3539         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3540                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3541                     sinfo->bytes_may_use;
3542
3543         if (sinfo->total_bytes > num_bytes) {
3544                 num_bytes = sinfo->total_bytes - num_bytes;
3545                 block_rsv->reserved += num_bytes;
3546                 sinfo->bytes_reserved += num_bytes;
3547         }
3548
3549         if (block_rsv->reserved >= block_rsv->size) {
3550                 num_bytes = block_rsv->reserved - block_rsv->size;
3551                 sinfo->bytes_reserved -= num_bytes;
3552                 block_rsv->reserved = block_rsv->size;
3553                 block_rsv->full = 1;
3554         }
3555 #if 0
3556         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3557                 block_rsv->size, block_rsv->reserved);
3558 #endif
3559         spin_unlock(&sinfo->lock);
3560         spin_unlock(&block_rsv->lock);
3561 }
3562
3563 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3564 {
3565         struct btrfs_space_info *space_info;
3566
3567         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3568         fs_info->chunk_block_rsv.space_info = space_info;
3569         fs_info->chunk_block_rsv.priority = 10;
3570
3571         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3572         fs_info->global_block_rsv.space_info = space_info;
3573         fs_info->global_block_rsv.priority = 10;
3574         fs_info->global_block_rsv.refill_used = 1;
3575         fs_info->delalloc_block_rsv.space_info = space_info;
3576         fs_info->trans_block_rsv.space_info = space_info;
3577         fs_info->empty_block_rsv.space_info = space_info;
3578         fs_info->empty_block_rsv.priority = 10;
3579
3580         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3581         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3582         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3583         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3584         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3585
3586         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3587
3588         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3589
3590         update_global_block_rsv(fs_info);
3591 }
3592
3593 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3594 {
3595         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3596         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3597         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3598         WARN_ON(fs_info->trans_block_rsv.size > 0);
3599         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3600         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3601         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3602 }
3603
3604 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3605 {
3606         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3607                 3 * num_items;
3608 }
3609
3610 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3611                                  struct btrfs_root *root,
3612                                  int num_items, int *retries)
3613 {
3614         u64 num_bytes;
3615         int ret;
3616
3617         if (num_items == 0 || root->fs_info->chunk_root == root)
3618                 return 0;
3619
3620         num_bytes = calc_trans_metadata_size(root, num_items);
3621         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3622                                   num_bytes, retries);
3623         if (!ret) {
3624                 trans->bytes_reserved += num_bytes;
3625                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3626         }
3627         return ret;
3628 }
3629
3630 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3631                                   struct btrfs_root *root)
3632 {
3633         if (!trans->bytes_reserved)
3634                 return;
3635
3636         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3637         btrfs_block_rsv_release(root, trans->block_rsv,
3638                                 trans->bytes_reserved);
3639         trans->bytes_reserved = 0;
3640 }
3641
3642 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3643                                   struct inode *inode)
3644 {
3645         struct btrfs_root *root = BTRFS_I(inode)->root;
3646         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3647         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3648
3649         /*
3650          * one for deleting orphan item, one for updating inode and
3651          * two for calling btrfs_truncate_inode_items.
3652          *
3653          * btrfs_truncate_inode_items is a delete operation, it frees
3654          * more space than it uses in most cases. So two units of
3655          * metadata space should be enough for calling it many times.
3656          * If all of the metadata space is used, we can commit
3657          * transaction and use space it freed.
3658          */
3659         u64 num_bytes = calc_trans_metadata_size(root, 4);
3660         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3661 }
3662
3663 void btrfs_orphan_release_metadata(struct inode *inode)
3664 {
3665         struct btrfs_root *root = BTRFS_I(inode)->root;
3666         u64 num_bytes = calc_trans_metadata_size(root, 4);
3667         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3668 }
3669
3670 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3671                                 struct btrfs_pending_snapshot *pending)
3672 {
3673         struct btrfs_root *root = pending->root;
3674         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3675         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3676         /*
3677          * two for root back/forward refs, two for directory entries
3678          * and one for root of the snapshot.
3679          */
3680         u64 num_bytes = calc_trans_metadata_size(root, 5);
3681         dst_rsv->space_info = src_rsv->space_info;
3682         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3683 }
3684
3685 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3686 {
3687         return num_bytes >>= 3;
3688 }
3689
3690 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3691 {
3692         struct btrfs_root *root = BTRFS_I(inode)->root;
3693         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3694         u64 to_reserve;
3695         int nr_extents;
3696         int retries = 0;
3697         int ret;
3698
3699         if (btrfs_transaction_in_commit(root->fs_info))
3700                 schedule_timeout(1);
3701
3702         num_bytes = ALIGN(num_bytes, root->sectorsize);
3703 again:
3704         spin_lock(&BTRFS_I(inode)->accounting_lock);
3705         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3706         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3707                 nr_extents -= BTRFS_I(inode)->reserved_extents;
3708                 to_reserve = calc_trans_metadata_size(root, nr_extents);
3709         } else {
3710                 nr_extents = 0;
3711                 to_reserve = 0;
3712         }
3713
3714         to_reserve += calc_csum_metadata_size(inode, num_bytes);
3715         ret = reserve_metadata_bytes(block_rsv, to_reserve);
3716         if (ret) {
3717                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3718                 ret = should_retry_reserve(NULL, root, block_rsv, to_reserve,
3719                                            &retries);
3720                 if (ret > 0)
3721                         goto again;
3722                 return ret;
3723         }
3724
3725         BTRFS_I(inode)->reserved_extents += nr_extents;
3726         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3727         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3728
3729         block_rsv_add_bytes(block_rsv, to_reserve, 1);
3730
3731         if (block_rsv->size > 512 * 1024 * 1024)
3732                 shrink_delalloc(NULL, root, to_reserve);
3733
3734         return 0;
3735 }
3736
3737 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
3738 {
3739         struct btrfs_root *root = BTRFS_I(inode)->root;
3740         u64 to_free;
3741         int nr_extents;
3742
3743         num_bytes = ALIGN(num_bytes, root->sectorsize);
3744         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
3745
3746         spin_lock(&BTRFS_I(inode)->accounting_lock);
3747         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
3748         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
3749                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
3750                 BTRFS_I(inode)->reserved_extents -= nr_extents;
3751         } else {
3752                 nr_extents = 0;
3753         }
3754         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3755
3756         to_free = calc_csum_metadata_size(inode, num_bytes);
3757         if (nr_extents > 0)
3758                 to_free += calc_trans_metadata_size(root, nr_extents);
3759
3760         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
3761                                 to_free);
3762 }
3763
3764 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
3765 {
3766         int ret;
3767
3768         ret = btrfs_check_data_free_space(inode, num_bytes);
3769         if (ret)
3770                 return ret;
3771
3772         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
3773         if (ret) {
3774                 btrfs_free_reserved_data_space(inode, num_bytes);
3775                 return ret;
3776         }
3777
3778         return 0;
3779 }
3780
3781 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
3782 {
3783         btrfs_delalloc_release_metadata(inode, num_bytes);
3784         btrfs_free_reserved_data_space(inode, num_bytes);
3785 }
3786
3787 static int update_block_group(struct btrfs_trans_handle *trans,
3788                               struct btrfs_root *root,
3789                               u64 bytenr, u64 num_bytes, int alloc)
3790 {
3791         struct btrfs_block_group_cache *cache;
3792         struct btrfs_fs_info *info = root->fs_info;
3793         int factor;
3794         u64 total = num_bytes;
3795         u64 old_val;
3796         u64 byte_in_group;
3797
3798         /* block accounting for super block */
3799         spin_lock(&info->delalloc_lock);
3800         old_val = btrfs_super_bytes_used(&info->super_copy);
3801         if (alloc)
3802                 old_val += num_bytes;
3803         else
3804                 old_val -= num_bytes;
3805         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3806         spin_unlock(&info->delalloc_lock);
3807
3808         while (total) {
3809                 cache = btrfs_lookup_block_group(info, bytenr);
3810                 if (!cache)
3811                         return -1;
3812                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
3813                                     BTRFS_BLOCK_GROUP_RAID1 |
3814                                     BTRFS_BLOCK_GROUP_RAID10))
3815                         factor = 2;
3816                 else
3817                         factor = 1;
3818                 byte_in_group = bytenr - cache->key.objectid;
3819                 WARN_ON(byte_in_group > cache->key.offset);
3820
3821                 spin_lock(&cache->space_info->lock);
3822                 spin_lock(&cache->lock);
3823                 cache->dirty = 1;
3824                 old_val = btrfs_block_group_used(&cache->item);
3825                 num_bytes = min(total, cache->key.offset - byte_in_group);
3826                 if (alloc) {
3827                         old_val += num_bytes;
3828                         btrfs_set_block_group_used(&cache->item, old_val);
3829                         cache->reserved -= num_bytes;
3830                         cache->space_info->bytes_reserved -= num_bytes;
3831                         cache->space_info->bytes_used += num_bytes;
3832                         cache->space_info->disk_used += num_bytes * factor;
3833                         spin_unlock(&cache->lock);
3834                         spin_unlock(&cache->space_info->lock);
3835                 } else {
3836                         old_val -= num_bytes;
3837                         btrfs_set_block_group_used(&cache->item, old_val);
3838                         cache->pinned += num_bytes;
3839                         cache->space_info->bytes_pinned += num_bytes;
3840                         cache->space_info->bytes_used -= num_bytes;
3841                         cache->space_info->disk_used -= num_bytes * factor;
3842                         spin_unlock(&cache->lock);
3843                         spin_unlock(&cache->space_info->lock);
3844
3845                         set_extent_dirty(info->pinned_extents,
3846                                          bytenr, bytenr + num_bytes - 1,
3847                                          GFP_NOFS | __GFP_NOFAIL);
3848                 }
3849                 btrfs_put_block_group(cache);
3850                 total -= num_bytes;
3851                 bytenr += num_bytes;
3852         }
3853         return 0;
3854 }
3855
3856 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3857 {
3858         struct btrfs_block_group_cache *cache;
3859         u64 bytenr;
3860
3861         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3862         if (!cache)
3863                 return 0;
3864
3865         bytenr = cache->key.objectid;
3866         btrfs_put_block_group(cache);
3867
3868         return bytenr;
3869 }
3870
3871 static int pin_down_extent(struct btrfs_root *root,
3872                            struct btrfs_block_group_cache *cache,
3873                            u64 bytenr, u64 num_bytes, int reserved)
3874 {
3875         spin_lock(&cache->space_info->lock);
3876         spin_lock(&cache->lock);
3877         cache->pinned += num_bytes;
3878         cache->space_info->bytes_pinned += num_bytes;
3879         if (reserved) {
3880                 cache->reserved -= num_bytes;
3881                 cache->space_info->bytes_reserved -= num_bytes;
3882         }
3883         spin_unlock(&cache->lock);
3884         spin_unlock(&cache->space_info->lock);
3885
3886         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
3887                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
3888         return 0;
3889 }
3890
3891 /*
3892  * this function must be called within transaction
3893  */
3894 int btrfs_pin_extent(struct btrfs_root *root,
3895                      u64 bytenr, u64 num_bytes, int reserved)
3896 {
3897         struct btrfs_block_group_cache *cache;
3898
3899         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
3900         BUG_ON(!cache);
3901
3902         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
3903
3904         btrfs_put_block_group(cache);
3905         return 0;
3906 }
3907
3908 /*
3909  * update size of reserved extents. this function may return -EAGAIN
3910  * if 'reserve' is true or 'sinfo' is false.
3911  */
3912 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
3913                                  u64 num_bytes, int reserve, int sinfo)
3914 {
3915         int ret = 0;
3916         if (sinfo) {
3917                 struct btrfs_space_info *space_info = cache->space_info;
3918                 spin_lock(&space_info->lock);
3919                 spin_lock(&cache->lock);
3920                 if (reserve) {
3921                         if (cache->ro) {
3922                                 ret = -EAGAIN;
3923                         } else {
3924                                 cache->reserved += num_bytes;
3925                                 space_info->bytes_reserved += num_bytes;
3926                         }
3927                 } else {
3928                         if (cache->ro)
3929                                 space_info->bytes_readonly += num_bytes;
3930                         cache->reserved -= num_bytes;
3931                         space_info->bytes_reserved -= num_bytes;
3932                 }
3933                 spin_unlock(&cache->lock);
3934                 spin_unlock(&space_info->lock);
3935         } else {
3936                 spin_lock(&cache->lock);
3937                 if (cache->ro) {
3938                         ret = -EAGAIN;
3939                 } else {
3940                         if (reserve)
3941                                 cache->reserved += num_bytes;
3942                         else
3943                                 cache->reserved -= num_bytes;
3944                 }
3945                 spin_unlock(&cache->lock);
3946         }
3947         return ret;
3948 }
3949
3950 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3951                                 struct btrfs_root *root)
3952 {
3953         struct btrfs_fs_info *fs_info = root->fs_info;
3954         struct btrfs_caching_control *next;
3955         struct btrfs_caching_control *caching_ctl;
3956         struct btrfs_block_group_cache *cache;
3957
3958         down_write(&fs_info->extent_commit_sem);
3959
3960         list_for_each_entry_safe(caching_ctl, next,
3961                                  &fs_info->caching_block_groups, list) {
3962                 cache = caching_ctl->block_group;
3963                 if (block_group_cache_done(cache)) {
3964                         cache->last_byte_to_unpin = (u64)-1;
3965                         list_del_init(&caching_ctl->list);
3966                         put_caching_control(caching_ctl);
3967                 } else {
3968                         cache->last_byte_to_unpin = caching_ctl->progress;
3969                 }
3970         }
3971
3972         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3973                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3974         else
3975                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3976
3977         up_write(&fs_info->extent_commit_sem);
3978
3979         update_global_block_rsv(fs_info);
3980         return 0;
3981 }
3982
3983 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3984 {
3985         struct btrfs_fs_info *fs_info = root->fs_info;
3986         struct btrfs_block_group_cache *cache = NULL;
3987         u64 len;
3988
3989         while (start <= end) {
3990                 if (!cache ||
3991                     start >= cache->key.objectid + cache->key.offset) {
3992                         if (cache)
3993                                 btrfs_put_block_group(cache);
3994                         cache = btrfs_lookup_block_group(fs_info, start);
3995                         BUG_ON(!cache);
3996                 }
3997
3998                 len = cache->key.objectid + cache->key.offset - start;
3999                 len = min(len, end + 1 - start);
4000
4001                 if (start < cache->last_byte_to_unpin) {
4002                         len = min(len, cache->last_byte_to_unpin - start);
4003                         btrfs_add_free_space(cache, start, len);
4004                 }
4005
4006                 start += len;
4007
4008                 spin_lock(&cache->space_info->lock);
4009                 spin_lock(&cache->lock);
4010                 cache->pinned -= len;
4011                 cache->space_info->bytes_pinned -= len;
4012                 if (cache->ro) {
4013                         cache->space_info->bytes_readonly += len;
4014                 } else if (cache->reserved_pinned > 0) {
4015                         len = min(len, cache->reserved_pinned);
4016                         cache->reserved_pinned -= len;
4017                         cache->space_info->bytes_reserved += len;
4018                 }
4019                 spin_unlock(&cache->lock);
4020                 spin_unlock(&cache->space_info->lock);
4021         }
4022
4023         if (cache)
4024                 btrfs_put_block_group(cache);
4025         return 0;
4026 }
4027
4028 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4029                                struct btrfs_root *root)
4030 {
4031         struct btrfs_fs_info *fs_info = root->fs_info;
4032         struct extent_io_tree *unpin;
4033         struct btrfs_block_rsv *block_rsv;
4034         struct btrfs_block_rsv *next_rsv;
4035         u64 start;
4036         u64 end;
4037         int idx;
4038         int ret;
4039
4040         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4041                 unpin = &fs_info->freed_extents[1];
4042         else
4043                 unpin = &fs_info->freed_extents[0];
4044
4045         while (1) {
4046                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4047                                             EXTENT_DIRTY);
4048                 if (ret)
4049                         break;
4050
4051                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4052
4053                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4054                 unpin_extent_range(root, start, end);
4055                 cond_resched();
4056         }
4057
4058         mutex_lock(&fs_info->durable_block_rsv_mutex);
4059         list_for_each_entry_safe(block_rsv, next_rsv,
4060                                  &fs_info->durable_block_rsv_list, list) {
4061
4062                 idx = trans->transid & 0x1;
4063                 if (block_rsv->freed[idx] > 0) {
4064                         block_rsv_add_bytes(block_rsv,
4065                                             block_rsv->freed[idx], 0);
4066                         block_rsv->freed[idx] = 0;
4067                 }
4068                 if (atomic_read(&block_rsv->usage) == 0) {
4069                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4070
4071                         if (block_rsv->freed[0] == 0 &&
4072                             block_rsv->freed[1] == 0) {
4073                                 list_del_init(&block_rsv->list);
4074                                 kfree(block_rsv);
4075                         }
4076                 } else {
4077                         btrfs_block_rsv_release(root, block_rsv, 0);
4078                 }
4079         }
4080         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4081
4082         return 0;
4083 }
4084
4085 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4086                                 struct btrfs_root *root,
4087                                 u64 bytenr, u64 num_bytes, u64 parent,
4088                                 u64 root_objectid, u64 owner_objectid,
4089                                 u64 owner_offset, int refs_to_drop,
4090                                 struct btrfs_delayed_extent_op *extent_op)
4091 {
4092         struct btrfs_key key;
4093         struct btrfs_path *path;
4094         struct btrfs_fs_info *info = root->fs_info;
4095         struct btrfs_root *extent_root = info->extent_root;
4096         struct extent_buffer *leaf;
4097         struct btrfs_extent_item *ei;
4098         struct btrfs_extent_inline_ref *iref;
4099         int ret;
4100         int is_data;
4101         int extent_slot = 0;
4102         int found_extent = 0;
4103         int num_to_del = 1;
4104         u32 item_size;
4105         u64 refs;
4106
4107         path = btrfs_alloc_path();
4108         if (!path)
4109                 return -ENOMEM;
4110
4111         path->reada = 1;
4112         path->leave_spinning = 1;
4113
4114         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4115         BUG_ON(!is_data && refs_to_drop != 1);
4116
4117         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4118                                     bytenr, num_bytes, parent,
4119                                     root_objectid, owner_objectid,
4120                                     owner_offset);
4121         if (ret == 0) {
4122                 extent_slot = path->slots[0];
4123                 while (extent_slot >= 0) {
4124                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4125                                               extent_slot);
4126                         if (key.objectid != bytenr)
4127                                 break;
4128                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4129                             key.offset == num_bytes) {
4130                                 found_extent = 1;
4131                                 break;
4132                         }
4133                         if (path->slots[0] - extent_slot > 5)
4134                                 break;
4135                         extent_slot--;
4136                 }
4137 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4138                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4139                 if (found_extent && item_size < sizeof(*ei))
4140                         found_extent = 0;
4141 #endif
4142                 if (!found_extent) {
4143                         BUG_ON(iref);
4144                         ret = remove_extent_backref(trans, extent_root, path,
4145                                                     NULL, refs_to_drop,
4146                                                     is_data);
4147                         BUG_ON(ret);
4148                         btrfs_release_path(extent_root, path);
4149                         path->leave_spinning = 1;
4150
4151                         key.objectid = bytenr;
4152                         key.type = BTRFS_EXTENT_ITEM_KEY;
4153                         key.offset = num_bytes;
4154
4155                         ret = btrfs_search_slot(trans, extent_root,
4156                                                 &key, path, -1, 1);
4157                         if (ret) {
4158                                 printk(KERN_ERR "umm, got %d back from search"
4159                                        ", was looking for %llu\n", ret,
4160                                        (unsigned long long)bytenr);
4161                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4162                         }
4163                         BUG_ON(ret);
4164                         extent_slot = path->slots[0];
4165                 }
4166         } else {
4167                 btrfs_print_leaf(extent_root, path->nodes[0]);
4168                 WARN_ON(1);
4169                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4170                        "parent %llu root %llu  owner %llu offset %llu\n",
4171                        (unsigned long long)bytenr,
4172                        (unsigned long long)parent,
4173                        (unsigned long long)root_objectid,
4174                        (unsigned long long)owner_objectid,
4175                        (unsigned long long)owner_offset);
4176         }
4177
4178         leaf = path->nodes[0];
4179         item_size = btrfs_item_size_nr(leaf, extent_slot);
4180 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4181         if (item_size < sizeof(*ei)) {
4182                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4183                 ret = convert_extent_item_v0(trans, extent_root, path,
4184                                              owner_objectid, 0);
4185                 BUG_ON(ret < 0);
4186
4187                 btrfs_release_path(extent_root, path);
4188                 path->leave_spinning = 1;
4189
4190                 key.objectid = bytenr;
4191                 key.type = BTRFS_EXTENT_ITEM_KEY;
4192                 key.offset = num_bytes;
4193
4194                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4195                                         -1, 1);
4196                 if (ret) {
4197                         printk(KERN_ERR "umm, got %d back from search"
4198                                ", was looking for %llu\n", ret,
4199                                (unsigned long long)bytenr);
4200                         btrfs_print_leaf(extent_root, path->nodes[0]);
4201                 }
4202                 BUG_ON(ret);
4203                 extent_slot = path->slots[0];
4204                 leaf = path->nodes[0];
4205                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4206         }
4207 #endif
4208         BUG_ON(item_size < sizeof(*ei));
4209         ei = btrfs_item_ptr(leaf, extent_slot,
4210                             struct btrfs_extent_item);
4211         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4212                 struct btrfs_tree_block_info *bi;
4213                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4214                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4215                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4216         }
4217
4218         refs = btrfs_extent_refs(leaf, ei);
4219         BUG_ON(refs < refs_to_drop);
4220         refs -= refs_to_drop;
4221
4222         if (refs > 0) {
4223                 if (extent_op)
4224                         __run_delayed_extent_op(extent_op, leaf, ei);
4225                 /*
4226                  * In the case of inline back ref, reference count will
4227                  * be updated by remove_extent_backref
4228                  */
4229                 if (iref) {
4230                         BUG_ON(!found_extent);
4231                 } else {
4232                         btrfs_set_extent_refs(leaf, ei, refs);
4233                         btrfs_mark_buffer_dirty(leaf);
4234                 }
4235                 if (found_extent) {
4236                         ret = remove_extent_backref(trans, extent_root, path,
4237                                                     iref, refs_to_drop,
4238                                                     is_data);
4239                         BUG_ON(ret);
4240                 }
4241         } else {
4242                 if (found_extent) {
4243                         BUG_ON(is_data && refs_to_drop !=
4244                                extent_data_ref_count(root, path, iref));
4245                         if (iref) {
4246                                 BUG_ON(path->slots[0] != extent_slot);
4247                         } else {
4248                                 BUG_ON(path->slots[0] != extent_slot + 1);
4249                                 path->slots[0] = extent_slot;
4250                                 num_to_del = 2;
4251                         }
4252                 }
4253
4254                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4255                                       num_to_del);
4256                 BUG_ON(ret);
4257                 btrfs_release_path(extent_root, path);
4258
4259                 if (is_data) {
4260                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4261                         BUG_ON(ret);
4262                 } else {
4263                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4264                              bytenr >> PAGE_CACHE_SHIFT,
4265                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4266                 }
4267
4268                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4269                 BUG_ON(ret);
4270         }
4271         btrfs_free_path(path);
4272         return ret;
4273 }
4274
4275 /*
4276  * when we free an block, it is possible (and likely) that we free the last
4277  * delayed ref for that extent as well.  This searches the delayed ref tree for
4278  * a given extent, and if there are no other delayed refs to be processed, it
4279  * removes it from the tree.
4280  */
4281 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4282                                       struct btrfs_root *root, u64 bytenr)
4283 {
4284         struct btrfs_delayed_ref_head *head;
4285         struct btrfs_delayed_ref_root *delayed_refs;
4286         struct btrfs_delayed_ref_node *ref;
4287         struct rb_node *node;
4288         int ret = 0;
4289
4290         delayed_refs = &trans->transaction->delayed_refs;
4291         spin_lock(&delayed_refs->lock);
4292         head = btrfs_find_delayed_ref_head(trans, bytenr);
4293         if (!head)
4294                 goto out;
4295
4296         node = rb_prev(&head->node.rb_node);
4297         if (!node)
4298                 goto out;
4299
4300         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4301
4302         /* there are still entries for this ref, we can't drop it */
4303         if (ref->bytenr == bytenr)
4304                 goto out;
4305
4306         if (head->extent_op) {
4307                 if (!head->must_insert_reserved)
4308                         goto out;
4309                 kfree(head->extent_op);
4310                 head->extent_op = NULL;
4311         }
4312
4313         /*
4314          * waiting for the lock here would deadlock.  If someone else has it
4315          * locked they are already in the process of dropping it anyway
4316          */
4317         if (!mutex_trylock(&head->mutex))
4318                 goto out;
4319
4320         /*
4321          * at this point we have a head with no other entries.  Go
4322          * ahead and process it.
4323          */
4324         head->node.in_tree = 0;
4325         rb_erase(&head->node.rb_node, &delayed_refs->root);
4326
4327         delayed_refs->num_entries--;
4328
4329         /*
4330          * we don't take a ref on the node because we're removing it from the
4331          * tree, so we just steal the ref the tree was holding.
4332          */
4333         delayed_refs->num_heads--;
4334         if (list_empty(&head->cluster))
4335                 delayed_refs->num_heads_ready--;
4336
4337         list_del_init(&head->cluster);
4338         spin_unlock(&delayed_refs->lock);
4339
4340         BUG_ON(head->extent_op);
4341         if (head->must_insert_reserved)
4342                 ret = 1;
4343
4344         mutex_unlock(&head->mutex);
4345         btrfs_put_delayed_ref(&head->node);
4346         return ret;
4347 out:
4348         spin_unlock(&delayed_refs->lock);
4349         return 0;
4350 }
4351
4352 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4353                            struct btrfs_root *root,
4354                            struct extent_buffer *buf,
4355                            u64 parent, int last_ref)
4356 {
4357         struct btrfs_block_rsv *block_rsv;
4358         struct btrfs_block_group_cache *cache = NULL;
4359         int ret;
4360
4361         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4362                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4363                                                 parent, root->root_key.objectid,
4364                                                 btrfs_header_level(buf),
4365                                                 BTRFS_DROP_DELAYED_REF, NULL);
4366                 BUG_ON(ret);
4367         }
4368
4369         if (!last_ref)
4370                 return;
4371
4372         block_rsv = get_block_rsv(trans, root);
4373         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4374         if (block_rsv->space_info != cache->space_info)
4375                 goto out;
4376
4377         if (btrfs_header_generation(buf) == trans->transid) {
4378                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4379                         ret = check_ref_cleanup(trans, root, buf->start);
4380                         if (!ret)
4381                                 goto pin;
4382                 }
4383
4384                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4385                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4386                         goto pin;
4387                 }
4388
4389                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4390
4391                 btrfs_add_free_space(cache, buf->start, buf->len);
4392                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4393                 if (ret == -EAGAIN) {
4394                         /* block group became read-only */
4395                         update_reserved_bytes(cache, buf->len, 0, 1);
4396                         goto out;
4397                 }
4398
4399                 ret = 1;
4400                 spin_lock(&block_rsv->lock);
4401                 if (block_rsv->reserved < block_rsv->size) {
4402                         block_rsv->reserved += buf->len;
4403                         ret = 0;
4404                 }
4405                 spin_unlock(&block_rsv->lock);
4406
4407                 if (ret) {
4408                         spin_lock(&cache->space_info->lock);
4409                         cache->space_info->bytes_reserved -= buf->len;
4410                         spin_unlock(&cache->space_info->lock);
4411                 }
4412                 goto out;
4413         }
4414 pin:
4415         if (block_rsv->durable && !cache->ro) {
4416                 ret = 0;
4417                 spin_lock(&cache->lock);
4418                 if (!cache->ro) {
4419                         cache->reserved_pinned += buf->len;
4420                         ret = 1;
4421                 }
4422                 spin_unlock(&cache->lock);
4423
4424                 if (ret) {
4425                         spin_lock(&block_rsv->lock);
4426                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4427                         spin_unlock(&block_rsv->lock);
4428                 }
4429         }
4430 out:
4431         btrfs_put_block_group(cache);
4432 }
4433
4434 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4435                       struct btrfs_root *root,
4436                       u64 bytenr, u64 num_bytes, u64 parent,
4437                       u64 root_objectid, u64 owner, u64 offset)
4438 {
4439         int ret;
4440
4441         /*
4442          * tree log blocks never actually go into the extent allocation
4443          * tree, just update pinning info and exit early.
4444          */
4445         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4446                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4447                 /* unlocks the pinned mutex */
4448                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4449                 ret = 0;
4450         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4451                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4452                                         parent, root_objectid, (int)owner,
4453                                         BTRFS_DROP_DELAYED_REF, NULL);
4454                 BUG_ON(ret);
4455         } else {
4456                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4457                                         parent, root_objectid, owner,
4458                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4459                 BUG_ON(ret);
4460         }
4461         return ret;
4462 }
4463
4464 static u64 stripe_align(struct btrfs_root *root, u64 val)
4465 {
4466         u64 mask = ((u64)root->stripesize - 1);
4467         u64 ret = (val + mask) & ~mask;
4468         return ret;
4469 }
4470
4471 /*
4472  * when we wait for progress in the block group caching, its because
4473  * our allocation attempt failed at least once.  So, we must sleep
4474  * and let some progress happen before we try again.
4475  *
4476  * This function will sleep at least once waiting for new free space to
4477  * show up, and then it will check the block group free space numbers
4478  * for our min num_bytes.  Another option is to have it go ahead
4479  * and look in the rbtree for a free extent of a given size, but this
4480  * is a good start.
4481  */
4482 static noinline int
4483 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4484                                 u64 num_bytes)
4485 {
4486         struct btrfs_caching_control *caching_ctl;
4487         DEFINE_WAIT(wait);
4488
4489         caching_ctl = get_caching_control(cache);
4490         if (!caching_ctl)
4491                 return 0;
4492
4493         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4494                    (cache->free_space >= num_bytes));
4495
4496         put_caching_control(caching_ctl);
4497         return 0;
4498 }
4499
4500 static noinline int
4501 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4502 {
4503         struct btrfs_caching_control *caching_ctl;
4504         DEFINE_WAIT(wait);
4505
4506         caching_ctl = get_caching_control(cache);
4507         if (!caching_ctl)
4508                 return 0;
4509
4510         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4511
4512         put_caching_control(caching_ctl);
4513         return 0;
4514 }
4515
4516 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4517 {
4518         int index;
4519         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4520                 index = 0;
4521         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4522                 index = 1;
4523         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4524                 index = 2;
4525         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4526                 index = 3;
4527         else
4528                 index = 4;
4529         return index;
4530 }
4531
4532 enum btrfs_loop_type {
4533         LOOP_FIND_IDEAL = 0,
4534         LOOP_CACHING_NOWAIT = 1,
4535         LOOP_CACHING_WAIT = 2,
4536         LOOP_ALLOC_CHUNK = 3,
4537         LOOP_NO_EMPTY_SIZE = 4,
4538 };
4539
4540 /*
4541  * walks the btree of allocated extents and find a hole of a given size.
4542  * The key ins is changed to record the hole:
4543  * ins->objectid == block start
4544  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4545  * ins->offset == number of blocks
4546  * Any available blocks before search_start are skipped.
4547  */
4548 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4549                                      struct btrfs_root *orig_root,
4550                                      u64 num_bytes, u64 empty_size,
4551                                      u64 search_start, u64 search_end,
4552                                      u64 hint_byte, struct btrfs_key *ins,
4553                                      int data)
4554 {
4555         int ret = 0;
4556         struct btrfs_root *root = orig_root->fs_info->extent_root;
4557         struct btrfs_free_cluster *last_ptr = NULL;
4558         struct btrfs_block_group_cache *block_group = NULL;
4559         int empty_cluster = 2 * 1024 * 1024;
4560         int allowed_chunk_alloc = 0;
4561         int done_chunk_alloc = 0;
4562         struct btrfs_space_info *space_info;
4563         int last_ptr_loop = 0;
4564         int loop = 0;
4565         int index = 0;
4566         bool found_uncached_bg = false;
4567         bool failed_cluster_refill = false;
4568         bool failed_alloc = false;
4569         u64 ideal_cache_percent = 0;
4570         u64 ideal_cache_offset = 0;
4571
4572         WARN_ON(num_bytes < root->sectorsize);
4573         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4574         ins->objectid = 0;
4575         ins->offset = 0;
4576
4577         space_info = __find_space_info(root->fs_info, data);
4578         if (!space_info) {
4579                 printk(KERN_ERR "No space info for %d\n", data);
4580                 return -ENOSPC;
4581         }
4582
4583         if (orig_root->ref_cows || empty_size)
4584                 allowed_chunk_alloc = 1;
4585
4586         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4587                 last_ptr = &root->fs_info->meta_alloc_cluster;
4588                 if (!btrfs_test_opt(root, SSD))
4589                         empty_cluster = 64 * 1024;
4590         }
4591
4592         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4593                 last_ptr = &root->fs_info->data_alloc_cluster;
4594         }
4595
4596         if (last_ptr) {
4597                 spin_lock(&last_ptr->lock);
4598                 if (last_ptr->block_group)
4599                         hint_byte = last_ptr->window_start;
4600                 spin_unlock(&last_ptr->lock);
4601         }
4602
4603         search_start = max(search_start, first_logical_byte(root, 0));
4604         search_start = max(search_start, hint_byte);
4605
4606         if (!last_ptr)
4607                 empty_cluster = 0;
4608
4609         if (search_start == hint_byte) {
4610 ideal_cache:
4611                 block_group = btrfs_lookup_block_group(root->fs_info,
4612                                                        search_start);
4613                 /*
4614                  * we don't want to use the block group if it doesn't match our
4615                  * allocation bits, or if its not cached.
4616                  *
4617                  * However if we are re-searching with an ideal block group
4618                  * picked out then we don't care that the block group is cached.
4619                  */
4620                 if (block_group && block_group_bits(block_group, data) &&
4621                     (block_group->cached != BTRFS_CACHE_NO ||
4622                      search_start == ideal_cache_offset)) {
4623                         down_read(&space_info->groups_sem);
4624                         if (list_empty(&block_group->list) ||
4625                             block_group->ro) {
4626                                 /*
4627                                  * someone is removing this block group,
4628                                  * we can't jump into the have_block_group
4629                                  * target because our list pointers are not
4630                                  * valid
4631                                  */
4632                                 btrfs_put_block_group(block_group);
4633                                 up_read(&space_info->groups_sem);
4634                         } else {
4635                                 index = get_block_group_index(block_group);
4636                                 goto have_block_group;
4637                         }
4638                 } else if (block_group) {
4639                         btrfs_put_block_group(block_group);
4640                 }
4641         }
4642 search:
4643         down_read(&space_info->groups_sem);
4644         list_for_each_entry(block_group, &space_info->block_groups[index],
4645                             list) {
4646                 u64 offset;
4647                 int cached;
4648
4649                 btrfs_get_block_group(block_group);
4650                 search_start = block_group->key.objectid;
4651
4652 have_block_group:
4653                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4654                         u64 free_percent;
4655
4656                         free_percent = btrfs_block_group_used(&block_group->item);
4657                         free_percent *= 100;
4658                         free_percent = div64_u64(free_percent,
4659                                                  block_group->key.offset);
4660                         free_percent = 100 - free_percent;
4661                         if (free_percent > ideal_cache_percent &&
4662                             likely(!block_group->ro)) {
4663                                 ideal_cache_offset = block_group->key.objectid;
4664                                 ideal_cache_percent = free_percent;
4665                         }
4666
4667                         /*
4668                          * We only want to start kthread caching if we are at
4669                          * the point where we will wait for caching to make
4670                          * progress, or if our ideal search is over and we've
4671                          * found somebody to start caching.
4672                          */
4673                         if (loop > LOOP_CACHING_NOWAIT ||
4674                             (loop > LOOP_FIND_IDEAL &&
4675                              atomic_read(&space_info->caching_threads) < 2)) {
4676                                 ret = cache_block_group(block_group);
4677                                 BUG_ON(ret);
4678                         }
4679                         found_uncached_bg = true;
4680
4681                         /*
4682                          * If loop is set for cached only, try the next block
4683                          * group.
4684                          */
4685                         if (loop == LOOP_FIND_IDEAL)
4686                                 goto loop;
4687                 }
4688
4689                 cached = block_group_cache_done(block_group);
4690                 if (unlikely(!cached))
4691                         found_uncached_bg = true;
4692
4693                 if (unlikely(block_group->ro))
4694                         goto loop;
4695
4696                 /*
4697                  * Ok we want to try and use the cluster allocator, so lets look
4698                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4699                  * have tried the cluster allocator plenty of times at this
4700                  * point and not have found anything, so we are likely way too
4701                  * fragmented for the clustering stuff to find anything, so lets
4702                  * just skip it and let the allocator find whatever block it can
4703                  * find
4704                  */
4705                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4706                         /*
4707                          * the refill lock keeps out other
4708                          * people trying to start a new cluster
4709                          */
4710                         spin_lock(&last_ptr->refill_lock);
4711                         if (last_ptr->block_group &&
4712                             (last_ptr->block_group->ro ||
4713                             !block_group_bits(last_ptr->block_group, data))) {
4714                                 offset = 0;
4715                                 goto refill_cluster;
4716                         }
4717
4718                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4719                                                  num_bytes, search_start);
4720                         if (offset) {
4721                                 /* we have a block, we're done */
4722                                 spin_unlock(&last_ptr->refill_lock);
4723                                 goto checks;
4724                         }
4725
4726                         spin_lock(&last_ptr->lock);
4727                         /*
4728                          * whoops, this cluster doesn't actually point to
4729                          * this block group.  Get a ref on the block
4730                          * group is does point to and try again
4731                          */
4732                         if (!last_ptr_loop && last_ptr->block_group &&
4733                             last_ptr->block_group != block_group) {
4734
4735                                 btrfs_put_block_group(block_group);
4736                                 block_group = last_ptr->block_group;
4737                                 btrfs_get_block_group(block_group);
4738                                 spin_unlock(&last_ptr->lock);
4739                                 spin_unlock(&last_ptr->refill_lock);
4740
4741                                 last_ptr_loop = 1;
4742                                 search_start = block_group->key.objectid;
4743                                 /*
4744                                  * we know this block group is properly
4745                                  * in the list because
4746                                  * btrfs_remove_block_group, drops the
4747                                  * cluster before it removes the block
4748                                  * group from the list
4749                                  */
4750                                 goto have_block_group;
4751                         }
4752                         spin_unlock(&last_ptr->lock);
4753 refill_cluster:
4754                         /*
4755                          * this cluster didn't work out, free it and
4756                          * start over
4757                          */
4758                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4759
4760                         last_ptr_loop = 0;
4761
4762                         /* allocate a cluster in this block group */
4763                         ret = btrfs_find_space_cluster(trans, root,
4764                                                block_group, last_ptr,
4765                                                offset, num_bytes,
4766                                                empty_cluster + empty_size);
4767                         if (ret == 0) {
4768                                 /*
4769                                  * now pull our allocation out of this
4770                                  * cluster
4771                                  */
4772                                 offset = btrfs_alloc_from_cluster(block_group,
4773                                                   last_ptr, num_bytes,
4774                                                   search_start);
4775                                 if (offset) {
4776                                         /* we found one, proceed */
4777                                         spin_unlock(&last_ptr->refill_lock);
4778                                         goto checks;
4779                                 }
4780                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4781                                    && !failed_cluster_refill) {
4782                                 spin_unlock(&last_ptr->refill_lock);
4783
4784                                 failed_cluster_refill = true;
4785                                 wait_block_group_cache_progress(block_group,
4786                                        num_bytes + empty_cluster + empty_size);
4787                                 goto have_block_group;
4788                         }
4789
4790                         /*
4791                          * at this point we either didn't find a cluster
4792                          * or we weren't able to allocate a block from our
4793                          * cluster.  Free the cluster we've been trying
4794                          * to use, and go to the next block group
4795                          */
4796                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4797                         spin_unlock(&last_ptr->refill_lock);
4798                         goto loop;
4799                 }
4800
4801                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4802                                                     num_bytes, empty_size);
4803                 /*
4804                  * If we didn't find a chunk, and we haven't failed on this
4805                  * block group before, and this block group is in the middle of
4806                  * caching and we are ok with waiting, then go ahead and wait
4807                  * for progress to be made, and set failed_alloc to true.
4808                  *
4809                  * If failed_alloc is true then we've already waited on this
4810                  * block group once and should move on to the next block group.
4811                  */
4812                 if (!offset && !failed_alloc && !cached &&
4813                     loop > LOOP_CACHING_NOWAIT) {
4814                         wait_block_group_cache_progress(block_group,
4815                                                 num_bytes + empty_size);
4816                         failed_alloc = true;
4817                         goto have_block_group;
4818                 } else if (!offset) {
4819                         goto loop;
4820                 }
4821 checks:
4822                 search_start = stripe_align(root, offset);
4823                 /* move on to the next group */
4824                 if (search_start + num_bytes >= search_end) {
4825                         btrfs_add_free_space(block_group, offset, num_bytes);
4826                         goto loop;
4827                 }
4828
4829                 /* move on to the next group */
4830                 if (search_start + num_bytes >
4831                     block_group->key.objectid + block_group->key.offset) {
4832                         btrfs_add_free_space(block_group, offset, num_bytes);
4833                         goto loop;
4834                 }
4835
4836                 ins->objectid = search_start;
4837                 ins->offset = num_bytes;
4838
4839                 if (offset < search_start)
4840                         btrfs_add_free_space(block_group, offset,
4841                                              search_start - offset);
4842                 BUG_ON(offset > search_start);
4843
4844                 ret = update_reserved_bytes(block_group, num_bytes, 1,
4845                                             (data & BTRFS_BLOCK_GROUP_DATA));
4846                 if (ret == -EAGAIN) {
4847                         btrfs_add_free_space(block_group, offset, num_bytes);
4848                         goto loop;
4849                 }
4850
4851                 /* we are all good, lets return */
4852                 ins->objectid = search_start;
4853                 ins->offset = num_bytes;
4854
4855                 if (offset < search_start)
4856                         btrfs_add_free_space(block_group, offset,
4857                                              search_start - offset);
4858                 BUG_ON(offset > search_start);
4859                 break;
4860 loop:
4861                 failed_cluster_refill = false;
4862                 failed_alloc = false;
4863                 BUG_ON(index != get_block_group_index(block_group));
4864                 btrfs_put_block_group(block_group);
4865         }
4866         up_read(&space_info->groups_sem);
4867
4868         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
4869                 goto search;
4870
4871         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4872          *                      for them to make caching progress.  Also
4873          *                      determine the best possible bg to cache
4874          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4875          *                      caching kthreads as we move along
4876          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4877          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4878          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4879          *                      again
4880          */
4881         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4882             (found_uncached_bg || empty_size || empty_cluster ||
4883              allowed_chunk_alloc)) {
4884                 index = 0;
4885                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4886                         found_uncached_bg = false;
4887                         loop++;
4888                         if (!ideal_cache_percent &&
4889                             atomic_read(&space_info->caching_threads))
4890                                 goto search;
4891
4892                         /*
4893                          * 1 of the following 2 things have happened so far
4894                          *
4895                          * 1) We found an ideal block group for caching that
4896                          * is mostly full and will cache quickly, so we might
4897                          * as well wait for it.
4898                          *
4899                          * 2) We searched for cached only and we didn't find
4900                          * anything, and we didn't start any caching kthreads
4901                          * either, so chances are we will loop through and
4902                          * start a couple caching kthreads, and then come back
4903                          * around and just wait for them.  This will be slower
4904                          * because we will have 2 caching kthreads reading at
4905                          * the same time when we could have just started one
4906                          * and waited for it to get far enough to give us an
4907                          * allocation, so go ahead and go to the wait caching
4908                          * loop.
4909                          */
4910                         loop = LOOP_CACHING_WAIT;
4911                         search_start = ideal_cache_offset;
4912                         ideal_cache_percent = 0;
4913                         goto ideal_cache;
4914                 } else if (loop == LOOP_FIND_IDEAL) {
4915                         /*
4916                          * Didn't find a uncached bg, wait on anything we find
4917                          * next.
4918                          */
4919                         loop = LOOP_CACHING_WAIT;
4920                         goto search;
4921                 }
4922
4923                 if (loop < LOOP_CACHING_WAIT) {
4924                         loop++;
4925                         goto search;
4926                 }
4927
4928                 if (loop == LOOP_ALLOC_CHUNK) {
4929                         empty_size = 0;
4930                         empty_cluster = 0;
4931                 }
4932
4933                 if (allowed_chunk_alloc) {
4934                         ret = do_chunk_alloc(trans, root, num_bytes +
4935                                              2 * 1024 * 1024, data, 1);
4936                         allowed_chunk_alloc = 0;
4937                         done_chunk_alloc = 1;
4938                 } else if (!done_chunk_alloc) {
4939                         space_info->force_alloc = 1;
4940                 }
4941
4942                 if (loop < LOOP_NO_EMPTY_SIZE) {
4943                         loop++;
4944                         goto search;
4945                 }
4946                 ret = -ENOSPC;
4947         } else if (!ins->objectid) {
4948                 ret = -ENOSPC;
4949         }
4950
4951         /* we found what we needed */
4952         if (ins->objectid) {
4953                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4954                         trans->block_group = block_group->key.objectid;
4955
4956                 btrfs_put_block_group(block_group);
4957                 ret = 0;
4958         }
4959
4960         return ret;
4961 }
4962
4963 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4964                             int dump_block_groups)
4965 {
4966         struct btrfs_block_group_cache *cache;
4967         int index = 0;
4968
4969         spin_lock(&info->lock);
4970         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4971                (unsigned long long)(info->total_bytes - info->bytes_used -
4972                                     info->bytes_pinned - info->bytes_reserved -
4973                                     info->bytes_readonly),
4974                (info->full) ? "" : "not ");
4975         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
4976                "reserved=%llu, may_use=%llu, readonly=%llu\n",
4977                (unsigned long long)info->total_bytes,
4978                (unsigned long long)info->bytes_used,
4979                (unsigned long long)info->bytes_pinned,
4980                (unsigned long long)info->bytes_reserved,
4981                (unsigned long long)info->bytes_may_use,
4982                (unsigned long long)info->bytes_readonly);
4983         spin_unlock(&info->lock);
4984
4985         if (!dump_block_groups)
4986                 return;
4987
4988         down_read(&info->groups_sem);
4989 again:
4990         list_for_each_entry(cache, &info->block_groups[index], list) {
4991                 spin_lock(&cache->lock);
4992                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4993                        "%llu pinned %llu reserved\n",
4994                        (unsigned long long)cache->key.objectid,
4995                        (unsigned long long)cache->key.offset,
4996                        (unsigned long long)btrfs_block_group_used(&cache->item),
4997                        (unsigned long long)cache->pinned,
4998                        (unsigned long long)cache->reserved);
4999                 btrfs_dump_free_space(cache, bytes);
5000                 spin_unlock(&cache->lock);
5001         }
5002         if (++index < BTRFS_NR_RAID_TYPES)
5003                 goto again;
5004         up_read(&info->groups_sem);
5005 }
5006
5007 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5008                          struct btrfs_root *root,
5009                          u64 num_bytes, u64 min_alloc_size,
5010                          u64 empty_size, u64 hint_byte,
5011                          u64 search_end, struct btrfs_key *ins,
5012                          u64 data)
5013 {
5014         int ret;
5015         u64 search_start = 0;
5016
5017         data = btrfs_get_alloc_profile(root, data);
5018 again:
5019         /*
5020          * the only place that sets empty_size is btrfs_realloc_node, which
5021          * is not called recursively on allocations
5022          */
5023         if (empty_size || root->ref_cows)
5024                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5025                                      num_bytes + 2 * 1024 * 1024, data, 0);
5026
5027         WARN_ON(num_bytes < root->sectorsize);
5028         ret = find_free_extent(trans, root, num_bytes, empty_size,
5029                                search_start, search_end, hint_byte,
5030                                ins, data);
5031
5032         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5033                 num_bytes = num_bytes >> 1;
5034                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5035                 num_bytes = max(num_bytes, min_alloc_size);
5036                 do_chunk_alloc(trans, root->fs_info->extent_root,
5037                                num_bytes, data, 1);
5038                 goto again;
5039         }
5040         if (ret == -ENOSPC) {
5041                 struct btrfs_space_info *sinfo;
5042
5043                 sinfo = __find_space_info(root->fs_info, data);
5044                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5045                        "wanted %llu\n", (unsigned long long)data,
5046                        (unsigned long long)num_bytes);
5047                 dump_space_info(sinfo, num_bytes, 1);
5048         }
5049
5050         return ret;
5051 }
5052
5053 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5054 {
5055         struct btrfs_block_group_cache *cache;
5056         int ret = 0;
5057
5058         cache = btrfs_lookup_block_group(root->fs_info, start);
5059         if (!cache) {
5060                 printk(KERN_ERR "Unable to find block group for %llu\n",
5061                        (unsigned long long)start);
5062                 return -ENOSPC;
5063         }
5064
5065         ret = btrfs_discard_extent(root, start, len);
5066
5067         btrfs_add_free_space(cache, start, len);
5068         update_reserved_bytes(cache, len, 0, 1);
5069         btrfs_put_block_group(cache);
5070
5071         return ret;
5072 }
5073
5074 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5075                                       struct btrfs_root *root,
5076                                       u64 parent, u64 root_objectid,
5077                                       u64 flags, u64 owner, u64 offset,
5078                                       struct btrfs_key *ins, int ref_mod)
5079 {
5080         int ret;
5081         struct btrfs_fs_info *fs_info = root->fs_info;
5082         struct btrfs_extent_item *extent_item;
5083         struct btrfs_extent_inline_ref *iref;
5084         struct btrfs_path *path;
5085         struct extent_buffer *leaf;
5086         int type;
5087         u32 size;
5088
5089         if (parent > 0)
5090                 type = BTRFS_SHARED_DATA_REF_KEY;
5091         else
5092                 type = BTRFS_EXTENT_DATA_REF_KEY;
5093
5094         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5095
5096         path = btrfs_alloc_path();
5097         BUG_ON(!path);
5098
5099         path->leave_spinning = 1;
5100         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5101                                       ins, size);
5102         BUG_ON(ret);
5103
5104         leaf = path->nodes[0];
5105         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5106                                      struct btrfs_extent_item);
5107         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5108         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5109         btrfs_set_extent_flags(leaf, extent_item,
5110                                flags | BTRFS_EXTENT_FLAG_DATA);
5111
5112         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5113         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5114         if (parent > 0) {
5115                 struct btrfs_shared_data_ref *ref;
5116                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5117                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5118                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5119         } else {
5120                 struct btrfs_extent_data_ref *ref;
5121                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5122                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5123                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5124                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5125                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5126         }
5127
5128         btrfs_mark_buffer_dirty(path->nodes[0]);
5129         btrfs_free_path(path);
5130
5131         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5132         if (ret) {
5133                 printk(KERN_ERR "btrfs update block group failed for %llu "
5134                        "%llu\n", (unsigned long long)ins->objectid,
5135                        (unsigned long long)ins->offset);
5136                 BUG();
5137         }
5138         return ret;
5139 }
5140
5141 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5142                                      struct btrfs_root *root,
5143                                      u64 parent, u64 root_objectid,
5144                                      u64 flags, struct btrfs_disk_key *key,
5145                                      int level, struct btrfs_key *ins)
5146 {
5147         int ret;
5148         struct btrfs_fs_info *fs_info = root->fs_info;
5149         struct btrfs_extent_item *extent_item;
5150         struct btrfs_tree_block_info *block_info;
5151         struct btrfs_extent_inline_ref *iref;
5152         struct btrfs_path *path;
5153         struct extent_buffer *leaf;
5154         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5155
5156         path = btrfs_alloc_path();
5157         BUG_ON(!path);
5158
5159         path->leave_spinning = 1;
5160         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5161                                       ins, size);
5162         BUG_ON(ret);
5163
5164         leaf = path->nodes[0];
5165         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5166                                      struct btrfs_extent_item);
5167         btrfs_set_extent_refs(leaf, extent_item, 1);
5168         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5169         btrfs_set_extent_flags(leaf, extent_item,
5170                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5171         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5172
5173         btrfs_set_tree_block_key(leaf, block_info, key);
5174         btrfs_set_tree_block_level(leaf, block_info, level);
5175
5176         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5177         if (parent > 0) {
5178                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5179                 btrfs_set_extent_inline_ref_type(leaf, iref,
5180                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5181                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5182         } else {
5183                 btrfs_set_extent_inline_ref_type(leaf, iref,
5184                                                  BTRFS_TREE_BLOCK_REF_KEY);
5185                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5186         }
5187
5188         btrfs_mark_buffer_dirty(leaf);
5189         btrfs_free_path(path);
5190
5191         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5192         if (ret) {
5193                 printk(KERN_ERR "btrfs update block group failed for %llu "
5194                        "%llu\n", (unsigned long long)ins->objectid,
5195                        (unsigned long long)ins->offset);
5196                 BUG();
5197         }
5198         return ret;
5199 }
5200
5201 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5202                                      struct btrfs_root *root,
5203                                      u64 root_objectid, u64 owner,
5204                                      u64 offset, struct btrfs_key *ins)
5205 {
5206         int ret;
5207
5208         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5209
5210         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5211                                          0, root_objectid, owner, offset,
5212                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5213         return ret;
5214 }
5215
5216 /*
5217  * this is used by the tree logging recovery code.  It records that
5218  * an extent has been allocated and makes sure to clear the free
5219  * space cache bits as well
5220  */
5221 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5222                                    struct btrfs_root *root,
5223                                    u64 root_objectid, u64 owner, u64 offset,
5224                                    struct btrfs_key *ins)
5225 {
5226         int ret;
5227         struct btrfs_block_group_cache *block_group;
5228         struct btrfs_caching_control *caching_ctl;
5229         u64 start = ins->objectid;
5230         u64 num_bytes = ins->offset;
5231
5232         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5233         cache_block_group(block_group);
5234         caching_ctl = get_caching_control(block_group);
5235
5236         if (!caching_ctl) {
5237                 BUG_ON(!block_group_cache_done(block_group));
5238                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5239                 BUG_ON(ret);
5240         } else {
5241                 mutex_lock(&caching_ctl->mutex);
5242
5243                 if (start >= caching_ctl->progress) {
5244                         ret = add_excluded_extent(root, start, num_bytes);
5245                         BUG_ON(ret);
5246                 } else if (start + num_bytes <= caching_ctl->progress) {
5247                         ret = btrfs_remove_free_space(block_group,
5248                                                       start, num_bytes);
5249                         BUG_ON(ret);
5250                 } else {
5251                         num_bytes = caching_ctl->progress - start;
5252                         ret = btrfs_remove_free_space(block_group,
5253                                                       start, num_bytes);
5254                         BUG_ON(ret);
5255
5256                         start = caching_ctl->progress;
5257                         num_bytes = ins->objectid + ins->offset -
5258                                     caching_ctl->progress;
5259                         ret = add_excluded_extent(root, start, num_bytes);
5260                         BUG_ON(ret);
5261                 }
5262
5263                 mutex_unlock(&caching_ctl->mutex);
5264                 put_caching_control(caching_ctl);
5265         }
5266
5267         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5268         BUG_ON(ret);
5269         btrfs_put_block_group(block_group);
5270         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5271                                          0, owner, offset, ins, 1);
5272         return ret;
5273 }
5274
5275 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5276                                             struct btrfs_root *root,
5277                                             u64 bytenr, u32 blocksize,
5278                                             int level)
5279 {
5280         struct extent_buffer *buf;
5281
5282         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5283         if (!buf)
5284                 return ERR_PTR(-ENOMEM);
5285         btrfs_set_header_generation(buf, trans->transid);
5286         btrfs_set_buffer_lockdep_class(buf, level);
5287         btrfs_tree_lock(buf);
5288         clean_tree_block(trans, root, buf);
5289
5290         btrfs_set_lock_blocking(buf);
5291         btrfs_set_buffer_uptodate(buf);
5292
5293         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5294                 /*
5295                  * we allow two log transactions at a time, use different
5296                  * EXENT bit to differentiate dirty pages.
5297                  */
5298                 if (root->log_transid % 2 == 0)
5299                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5300                                         buf->start + buf->len - 1, GFP_NOFS);
5301                 else
5302                         set_extent_new(&root->dirty_log_pages, buf->start,
5303                                         buf->start + buf->len - 1, GFP_NOFS);
5304         } else {
5305                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5306                          buf->start + buf->len - 1, GFP_NOFS);
5307         }
5308         trans->blocks_used++;
5309         /* this returns a buffer locked for blocking */
5310         return buf;
5311 }
5312
5313 static struct btrfs_block_rsv *
5314 use_block_rsv(struct btrfs_trans_handle *trans,
5315               struct btrfs_root *root, u32 blocksize)
5316 {
5317         struct btrfs_block_rsv *block_rsv;
5318         int ret;
5319
5320         block_rsv = get_block_rsv(trans, root);
5321
5322         if (block_rsv->size == 0) {
5323                 ret = reserve_metadata_bytes(block_rsv, blocksize);
5324                 if (ret)
5325                         return ERR_PTR(ret);
5326                 return block_rsv;
5327         }
5328
5329         ret = block_rsv_use_bytes(block_rsv, blocksize);
5330         if (!ret)
5331                 return block_rsv;
5332
5333         WARN_ON(1);
5334         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
5335                 block_rsv->size, block_rsv->reserved,
5336                 block_rsv->freed[0], block_rsv->freed[1]);
5337
5338         return ERR_PTR(-ENOSPC);
5339 }
5340
5341 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5342 {
5343         block_rsv_add_bytes(block_rsv, blocksize, 0);
5344         block_rsv_release_bytes(block_rsv, NULL, 0);
5345 }
5346
5347 /*
5348  * finds a free extent and does all the dirty work required for allocation
5349  * returns the key for the extent through ins, and a tree buffer for
5350  * the first block of the extent through buf.
5351  *
5352  * returns the tree buffer or NULL.
5353  */
5354 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5355                                         struct btrfs_root *root, u32 blocksize,
5356                                         u64 parent, u64 root_objectid,
5357                                         struct btrfs_disk_key *key, int level,
5358                                         u64 hint, u64 empty_size)
5359 {
5360         struct btrfs_key ins;
5361         struct btrfs_block_rsv *block_rsv;
5362         struct extent_buffer *buf;
5363         u64 flags = 0;
5364         int ret;
5365
5366
5367         block_rsv = use_block_rsv(trans, root, blocksize);
5368         if (IS_ERR(block_rsv))
5369                 return ERR_CAST(block_rsv);
5370
5371         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5372                                    empty_size, hint, (u64)-1, &ins, 0);
5373         if (ret) {
5374                 unuse_block_rsv(block_rsv, blocksize);
5375                 return ERR_PTR(ret);
5376         }
5377
5378         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5379                                     blocksize, level);
5380         BUG_ON(IS_ERR(buf));
5381
5382         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5383                 if (parent == 0)
5384                         parent = ins.objectid;
5385                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5386         } else
5387                 BUG_ON(parent > 0);
5388
5389         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5390                 struct btrfs_delayed_extent_op *extent_op;
5391                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5392                 BUG_ON(!extent_op);
5393                 if (key)
5394                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5395                 else
5396                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5397                 extent_op->flags_to_set = flags;
5398                 extent_op->update_key = 1;
5399                 extent_op->update_flags = 1;
5400                 extent_op->is_data = 0;
5401
5402                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5403                                         ins.offset, parent, root_objectid,
5404                                         level, BTRFS_ADD_DELAYED_EXTENT,
5405                                         extent_op);
5406                 BUG_ON(ret);
5407         }
5408         return buf;
5409 }
5410
5411 struct walk_control {
5412         u64 refs[BTRFS_MAX_LEVEL];
5413         u64 flags[BTRFS_MAX_LEVEL];
5414         struct btrfs_key update_progress;
5415         int stage;
5416         int level;
5417         int shared_level;
5418         int update_ref;
5419         int keep_locks;
5420         int reada_slot;
5421         int reada_count;
5422 };
5423
5424 #define DROP_REFERENCE  1
5425 #define UPDATE_BACKREF  2
5426
5427 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5428                                      struct btrfs_root *root,
5429                                      struct walk_control *wc,
5430                                      struct btrfs_path *path)
5431 {
5432         u64 bytenr;
5433         u64 generation;
5434         u64 refs;
5435         u64 flags;
5436         u64 last = 0;
5437         u32 nritems;
5438         u32 blocksize;
5439         struct btrfs_key key;
5440         struct extent_buffer *eb;
5441         int ret;
5442         int slot;
5443         int nread = 0;
5444
5445         if (path->slots[wc->level] < wc->reada_slot) {
5446                 wc->reada_count = wc->reada_count * 2 / 3;
5447                 wc->reada_count = max(wc->reada_count, 2);
5448         } else {
5449                 wc->reada_count = wc->reada_count * 3 / 2;
5450                 wc->reada_count = min_t(int, wc->reada_count,
5451                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5452         }
5453
5454         eb = path->nodes[wc->level];
5455         nritems = btrfs_header_nritems(eb);
5456         blocksize = btrfs_level_size(root, wc->level - 1);
5457
5458         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5459                 if (nread >= wc->reada_count)
5460                         break;
5461
5462                 cond_resched();
5463                 bytenr = btrfs_node_blockptr(eb, slot);
5464                 generation = btrfs_node_ptr_generation(eb, slot);
5465
5466                 if (slot == path->slots[wc->level])
5467                         goto reada;
5468
5469                 if (wc->stage == UPDATE_BACKREF &&
5470                     generation <= root->root_key.offset)
5471                         continue;
5472
5473                 /* We don't lock the tree block, it's OK to be racy here */
5474                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5475                                                &refs, &flags);
5476                 BUG_ON(ret);
5477                 BUG_ON(refs == 0);
5478
5479                 if (wc->stage == DROP_REFERENCE) {
5480                         if (refs == 1)
5481                                 goto reada;
5482
5483                         if (wc->level == 1 &&
5484                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5485                                 continue;
5486                         if (!wc->update_ref ||
5487                             generation <= root->root_key.offset)
5488                                 continue;
5489                         btrfs_node_key_to_cpu(eb, &key, slot);
5490                         ret = btrfs_comp_cpu_keys(&key,
5491                                                   &wc->update_progress);
5492                         if (ret < 0)
5493                                 continue;
5494                 } else {
5495                         if (wc->level == 1 &&
5496                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5497                                 continue;
5498                 }
5499 reada:
5500                 ret = readahead_tree_block(root, bytenr, blocksize,
5501                                            generation);
5502                 if (ret)
5503                         break;
5504                 last = bytenr + blocksize;
5505                 nread++;
5506         }
5507         wc->reada_slot = slot;
5508 }
5509
5510 /*
5511  * hepler to process tree block while walking down the tree.
5512  *
5513  * when wc->stage == UPDATE_BACKREF, this function updates
5514  * back refs for pointers in the block.
5515  *
5516  * NOTE: return value 1 means we should stop walking down.
5517  */
5518 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5519                                    struct btrfs_root *root,
5520                                    struct btrfs_path *path,
5521                                    struct walk_control *wc, int lookup_info)
5522 {
5523         int level = wc->level;
5524         struct extent_buffer *eb = path->nodes[level];
5525         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5526         int ret;
5527
5528         if (wc->stage == UPDATE_BACKREF &&
5529             btrfs_header_owner(eb) != root->root_key.objectid)
5530                 return 1;
5531
5532         /*
5533          * when reference count of tree block is 1, it won't increase
5534          * again. once full backref flag is set, we never clear it.
5535          */
5536         if (lookup_info &&
5537             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5538              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5539                 BUG_ON(!path->locks[level]);
5540                 ret = btrfs_lookup_extent_info(trans, root,
5541                                                eb->start, eb->len,
5542                                                &wc->refs[level],
5543                                                &wc->flags[level]);
5544                 BUG_ON(ret);
5545                 BUG_ON(wc->refs[level] == 0);
5546         }
5547
5548         if (wc->stage == DROP_REFERENCE) {
5549                 if (wc->refs[level] > 1)
5550                         return 1;
5551
5552                 if (path->locks[level] && !wc->keep_locks) {
5553                         btrfs_tree_unlock(eb);
5554                         path->locks[level] = 0;
5555                 }
5556                 return 0;
5557         }
5558
5559         /* wc->stage == UPDATE_BACKREF */
5560         if (!(wc->flags[level] & flag)) {
5561                 BUG_ON(!path->locks[level]);
5562                 ret = btrfs_inc_ref(trans, root, eb, 1);
5563                 BUG_ON(ret);
5564                 ret = btrfs_dec_ref(trans, root, eb, 0);
5565                 BUG_ON(ret);
5566                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5567                                                   eb->len, flag, 0);
5568                 BUG_ON(ret);
5569                 wc->flags[level] |= flag;
5570         }
5571
5572         /*
5573          * the block is shared by multiple trees, so it's not good to
5574          * keep the tree lock
5575          */
5576         if (path->locks[level] && level > 0) {
5577                 btrfs_tree_unlock(eb);
5578                 path->locks[level] = 0;
5579         }
5580         return 0;
5581 }
5582
5583 /*
5584  * hepler to process tree block pointer.
5585  *
5586  * when wc->stage == DROP_REFERENCE, this function checks
5587  * reference count of the block pointed to. if the block
5588  * is shared and we need update back refs for the subtree
5589  * rooted at the block, this function changes wc->stage to
5590  * UPDATE_BACKREF. if the block is shared and there is no
5591  * need to update back, this function drops the reference
5592  * to the block.
5593  *
5594  * NOTE: return value 1 means we should stop walking down.
5595  */
5596 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5597                                  struct btrfs_root *root,
5598                                  struct btrfs_path *path,
5599                                  struct walk_control *wc, int *lookup_info)
5600 {
5601         u64 bytenr;
5602         u64 generation;
5603         u64 parent;
5604         u32 blocksize;
5605         struct btrfs_key key;
5606         struct extent_buffer *next;
5607         int level = wc->level;
5608         int reada = 0;
5609         int ret = 0;
5610
5611         generation = btrfs_node_ptr_generation(path->nodes[level],
5612                                                path->slots[level]);
5613         /*
5614          * if the lower level block was created before the snapshot
5615          * was created, we know there is no need to update back refs
5616          * for the subtree
5617          */
5618         if (wc->stage == UPDATE_BACKREF &&
5619             generation <= root->root_key.offset) {
5620                 *lookup_info = 1;
5621                 return 1;
5622         }
5623
5624         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5625         blocksize = btrfs_level_size(root, level - 1);
5626
5627         next = btrfs_find_tree_block(root, bytenr, blocksize);
5628         if (!next) {
5629                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5630                 if (!next)
5631                         return -ENOMEM;
5632                 reada = 1;
5633         }
5634         btrfs_tree_lock(next);
5635         btrfs_set_lock_blocking(next);
5636
5637         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5638                                        &wc->refs[level - 1],
5639                                        &wc->flags[level - 1]);
5640         BUG_ON(ret);
5641         BUG_ON(wc->refs[level - 1] == 0);
5642         *lookup_info = 0;
5643
5644         if (wc->stage == DROP_REFERENCE) {
5645                 if (wc->refs[level - 1] > 1) {
5646                         if (level == 1 &&
5647                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5648                                 goto skip;
5649
5650                         if (!wc->update_ref ||
5651                             generation <= root->root_key.offset)
5652                                 goto skip;
5653
5654                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5655                                               path->slots[level]);
5656                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5657                         if (ret < 0)
5658                                 goto skip;
5659
5660                         wc->stage = UPDATE_BACKREF;
5661                         wc->shared_level = level - 1;
5662                 }
5663         } else {
5664                 if (level == 1 &&
5665                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5666                         goto skip;
5667         }
5668
5669         if (!btrfs_buffer_uptodate(next, generation)) {
5670                 btrfs_tree_unlock(next);
5671                 free_extent_buffer(next);
5672                 next = NULL;
5673                 *lookup_info = 1;
5674         }
5675
5676         if (!next) {
5677                 if (reada && level == 1)
5678                         reada_walk_down(trans, root, wc, path);
5679                 next = read_tree_block(root, bytenr, blocksize, generation);
5680                 btrfs_tree_lock(next);
5681                 btrfs_set_lock_blocking(next);
5682         }
5683
5684         level--;
5685         BUG_ON(level != btrfs_header_level(next));
5686         path->nodes[level] = next;
5687         path->slots[level] = 0;
5688         path->locks[level] = 1;
5689         wc->level = level;
5690         if (wc->level == 1)
5691                 wc->reada_slot = 0;
5692         return 0;
5693 skip:
5694         wc->refs[level - 1] = 0;
5695         wc->flags[level - 1] = 0;
5696         if (wc->stage == DROP_REFERENCE) {
5697                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5698                         parent = path->nodes[level]->start;
5699                 } else {
5700                         BUG_ON(root->root_key.objectid !=
5701                                btrfs_header_owner(path->nodes[level]));
5702                         parent = 0;
5703                 }
5704
5705                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5706                                         root->root_key.objectid, level - 1, 0);
5707                 BUG_ON(ret);
5708         }
5709         btrfs_tree_unlock(next);
5710         free_extent_buffer(next);
5711         *lookup_info = 1;
5712         return 1;
5713 }
5714
5715 /*
5716  * hepler to process tree block while walking up the tree.
5717  *
5718  * when wc->stage == DROP_REFERENCE, this function drops
5719  * reference count on the block.
5720  *
5721  * when wc->stage == UPDATE_BACKREF, this function changes
5722  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5723  * to UPDATE_BACKREF previously while processing the block.
5724  *
5725  * NOTE: return value 1 means we should stop walking up.
5726  */
5727 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5728                                  struct btrfs_root *root,
5729                                  struct btrfs_path *path,
5730                                  struct walk_control *wc)
5731 {
5732         int ret;
5733         int level = wc->level;
5734         struct extent_buffer *eb = path->nodes[level];
5735         u64 parent = 0;
5736
5737         if (wc->stage == UPDATE_BACKREF) {
5738                 BUG_ON(wc->shared_level < level);
5739                 if (level < wc->shared_level)
5740                         goto out;
5741
5742                 ret = find_next_key(path, level + 1, &wc->update_progress);
5743                 if (ret > 0)
5744                         wc->update_ref = 0;
5745
5746                 wc->stage = DROP_REFERENCE;
5747                 wc->shared_level = -1;
5748                 path->slots[level] = 0;
5749
5750                 /*
5751                  * check reference count again if the block isn't locked.
5752                  * we should start walking down the tree again if reference
5753                  * count is one.
5754                  */
5755                 if (!path->locks[level]) {
5756                         BUG_ON(level == 0);
5757                         btrfs_tree_lock(eb);
5758                         btrfs_set_lock_blocking(eb);
5759                         path->locks[level] = 1;
5760
5761                         ret = btrfs_lookup_extent_info(trans, root,
5762                                                        eb->start, eb->len,
5763                                                        &wc->refs[level],
5764                                                        &wc->flags[level]);
5765                         BUG_ON(ret);
5766                         BUG_ON(wc->refs[level] == 0);
5767                         if (wc->refs[level] == 1) {
5768                                 btrfs_tree_unlock(eb);
5769                                 path->locks[level] = 0;
5770                                 return 1;
5771                         }
5772                 }
5773         }
5774
5775         /* wc->stage == DROP_REFERENCE */
5776         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5777
5778         if (wc->refs[level] == 1) {
5779                 if (level == 0) {
5780                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5781                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5782                         else
5783                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5784                         BUG_ON(ret);
5785                 }
5786                 /* make block locked assertion in clean_tree_block happy */
5787                 if (!path->locks[level] &&
5788                     btrfs_header_generation(eb) == trans->transid) {
5789                         btrfs_tree_lock(eb);
5790                         btrfs_set_lock_blocking(eb);
5791                         path->locks[level] = 1;
5792                 }
5793                 clean_tree_block(trans, root, eb);
5794         }
5795
5796         if (eb == root->node) {
5797                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5798                         parent = eb->start;
5799                 else
5800                         BUG_ON(root->root_key.objectid !=
5801                                btrfs_header_owner(eb));
5802         } else {
5803                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5804                         parent = path->nodes[level + 1]->start;
5805                 else
5806                         BUG_ON(root->root_key.objectid !=
5807                                btrfs_header_owner(path->nodes[level + 1]));
5808         }
5809
5810         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5811 out:
5812         wc->refs[level] = 0;
5813         wc->flags[level] = 0;
5814         return 0;
5815 }
5816
5817 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5818                                    struct btrfs_root *root,
5819                                    struct btrfs_path *path,
5820                                    struct walk_control *wc)
5821 {
5822         int level = wc->level;
5823         int lookup_info = 1;
5824         int ret;
5825
5826         while (level >= 0) {
5827                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5828                 if (ret > 0)
5829                         break;
5830
5831                 if (level == 0)
5832                         break;
5833
5834                 if (path->slots[level] >=
5835                     btrfs_header_nritems(path->nodes[level]))
5836                         break;
5837
5838                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5839                 if (ret > 0) {
5840                         path->slots[level]++;
5841                         continue;
5842                 } else if (ret < 0)
5843                         return ret;
5844                 level = wc->level;
5845         }
5846         return 0;
5847 }
5848
5849 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5850                                  struct btrfs_root *root,
5851                                  struct btrfs_path *path,
5852                                  struct walk_control *wc, int max_level)
5853 {
5854         int level = wc->level;
5855         int ret;
5856
5857         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5858         while (level < max_level && path->nodes[level]) {
5859                 wc->level = level;
5860                 if (path->slots[level] + 1 <
5861                     btrfs_header_nritems(path->nodes[level])) {
5862                         path->slots[level]++;
5863                         return 0;
5864                 } else {
5865                         ret = walk_up_proc(trans, root, path, wc);
5866                         if (ret > 0)
5867                                 return 0;
5868
5869                         if (path->locks[level]) {
5870                                 btrfs_tree_unlock(path->nodes[level]);
5871                                 path->locks[level] = 0;
5872                         }
5873                         free_extent_buffer(path->nodes[level]);
5874                         path->nodes[level] = NULL;
5875                         level++;
5876                 }
5877         }
5878         return 1;
5879 }
5880
5881 /*
5882  * drop a subvolume tree.
5883  *
5884  * this function traverses the tree freeing any blocks that only
5885  * referenced by the tree.
5886  *
5887  * when a shared tree block is found. this function decreases its
5888  * reference count by one. if update_ref is true, this function
5889  * also make sure backrefs for the shared block and all lower level
5890  * blocks are properly updated.
5891  */
5892 int btrfs_drop_snapshot(struct btrfs_root *root,
5893                         struct btrfs_block_rsv *block_rsv, int update_ref)
5894 {
5895         struct btrfs_path *path;
5896         struct btrfs_trans_handle *trans;
5897         struct btrfs_root *tree_root = root->fs_info->tree_root;
5898         struct btrfs_root_item *root_item = &root->root_item;
5899         struct walk_control *wc;
5900         struct btrfs_key key;
5901         int err = 0;
5902         int ret;
5903         int level;
5904
5905         path = btrfs_alloc_path();
5906         BUG_ON(!path);
5907
5908         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5909         BUG_ON(!wc);
5910
5911         trans = btrfs_start_transaction(tree_root, 0);
5912         if (block_rsv)
5913                 trans->block_rsv = block_rsv;
5914
5915         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5916                 level = btrfs_header_level(root->node);
5917                 path->nodes[level] = btrfs_lock_root_node(root);
5918                 btrfs_set_lock_blocking(path->nodes[level]);
5919                 path->slots[level] = 0;
5920                 path->locks[level] = 1;
5921                 memset(&wc->update_progress, 0,
5922                        sizeof(wc->update_progress));
5923         } else {
5924                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5925                 memcpy(&wc->update_progress, &key,
5926                        sizeof(wc->update_progress));
5927
5928                 level = root_item->drop_level;
5929                 BUG_ON(level == 0);
5930                 path->lowest_level = level;
5931                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5932                 path->lowest_level = 0;
5933                 if (ret < 0) {
5934                         err = ret;
5935                         goto out;
5936                 }
5937                 WARN_ON(ret > 0);
5938
5939                 /*
5940                  * unlock our path, this is safe because only this
5941                  * function is allowed to delete this snapshot
5942                  */
5943                 btrfs_unlock_up_safe(path, 0);
5944
5945                 level = btrfs_header_level(root->node);
5946                 while (1) {
5947                         btrfs_tree_lock(path->nodes[level]);
5948                         btrfs_set_lock_blocking(path->nodes[level]);
5949
5950                         ret = btrfs_lookup_extent_info(trans, root,
5951                                                 path->nodes[level]->start,
5952                                                 path->nodes[level]->len,
5953                                                 &wc->refs[level],
5954                                                 &wc->flags[level]);
5955                         BUG_ON(ret);
5956                         BUG_ON(wc->refs[level] == 0);
5957
5958                         if (level == root_item->drop_level)
5959                                 break;
5960
5961                         btrfs_tree_unlock(path->nodes[level]);
5962                         WARN_ON(wc->refs[level] != 1);
5963                         level--;
5964                 }
5965         }
5966
5967         wc->level = level;
5968         wc->shared_level = -1;
5969         wc->stage = DROP_REFERENCE;
5970         wc->update_ref = update_ref;
5971         wc->keep_locks = 0;
5972         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5973
5974         while (1) {
5975                 ret = walk_down_tree(trans, root, path, wc);
5976                 if (ret < 0) {
5977                         err = ret;
5978                         break;
5979                 }
5980
5981                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5982                 if (ret < 0) {
5983                         err = ret;
5984                         break;
5985                 }
5986
5987                 if (ret > 0) {
5988                         BUG_ON(wc->stage != DROP_REFERENCE);
5989                         break;
5990                 }
5991
5992                 if (wc->stage == DROP_REFERENCE) {
5993                         level = wc->level;
5994                         btrfs_node_key(path->nodes[level],
5995                                        &root_item->drop_progress,
5996                                        path->slots[level]);
5997                         root_item->drop_level = level;
5998                 }
5999
6000                 BUG_ON(wc->level == 0);
6001                 if (btrfs_should_end_transaction(trans, tree_root)) {
6002                         ret = btrfs_update_root(trans, tree_root,
6003                                                 &root->root_key,
6004                                                 root_item);
6005                         BUG_ON(ret);
6006
6007                         btrfs_end_transaction_throttle(trans, tree_root);
6008                         trans = btrfs_start_transaction(tree_root, 0);
6009                         if (block_rsv)
6010                                 trans->block_rsv = block_rsv;
6011                 }
6012         }
6013         btrfs_release_path(root, path);
6014         BUG_ON(err);
6015
6016         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6017         BUG_ON(ret);
6018
6019         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6020                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6021                                            NULL, NULL);
6022                 BUG_ON(ret < 0);
6023                 if (ret > 0) {
6024                         ret = btrfs_del_orphan_item(trans, tree_root,
6025                                                     root->root_key.objectid);
6026                         BUG_ON(ret);
6027                 }
6028         }
6029
6030         if (root->in_radix) {
6031                 btrfs_free_fs_root(tree_root->fs_info, root);
6032         } else {
6033                 free_extent_buffer(root->node);
6034                 free_extent_buffer(root->commit_root);
6035                 kfree(root);
6036         }
6037 out:
6038         btrfs_end_transaction_throttle(trans, tree_root);
6039         kfree(wc);
6040         btrfs_free_path(path);
6041         return err;
6042 }
6043
6044 /*
6045  * drop subtree rooted at tree block 'node'.
6046  *
6047  * NOTE: this function will unlock and release tree block 'node'
6048  */
6049 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6050                         struct btrfs_root *root,
6051                         struct extent_buffer *node,
6052                         struct extent_buffer *parent)
6053 {
6054         struct btrfs_path *path;
6055         struct walk_control *wc;
6056         int level;
6057         int parent_level;
6058         int ret = 0;
6059         int wret;
6060
6061         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6062
6063         path = btrfs_alloc_path();
6064         BUG_ON(!path);
6065
6066         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6067         BUG_ON(!wc);
6068
6069         btrfs_assert_tree_locked(parent);
6070         parent_level = btrfs_header_level(parent);
6071         extent_buffer_get(parent);
6072         path->nodes[parent_level] = parent;
6073         path->slots[parent_level] = btrfs_header_nritems(parent);
6074
6075         btrfs_assert_tree_locked(node);
6076         level = btrfs_header_level(node);
6077         path->nodes[level] = node;
6078         path->slots[level] = 0;
6079         path->locks[level] = 1;
6080
6081         wc->refs[parent_level] = 1;
6082         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6083         wc->level = level;
6084         wc->shared_level = -1;
6085         wc->stage = DROP_REFERENCE;
6086         wc->update_ref = 0;
6087         wc->keep_locks = 1;
6088         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6089
6090         while (1) {
6091                 wret = walk_down_tree(trans, root, path, wc);
6092                 if (wret < 0) {
6093                         ret = wret;
6094                         break;
6095                 }
6096
6097                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6098                 if (wret < 0)
6099                         ret = wret;
6100                 if (wret != 0)
6101                         break;
6102         }
6103
6104         kfree(wc);
6105         btrfs_free_path(path);
6106         return ret;
6107 }
6108
6109 #if 0
6110 static unsigned long calc_ra(unsigned long start, unsigned long last,
6111                              unsigned long nr)
6112 {
6113         return min(last, start + nr - 1);
6114 }
6115
6116 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6117                                          u64 len)
6118 {
6119         u64 page_start;
6120         u64 page_end;
6121         unsigned long first_index;
6122         unsigned long last_index;
6123         unsigned long i;
6124         struct page *page;
6125         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6126         struct file_ra_state *ra;
6127         struct btrfs_ordered_extent *ordered;
6128         unsigned int total_read = 0;
6129         unsigned int total_dirty = 0;
6130         int ret = 0;
6131
6132         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6133
6134         mutex_lock(&inode->i_mutex);
6135         first_index = start >> PAGE_CACHE_SHIFT;
6136         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6137
6138         /* make sure the dirty trick played by the caller work */
6139         ret = invalidate_inode_pages2_range(inode->i_mapping,
6140                                             first_index, last_index);
6141         if (ret)
6142                 goto out_unlock;
6143
6144         file_ra_state_init(ra, inode->i_mapping);
6145
6146         for (i = first_index ; i <= last_index; i++) {
6147                 if (total_read % ra->ra_pages == 0) {
6148                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6149                                        calc_ra(i, last_index, ra->ra_pages));
6150                 }
6151                 total_read++;
6152 again:
6153                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6154                         BUG_ON(1);
6155                 page = grab_cache_page(inode->i_mapping, i);
6156                 if (!page) {
6157                         ret = -ENOMEM;
6158                         goto out_unlock;
6159                 }
6160                 if (!PageUptodate(page)) {
6161                         btrfs_readpage(NULL, page);
6162                         lock_page(page);
6163                         if (!PageUptodate(page)) {
6164                                 unlock_page(page);
6165                                 page_cache_release(page);
6166                                 ret = -EIO;
6167                                 goto out_unlock;
6168                         }
6169                 }
6170                 wait_on_page_writeback(page);
6171
6172                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6173                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6174                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6175
6176                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6177                 if (ordered) {
6178                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6179                         unlock_page(page);
6180                         page_cache_release(page);
6181                         btrfs_start_ordered_extent(inode, ordered, 1);
6182                         btrfs_put_ordered_extent(ordered);
6183                         goto again;
6184                 }
6185                 set_page_extent_mapped(page);
6186
6187                 if (i == first_index)
6188                         set_extent_bits(io_tree, page_start, page_end,
6189                                         EXTENT_BOUNDARY, GFP_NOFS);
6190                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6191
6192                 set_page_dirty(page);
6193                 total_dirty++;
6194
6195                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6196                 unlock_page(page);
6197                 page_cache_release(page);
6198         }
6199
6200 out_unlock:
6201         kfree(ra);
6202         mutex_unlock(&inode->i_mutex);
6203         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6204         return ret;
6205 }
6206
6207 static noinline int relocate_data_extent(struct inode *reloc_inode,
6208                                          struct btrfs_key *extent_key,
6209                                          u64 offset)
6210 {
6211         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6212         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6213         struct extent_map *em;
6214         u64 start = extent_key->objectid - offset;
6215         u64 end = start + extent_key->offset - 1;
6216
6217         em = alloc_extent_map(GFP_NOFS);
6218         BUG_ON(!em || IS_ERR(em));
6219
6220         em->start = start;
6221         em->len = extent_key->offset;
6222         em->block_len = extent_key->offset;
6223         em->block_start = extent_key->objectid;
6224         em->bdev = root->fs_info->fs_devices->latest_bdev;
6225         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6226
6227         /* setup extent map to cheat btrfs_readpage */
6228         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6229         while (1) {
6230                 int ret;
6231                 write_lock(&em_tree->lock);
6232                 ret = add_extent_mapping(em_tree, em);
6233                 write_unlock(&em_tree->lock);
6234                 if (ret != -EEXIST) {
6235                         free_extent_map(em);
6236                         break;
6237                 }
6238                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6239         }
6240         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6241
6242         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6243 }
6244
6245 struct btrfs_ref_path {
6246         u64 extent_start;
6247         u64 nodes[BTRFS_MAX_LEVEL];
6248         u64 root_objectid;
6249         u64 root_generation;
6250         u64 owner_objectid;
6251         u32 num_refs;
6252         int lowest_level;
6253         int current_level;
6254         int shared_level;
6255
6256         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6257         u64 new_nodes[BTRFS_MAX_LEVEL];
6258 };
6259
6260 struct disk_extent {
6261         u64 ram_bytes;
6262         u64 disk_bytenr;
6263         u64 disk_num_bytes;
6264         u64 offset;
6265         u64 num_bytes;
6266         u8 compression;
6267         u8 encryption;
6268         u16 other_encoding;
6269 };
6270
6271 static int is_cowonly_root(u64 root_objectid)
6272 {
6273         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6274             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6275             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6276             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6277             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6278             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6279                 return 1;
6280         return 0;
6281 }
6282
6283 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6284                                     struct btrfs_root *extent_root,
6285                                     struct btrfs_ref_path *ref_path,
6286                                     int first_time)
6287 {
6288         struct extent_buffer *leaf;
6289         struct btrfs_path *path;
6290         struct btrfs_extent_ref *ref;
6291         struct btrfs_key key;
6292         struct btrfs_key found_key;
6293         u64 bytenr;
6294         u32 nritems;
6295         int level;
6296         int ret = 1;
6297
6298         path = btrfs_alloc_path();
6299         if (!path)
6300                 return -ENOMEM;
6301
6302         if (first_time) {
6303                 ref_path->lowest_level = -1;
6304                 ref_path->current_level = -1;
6305                 ref_path->shared_level = -1;
6306                 goto walk_up;
6307         }
6308 walk_down:
6309         level = ref_path->current_level - 1;
6310         while (level >= -1) {
6311                 u64 parent;
6312                 if (level < ref_path->lowest_level)
6313                         break;
6314
6315                 if (level >= 0)
6316                         bytenr = ref_path->nodes[level];
6317                 else
6318                         bytenr = ref_path->extent_start;
6319                 BUG_ON(bytenr == 0);
6320
6321                 parent = ref_path->nodes[level + 1];
6322                 ref_path->nodes[level + 1] = 0;
6323                 ref_path->current_level = level;
6324                 BUG_ON(parent == 0);
6325
6326                 key.objectid = bytenr;
6327                 key.offset = parent + 1;
6328                 key.type = BTRFS_EXTENT_REF_KEY;
6329
6330                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6331                 if (ret < 0)
6332                         goto out;
6333                 BUG_ON(ret == 0);
6334
6335                 leaf = path->nodes[0];
6336                 nritems = btrfs_header_nritems(leaf);
6337                 if (path->slots[0] >= nritems) {
6338                         ret = btrfs_next_leaf(extent_root, path);
6339                         if (ret < 0)
6340                                 goto out;
6341                         if (ret > 0)
6342                                 goto next;
6343                         leaf = path->nodes[0];
6344                 }
6345
6346                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6347                 if (found_key.objectid == bytenr &&
6348                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6349                         if (level < ref_path->shared_level)
6350                                 ref_path->shared_level = level;
6351                         goto found;
6352                 }
6353 next:
6354                 level--;
6355                 btrfs_release_path(extent_root, path);
6356                 cond_resched();
6357         }
6358         /* reached lowest level */
6359         ret = 1;
6360         goto out;
6361 walk_up:
6362         level = ref_path->current_level;
6363         while (level < BTRFS_MAX_LEVEL - 1) {
6364                 u64 ref_objectid;
6365
6366                 if (level >= 0)
6367                         bytenr = ref_path->nodes[level];
6368                 else
6369                         bytenr = ref_path->extent_start;
6370
6371                 BUG_ON(bytenr == 0);
6372
6373                 key.objectid = bytenr;
6374                 key.offset = 0;
6375                 key.type = BTRFS_EXTENT_REF_KEY;
6376
6377                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6378                 if (ret < 0)
6379                         goto out;
6380
6381                 leaf = path->nodes[0];
6382                 nritems = btrfs_header_nritems(leaf);
6383                 if (path->slots[0] >= nritems) {
6384                         ret = btrfs_next_leaf(extent_root, path);
6385                         if (ret < 0)
6386                                 goto out;
6387                         if (ret > 0) {
6388                                 /* the extent was freed by someone */
6389                                 if (ref_path->lowest_level == level)
6390                                         goto out;
6391                                 btrfs_release_path(extent_root, path);
6392                                 goto walk_down;
6393                         }
6394                         leaf = path->nodes[0];
6395                 }
6396
6397                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6398                 if (found_key.objectid != bytenr ||
6399                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6400                         /* the extent was freed by someone */
6401                         if (ref_path->lowest_level == level) {
6402                                 ret = 1;
6403                                 goto out;
6404                         }
6405                         btrfs_release_path(extent_root, path);
6406                         goto walk_down;
6407                 }
6408 found:
6409                 ref = btrfs_item_ptr(leaf, path->slots[0],
6410                                 struct btrfs_extent_ref);
6411                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6412                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6413                         if (first_time) {
6414                                 level = (int)ref_objectid;
6415                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6416                                 ref_path->lowest_level = level;
6417                                 ref_path->current_level = level;
6418                                 ref_path->nodes[level] = bytenr;
6419                         } else {
6420                                 WARN_ON(ref_objectid != level);
6421                         }
6422                 } else {
6423                         WARN_ON(level != -1);
6424                 }
6425                 first_time = 0;
6426
6427                 if (ref_path->lowest_level == level) {
6428                         ref_path->owner_objectid = ref_objectid;
6429                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6430                 }
6431
6432                 /*
6433                  * the block is tree root or the block isn't in reference
6434                  * counted tree.
6435                  */
6436                 if (found_key.objectid == found_key.offset ||
6437                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6438                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6439                         ref_path->root_generation =
6440                                 btrfs_ref_generation(leaf, ref);
6441                         if (level < 0) {
6442                                 /* special reference from the tree log */
6443                                 ref_path->nodes[0] = found_key.offset;
6444                                 ref_path->current_level = 0;
6445                         }
6446                         ret = 0;
6447                         goto out;
6448                 }
6449
6450                 level++;
6451                 BUG_ON(ref_path->nodes[level] != 0);
6452                 ref_path->nodes[level] = found_key.offset;
6453                 ref_path->current_level = level;
6454
6455                 /*
6456                  * the reference was created in the running transaction,
6457                  * no need to continue walking up.
6458                  */
6459                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6460                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6461                         ref_path->root_generation =
6462                                 btrfs_ref_generation(leaf, ref);
6463                         ret = 0;
6464                         goto out;
6465                 }
6466
6467                 btrfs_release_path(extent_root, path);
6468                 cond_resched();
6469         }
6470         /* reached max tree level, but no tree root found. */
6471         BUG();
6472 out:
6473         btrfs_free_path(path);
6474         return ret;
6475 }
6476
6477 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6478                                 struct btrfs_root *extent_root,
6479                                 struct btrfs_ref_path *ref_path,
6480                                 u64 extent_start)
6481 {
6482         memset(ref_path, 0, sizeof(*ref_path));
6483         ref_path->extent_start = extent_start;
6484
6485         return __next_ref_path(trans, extent_root, ref_path, 1);
6486 }
6487
6488 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6489                                struct btrfs_root *extent_root,
6490                                struct btrfs_ref_path *ref_path)
6491 {
6492         return __next_ref_path(trans, extent_root, ref_path, 0);
6493 }
6494
6495 static noinline int get_new_locations(struct inode *reloc_inode,
6496                                       struct btrfs_key *extent_key,
6497                                       u64 offset, int no_fragment,
6498                                       struct disk_extent **extents,
6499                                       int *nr_extents)
6500 {
6501         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6502         struct btrfs_path *path;
6503         struct btrfs_file_extent_item *fi;
6504         struct extent_buffer *leaf;
6505         struct disk_extent *exts = *extents;
6506         struct btrfs_key found_key;
6507         u64 cur_pos;
6508         u64 last_byte;
6509         u32 nritems;
6510         int nr = 0;
6511         int max = *nr_extents;
6512         int ret;
6513
6514         WARN_ON(!no_fragment && *extents);
6515         if (!exts) {
6516                 max = 1;
6517                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6518                 if (!exts)
6519                         return -ENOMEM;
6520         }
6521
6522         path = btrfs_alloc_path();
6523         BUG_ON(!path);
6524
6525         cur_pos = extent_key->objectid - offset;
6526         last_byte = extent_key->objectid + extent_key->offset;
6527         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6528                                        cur_pos, 0);
6529         if (ret < 0)
6530                 goto out;
6531         if (ret > 0) {
6532                 ret = -ENOENT;
6533                 goto out;
6534         }
6535
6536         while (1) {
6537                 leaf = path->nodes[0];
6538                 nritems = btrfs_header_nritems(leaf);
6539                 if (path->slots[0] >= nritems) {
6540                         ret = btrfs_next_leaf(root, path);
6541                         if (ret < 0)
6542                                 goto out;
6543                         if (ret > 0)
6544                                 break;
6545                         leaf = path->nodes[0];
6546                 }
6547
6548                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6549                 if (found_key.offset != cur_pos ||
6550                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6551                     found_key.objectid != reloc_inode->i_ino)
6552                         break;
6553
6554                 fi = btrfs_item_ptr(leaf, path->slots[0],
6555                                     struct btrfs_file_extent_item);
6556                 if (btrfs_file_extent_type(leaf, fi) !=
6557                     BTRFS_FILE_EXTENT_REG ||
6558                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6559                         break;
6560
6561                 if (nr == max) {
6562                         struct disk_extent *old = exts;
6563                         max *= 2;
6564                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6565                         memcpy(exts, old, sizeof(*exts) * nr);
6566                         if (old != *extents)
6567                                 kfree(old);
6568                 }
6569
6570                 exts[nr].disk_bytenr =
6571                         btrfs_file_extent_disk_bytenr(leaf, fi);
6572                 exts[nr].disk_num_bytes =
6573                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6574                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6575                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6576                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6577                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6578                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6579                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6580                                                                            fi);
6581                 BUG_ON(exts[nr].offset > 0);
6582                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6583                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6584
6585                 cur_pos += exts[nr].num_bytes;
6586                 nr++;
6587
6588                 if (cur_pos + offset >= last_byte)
6589                         break;
6590
6591                 if (no_fragment) {
6592                         ret = 1;
6593                         goto out;
6594                 }
6595                 path->slots[0]++;
6596         }
6597
6598         BUG_ON(cur_pos + offset > last_byte);
6599         if (cur_pos + offset < last_byte) {
6600                 ret = -ENOENT;
6601                 goto out;
6602         }
6603         ret = 0;
6604 out:
6605         btrfs_free_path(path);
6606         if (ret) {
6607                 if (exts != *extents)
6608                         kfree(exts);
6609         } else {
6610                 *extents = exts;
6611                 *nr_extents = nr;
6612         }
6613         return ret;
6614 }
6615
6616 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6617                                         struct btrfs_root *root,
6618                                         struct btrfs_path *path,
6619                                         struct btrfs_key *extent_key,
6620                                         struct btrfs_key *leaf_key,
6621                                         struct btrfs_ref_path *ref_path,
6622                                         struct disk_extent *new_extents,
6623                                         int nr_extents)
6624 {
6625         struct extent_buffer *leaf;
6626         struct btrfs_file_extent_item *fi;
6627         struct inode *inode = NULL;
6628         struct btrfs_key key;
6629         u64 lock_start = 0;
6630         u64 lock_end = 0;
6631         u64 num_bytes;
6632         u64 ext_offset;
6633         u64 search_end = (u64)-1;
6634         u32 nritems;
6635         int nr_scaned = 0;
6636         int extent_locked = 0;
6637         int extent_type;
6638         int ret;
6639
6640         memcpy(&key, leaf_key, sizeof(key));
6641         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6642                 if (key.objectid < ref_path->owner_objectid ||
6643                     (key.objectid == ref_path->owner_objectid &&
6644                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6645                         key.objectid = ref_path->owner_objectid;
6646                         key.type = BTRFS_EXTENT_DATA_KEY;
6647                         key.offset = 0;
6648                 }
6649         }
6650
6651         while (1) {
6652                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6653                 if (ret < 0)
6654                         goto out;
6655
6656                 leaf = path->nodes[0];
6657                 nritems = btrfs_header_nritems(leaf);
6658 next:
6659                 if (extent_locked && ret > 0) {
6660                         /*
6661                          * the file extent item was modified by someone
6662                          * before the extent got locked.
6663                          */
6664                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6665                                       lock_end, GFP_NOFS);
6666                         extent_locked = 0;
6667                 }
6668
6669                 if (path->slots[0] >= nritems) {
6670                         if (++nr_scaned > 2)
6671                                 break;
6672
6673                         BUG_ON(extent_locked);
6674                         ret = btrfs_next_leaf(root, path);
6675                         if (ret < 0)
6676                                 goto out;
6677                         if (ret > 0)
6678                                 break;
6679                         leaf = path->nodes[0];
6680                         nritems = btrfs_header_nritems(leaf);
6681                 }
6682
6683                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6684
6685                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6686                         if ((key.objectid > ref_path->owner_objectid) ||
6687                             (key.objectid == ref_path->owner_objectid &&
6688                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6689                             key.offset >= search_end)
6690                                 break;
6691                 }
6692
6693                 if (inode && key.objectid != inode->i_ino) {
6694                         BUG_ON(extent_locked);
6695                         btrfs_release_path(root, path);
6696                         mutex_unlock(&inode->i_mutex);
6697                         iput(inode);
6698                         inode = NULL;
6699                         continue;
6700                 }
6701
6702                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6703                         path->slots[0]++;
6704                         ret = 1;
6705                         goto next;
6706                 }
6707                 fi = btrfs_item_ptr(leaf, path->slots[0],
6708                                     struct btrfs_file_extent_item);
6709                 extent_type = btrfs_file_extent_type(leaf, fi);
6710                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6711                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6712                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6713                      extent_key->objectid)) {
6714                         path->slots[0]++;
6715                         ret = 1;
6716                         goto next;
6717                 }
6718
6719                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6720                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6721
6722                 if (search_end == (u64)-1) {
6723                         search_end = key.offset - ext_offset +
6724                                 btrfs_file_extent_ram_bytes(leaf, fi);
6725                 }
6726
6727                 if (!extent_locked) {
6728                         lock_start = key.offset;
6729                         lock_end = lock_start + num_bytes - 1;
6730                 } else {
6731                         if (lock_start > key.offset ||
6732                             lock_end + 1 < key.offset + num_bytes) {
6733                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6734                                               lock_start, lock_end, GFP_NOFS);
6735                                 extent_locked = 0;
6736                         }
6737                 }
6738
6739                 if (!inode) {
6740                         btrfs_release_path(root, path);
6741
6742                         inode = btrfs_iget_locked(root->fs_info->sb,
6743                                                   key.objectid, root);
6744                         if (inode->i_state & I_NEW) {
6745                                 BTRFS_I(inode)->root = root;
6746                                 BTRFS_I(inode)->location.objectid =
6747                                         key.objectid;
6748                                 BTRFS_I(inode)->location.type =
6749                                         BTRFS_INODE_ITEM_KEY;
6750                                 BTRFS_I(inode)->location.offset = 0;
6751                                 btrfs_read_locked_inode(inode);
6752                                 unlock_new_inode(inode);
6753                         }
6754                         /*
6755                          * some code call btrfs_commit_transaction while
6756                          * holding the i_mutex, so we can't use mutex_lock
6757                          * here.
6758                          */
6759                         if (is_bad_inode(inode) ||
6760                             !mutex_trylock(&inode->i_mutex)) {
6761                                 iput(inode);
6762                                 inode = NULL;
6763                                 key.offset = (u64)-1;
6764                                 goto skip;
6765                         }
6766                 }
6767
6768                 if (!extent_locked) {
6769                         struct btrfs_ordered_extent *ordered;
6770
6771                         btrfs_release_path(root, path);
6772
6773                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6774                                     lock_end, GFP_NOFS);
6775                         ordered = btrfs_lookup_first_ordered_extent(inode,
6776                                                                     lock_end);
6777                         if (ordered &&
6778                             ordered->file_offset <= lock_end &&
6779                             ordered->file_offset + ordered->len > lock_start) {
6780                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6781                                               lock_start, lock_end, GFP_NOFS);
6782                                 btrfs_start_ordered_extent(inode, ordered, 1);
6783                                 btrfs_put_ordered_extent(ordered);
6784                                 key.offset += num_bytes;
6785                                 goto skip;
6786                         }
6787                         if (ordered)
6788                                 btrfs_put_ordered_extent(ordered);
6789
6790                         extent_locked = 1;
6791                         continue;
6792                 }
6793
6794                 if (nr_extents == 1) {
6795                         /* update extent pointer in place */
6796                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6797                                                 new_extents[0].disk_bytenr);
6798                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6799                                                 new_extents[0].disk_num_bytes);
6800                         btrfs_mark_buffer_dirty(leaf);
6801
6802                         btrfs_drop_extent_cache(inode, key.offset,
6803                                                 key.offset + num_bytes - 1, 0);
6804
6805                         ret = btrfs_inc_extent_ref(trans, root,
6806                                                 new_extents[0].disk_bytenr,
6807                                                 new_extents[0].disk_num_bytes,
6808                                                 leaf->start,
6809                                                 root->root_key.objectid,
6810                                                 trans->transid,
6811                                                 key.objectid);
6812                         BUG_ON(ret);
6813
6814                         ret = btrfs_free_extent(trans, root,
6815                                                 extent_key->objectid,
6816                                                 extent_key->offset,
6817                                                 leaf->start,
6818                                                 btrfs_header_owner(leaf),
6819                                                 btrfs_header_generation(leaf),
6820                                                 key.objectid, 0);
6821                         BUG_ON(ret);
6822
6823                         btrfs_release_path(root, path);
6824                         key.offset += num_bytes;
6825                 } else {
6826                         BUG_ON(1);
6827 #if 0
6828                         u64 alloc_hint;
6829                         u64 extent_len;
6830                         int i;
6831                         /*
6832                          * drop old extent pointer at first, then insert the
6833                          * new pointers one bye one
6834                          */
6835                         btrfs_release_path(root, path);
6836                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
6837                                                  key.offset + num_bytes,
6838                                                  key.offset, &alloc_hint);
6839                         BUG_ON(ret);
6840
6841                         for (i = 0; i < nr_extents; i++) {
6842                                 if (ext_offset >= new_extents[i].num_bytes) {
6843                                         ext_offset -= new_extents[i].num_bytes;
6844                                         continue;
6845                                 }
6846                                 extent_len = min(new_extents[i].num_bytes -
6847                                                  ext_offset, num_bytes);
6848
6849                                 ret = btrfs_insert_empty_item(trans, root,
6850                                                               path, &key,
6851                                                               sizeof(*fi));
6852                                 BUG_ON(ret);
6853
6854                                 leaf = path->nodes[0];
6855                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6856                                                 struct btrfs_file_extent_item);
6857                                 btrfs_set_file_extent_generation(leaf, fi,
6858                                                         trans->transid);
6859                                 btrfs_set_file_extent_type(leaf, fi,
6860                                                         BTRFS_FILE_EXTENT_REG);
6861                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6862                                                 new_extents[i].disk_bytenr);
6863                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6864                                                 new_extents[i].disk_num_bytes);
6865                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6866                                                 new_extents[i].ram_bytes);
6867
6868                                 btrfs_set_file_extent_compression(leaf, fi,
6869                                                 new_extents[i].compression);
6870                                 btrfs_set_file_extent_encryption(leaf, fi,
6871                                                 new_extents[i].encryption);
6872                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6873                                                 new_extents[i].other_encoding);
6874
6875                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6876                                                         extent_len);
6877                                 ext_offset += new_extents[i].offset;
6878                                 btrfs_set_file_extent_offset(leaf, fi,
6879                                                         ext_offset);
6880                                 btrfs_mark_buffer_dirty(leaf);
6881
6882                                 btrfs_drop_extent_cache(inode, key.offset,
6883                                                 key.offset + extent_len - 1, 0);
6884
6885                                 ret = btrfs_inc_extent_ref(trans, root,
6886                                                 new_extents[i].disk_bytenr,
6887                                                 new_extents[i].disk_num_bytes,
6888                                                 leaf->start,
6889                                                 root->root_key.objectid,
6890                                                 trans->transid, key.objectid);
6891                                 BUG_ON(ret);
6892                                 btrfs_release_path(root, path);
6893
6894                                 inode_add_bytes(inode, extent_len);
6895
6896                                 ext_offset = 0;
6897                                 num_bytes -= extent_len;
6898                                 key.offset += extent_len;
6899
6900                                 if (num_bytes == 0)
6901                                         break;
6902                         }
6903                         BUG_ON(i >= nr_extents);
6904 #endif
6905                 }
6906
6907                 if (extent_locked) {
6908                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6909                                       lock_end, GFP_NOFS);
6910                         extent_locked = 0;
6911                 }
6912 skip:
6913                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6914                     key.offset >= search_end)
6915                         break;
6916
6917                 cond_resched();
6918         }
6919         ret = 0;
6920 out:
6921         btrfs_release_path(root, path);
6922         if (inode) {
6923                 mutex_unlock(&inode->i_mutex);
6924                 if (extent_locked) {
6925                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6926                                       lock_end, GFP_NOFS);
6927                 }
6928                 iput(inode);
6929         }
6930         return ret;
6931 }
6932
6933 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6934                                struct btrfs_root *root,
6935                                struct extent_buffer *buf, u64 orig_start)
6936 {
6937         int level;
6938         int ret;
6939
6940         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6941         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6942
6943         level = btrfs_header_level(buf);
6944         if (level == 0) {
6945                 struct btrfs_leaf_ref *ref;
6946                 struct btrfs_leaf_ref *orig_ref;
6947
6948                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6949                 if (!orig_ref)
6950                         return -ENOENT;
6951
6952                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6953                 if (!ref) {
6954                         btrfs_free_leaf_ref(root, orig_ref);
6955                         return -ENOMEM;
6956                 }
6957
6958                 ref->nritems = orig_ref->nritems;
6959                 memcpy(ref->extents, orig_ref->extents,
6960                         sizeof(ref->extents[0]) * ref->nritems);
6961
6962                 btrfs_free_leaf_ref(root, orig_ref);
6963
6964                 ref->root_gen = trans->transid;
6965                 ref->bytenr = buf->start;
6966                 ref->owner = btrfs_header_owner(buf);
6967                 ref->generation = btrfs_header_generation(buf);
6968
6969                 ret = btrfs_add_leaf_ref(root, ref, 0);
6970                 WARN_ON(ret);
6971                 btrfs_free_leaf_ref(root, ref);
6972         }
6973         return 0;
6974 }
6975
6976 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6977                                         struct extent_buffer *leaf,
6978                                         struct btrfs_block_group_cache *group,
6979                                         struct btrfs_root *target_root)
6980 {
6981         struct btrfs_key key;
6982         struct inode *inode = NULL;
6983         struct btrfs_file_extent_item *fi;
6984         struct extent_state *cached_state = NULL;
6985         u64 num_bytes;
6986         u64 skip_objectid = 0;
6987         u32 nritems;
6988         u32 i;
6989
6990         nritems = btrfs_header_nritems(leaf);
6991         for (i = 0; i < nritems; i++) {
6992                 btrfs_item_key_to_cpu(leaf, &key, i);
6993                 if (key.objectid == skip_objectid ||
6994                     key.type != BTRFS_EXTENT_DATA_KEY)
6995                         continue;
6996                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6997                 if (btrfs_file_extent_type(leaf, fi) ==
6998                     BTRFS_FILE_EXTENT_INLINE)
6999                         continue;
7000                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7001                         continue;
7002                 if (!inode || inode->i_ino != key.objectid) {
7003                         iput(inode);
7004                         inode = btrfs_ilookup(target_root->fs_info->sb,
7005                                               key.objectid, target_root, 1);
7006                 }
7007                 if (!inode) {
7008                         skip_objectid = key.objectid;
7009                         continue;
7010                 }
7011                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7012
7013                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7014                                  key.offset + num_bytes - 1, 0, &cached_state,
7015                                  GFP_NOFS);
7016                 btrfs_drop_extent_cache(inode, key.offset,
7017                                         key.offset + num_bytes - 1, 1);
7018                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7019                                      key.offset + num_bytes - 1, &cached_state,
7020                                      GFP_NOFS);
7021                 cond_resched();
7022         }
7023         iput(inode);
7024         return 0;
7025 }
7026
7027 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7028                                         struct btrfs_root *root,
7029                                         struct extent_buffer *leaf,
7030                                         struct btrfs_block_group_cache *group,
7031                                         struct inode *reloc_inode)
7032 {
7033         struct btrfs_key key;
7034         struct btrfs_key extent_key;
7035         struct btrfs_file_extent_item *fi;
7036         struct btrfs_leaf_ref *ref;
7037         struct disk_extent *new_extent;
7038         u64 bytenr;
7039         u64 num_bytes;
7040         u32 nritems;
7041         u32 i;
7042         int ext_index;
7043         int nr_extent;
7044         int ret;
7045
7046         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7047         BUG_ON(!new_extent);
7048
7049         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7050         BUG_ON(!ref);
7051
7052         ext_index = -1;
7053         nritems = btrfs_header_nritems(leaf);
7054         for (i = 0; i < nritems; i++) {
7055                 btrfs_item_key_to_cpu(leaf, &key, i);
7056                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7057                         continue;
7058                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7059                 if (btrfs_file_extent_type(leaf, fi) ==
7060                     BTRFS_FILE_EXTENT_INLINE)
7061                         continue;
7062                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7063                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7064                 if (bytenr == 0)
7065                         continue;
7066
7067                 ext_index++;
7068                 if (bytenr >= group->key.objectid + group->key.offset ||
7069                     bytenr + num_bytes <= group->key.objectid)
7070                         continue;
7071
7072                 extent_key.objectid = bytenr;
7073                 extent_key.offset = num_bytes;
7074                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7075                 nr_extent = 1;
7076                 ret = get_new_locations(reloc_inode, &extent_key,
7077                                         group->key.objectid, 1,
7078                                         &new_extent, &nr_extent);
7079                 if (ret > 0)
7080                         continue;
7081                 BUG_ON(ret < 0);
7082
7083                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7084                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7085                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7086                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7087
7088                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7089                                                 new_extent->disk_bytenr);
7090                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7091                                                 new_extent->disk_num_bytes);
7092                 btrfs_mark_buffer_dirty(leaf);
7093
7094                 ret = btrfs_inc_extent_ref(trans, root,
7095                                         new_extent->disk_bytenr,
7096                                         new_extent->disk_num_bytes,
7097                                         leaf->start,
7098                                         root->root_key.objectid,
7099                                         trans->transid, key.objectid);
7100                 BUG_ON(ret);
7101
7102                 ret = btrfs_free_extent(trans, root,
7103                                         bytenr, num_bytes, leaf->start,
7104                                         btrfs_header_owner(leaf),
7105                                         btrfs_header_generation(leaf),
7106                                         key.objectid, 0);
7107                 BUG_ON(ret);
7108                 cond_resched();
7109         }
7110         kfree(new_extent);
7111         BUG_ON(ext_index + 1 != ref->nritems);
7112         btrfs_free_leaf_ref(root, ref);
7113         return 0;
7114 }
7115
7116 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7117                           struct btrfs_root *root)
7118 {
7119         struct btrfs_root *reloc_root;
7120         int ret;
7121
7122         if (root->reloc_root) {
7123                 reloc_root = root->reloc_root;
7124                 root->reloc_root = NULL;
7125                 list_add(&reloc_root->dead_list,
7126                          &root->fs_info->dead_reloc_roots);
7127
7128                 btrfs_set_root_bytenr(&reloc_root->root_item,
7129                                       reloc_root->node->start);
7130                 btrfs_set_root_level(&root->root_item,
7131                                      btrfs_header_level(reloc_root->node));
7132                 memset(&reloc_root->root_item.drop_progress, 0,
7133                         sizeof(struct btrfs_disk_key));
7134                 reloc_root->root_item.drop_level = 0;
7135
7136                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7137                                         &reloc_root->root_key,
7138                                         &reloc_root->root_item);
7139                 BUG_ON(ret);
7140         }
7141         return 0;
7142 }
7143
7144 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7145 {
7146         struct btrfs_trans_handle *trans;
7147         struct btrfs_root *reloc_root;
7148         struct btrfs_root *prev_root = NULL;
7149         struct list_head dead_roots;
7150         int ret;
7151         unsigned long nr;
7152
7153         INIT_LIST_HEAD(&dead_roots);
7154         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7155
7156         while (!list_empty(&dead_roots)) {
7157                 reloc_root = list_entry(dead_roots.prev,
7158                                         struct btrfs_root, dead_list);
7159                 list_del_init(&reloc_root->dead_list);
7160
7161                 BUG_ON(reloc_root->commit_root != NULL);
7162                 while (1) {
7163                         trans = btrfs_join_transaction(root, 1);
7164                         BUG_ON(!trans);
7165
7166                         mutex_lock(&root->fs_info->drop_mutex);
7167                         ret = btrfs_drop_snapshot(trans, reloc_root);
7168                         if (ret != -EAGAIN)
7169                                 break;
7170                         mutex_unlock(&root->fs_info->drop_mutex);
7171
7172                         nr = trans->blocks_used;
7173                         ret = btrfs_end_transaction(trans, root);
7174                         BUG_ON(ret);
7175                         btrfs_btree_balance_dirty(root, nr);
7176                 }
7177
7178                 free_extent_buffer(reloc_root->node);
7179
7180                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7181                                      &reloc_root->root_key);
7182                 BUG_ON(ret);
7183                 mutex_unlock(&root->fs_info->drop_mutex);
7184
7185                 nr = trans->blocks_used;
7186                 ret = btrfs_end_transaction(trans, root);
7187                 BUG_ON(ret);
7188                 btrfs_btree_balance_dirty(root, nr);
7189
7190                 kfree(prev_root);
7191                 prev_root = reloc_root;
7192         }
7193         if (prev_root) {
7194                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7195                 kfree(prev_root);
7196         }
7197         return 0;
7198 }
7199
7200 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7201 {
7202         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7203         return 0;
7204 }
7205
7206 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7207 {
7208         struct btrfs_root *reloc_root;
7209         struct btrfs_trans_handle *trans;
7210         struct btrfs_key location;
7211         int found;
7212         int ret;
7213
7214         mutex_lock(&root->fs_info->tree_reloc_mutex);
7215         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7216         BUG_ON(ret);
7217         found = !list_empty(&root->fs_info->dead_reloc_roots);
7218         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7219
7220         if (found) {
7221                 trans = btrfs_start_transaction(root, 1);
7222                 BUG_ON(!trans);
7223                 ret = btrfs_commit_transaction(trans, root);
7224                 BUG_ON(ret);
7225         }
7226
7227         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7228         location.offset = (u64)-1;
7229         location.type = BTRFS_ROOT_ITEM_KEY;
7230
7231         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7232         BUG_ON(!reloc_root);
7233         btrfs_orphan_cleanup(reloc_root);
7234         return 0;
7235 }
7236
7237 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7238                                     struct btrfs_root *root)
7239 {
7240         struct btrfs_root *reloc_root;
7241         struct extent_buffer *eb;
7242         struct btrfs_root_item *root_item;
7243         struct btrfs_key root_key;
7244         int ret;
7245
7246         BUG_ON(!root->ref_cows);
7247         if (root->reloc_root)
7248                 return 0;
7249
7250         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7251         BUG_ON(!root_item);
7252
7253         ret = btrfs_copy_root(trans, root, root->commit_root,
7254                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7255         BUG_ON(ret);
7256
7257         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7258         root_key.offset = root->root_key.objectid;
7259         root_key.type = BTRFS_ROOT_ITEM_KEY;
7260
7261         memcpy(root_item, &root->root_item, sizeof(root_item));
7262         btrfs_set_root_refs(root_item, 0);
7263         btrfs_set_root_bytenr(root_item, eb->start);
7264         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7265         btrfs_set_root_generation(root_item, trans->transid);
7266
7267         btrfs_tree_unlock(eb);
7268         free_extent_buffer(eb);
7269
7270         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7271                                 &root_key, root_item);
7272         BUG_ON(ret);
7273         kfree(root_item);
7274
7275         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7276                                                  &root_key);
7277         BUG_ON(!reloc_root);
7278         reloc_root->last_trans = trans->transid;
7279         reloc_root->commit_root = NULL;
7280         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7281
7282         root->reloc_root = reloc_root;
7283         return 0;
7284 }
7285
7286 /*
7287  * Core function of space balance.
7288  *
7289  * The idea is using reloc trees to relocate tree blocks in reference
7290  * counted roots. There is one reloc tree for each subvol, and all
7291  * reloc trees share same root key objectid. Reloc trees are snapshots
7292  * of the latest committed roots of subvols (root->commit_root).
7293  *
7294  * To relocate a tree block referenced by a subvol, there are two steps.
7295  * COW the block through subvol's reloc tree, then update block pointer
7296  * in the subvol to point to the new block. Since all reloc trees share
7297  * same root key objectid, doing special handing for tree blocks owned
7298  * by them is easy. Once a tree block has been COWed in one reloc tree,
7299  * we can use the resulting new block directly when the same block is
7300  * required to COW again through other reloc trees. By this way, relocated
7301  * tree blocks are shared between reloc trees, so they are also shared
7302  * between subvols.
7303  */
7304 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7305                                       struct btrfs_root *root,
7306                                       struct btrfs_path *path,
7307                                       struct btrfs_key *first_key,
7308                                       struct btrfs_ref_path *ref_path,
7309                                       struct btrfs_block_group_cache *group,
7310                                       struct inode *reloc_inode)
7311 {
7312         struct btrfs_root *reloc_root;
7313         struct extent_buffer *eb = NULL;
7314         struct btrfs_key *keys;
7315         u64 *nodes;
7316         int level;
7317         int shared_level;
7318         int lowest_level = 0;
7319         int ret;
7320
7321         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7322                 lowest_level = ref_path->owner_objectid;
7323
7324         if (!root->ref_cows) {
7325                 path->lowest_level = lowest_level;
7326                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7327                 BUG_ON(ret < 0);
7328                 path->lowest_level = 0;
7329                 btrfs_release_path(root, path);
7330                 return 0;
7331         }
7332
7333         mutex_lock(&root->fs_info->tree_reloc_mutex);
7334         ret = init_reloc_tree(trans, root);
7335         BUG_ON(ret);
7336         reloc_root = root->reloc_root;
7337
7338         shared_level = ref_path->shared_level;
7339         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7340
7341         keys = ref_path->node_keys;
7342         nodes = ref_path->new_nodes;
7343         memset(&keys[shared_level + 1], 0,
7344                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7345         memset(&nodes[shared_level + 1], 0,
7346                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7347
7348         if (nodes[lowest_level] == 0) {
7349                 path->lowest_level = lowest_level;
7350                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7351                                         0, 1);
7352                 BUG_ON(ret);
7353                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7354                         eb = path->nodes[level];
7355                         if (!eb || eb == reloc_root->node)
7356                                 break;
7357                         nodes[level] = eb->start;
7358                         if (level == 0)
7359                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7360                         else
7361                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7362                 }
7363                 if (nodes[0] &&
7364                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7365                         eb = path->nodes[0];
7366                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7367                                                       group, reloc_inode);
7368                         BUG_ON(ret);
7369                 }
7370                 btrfs_release_path(reloc_root, path);
7371         } else {
7372                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7373                                        lowest_level);
7374                 BUG_ON(ret);
7375         }
7376
7377         /*
7378          * replace tree blocks in the fs tree with tree blocks in
7379          * the reloc tree.
7380          */
7381         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7382         BUG_ON(ret < 0);
7383
7384         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7385                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7386                                         0, 0);
7387                 BUG_ON(ret);
7388                 extent_buffer_get(path->nodes[0]);
7389                 eb = path->nodes[0];
7390                 btrfs_release_path(reloc_root, path);
7391                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7392                 BUG_ON(ret);
7393                 free_extent_buffer(eb);
7394         }
7395
7396         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7397         path->lowest_level = 0;
7398         return 0;
7399 }
7400
7401 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7402                                         struct btrfs_root *root,
7403                                         struct btrfs_path *path,
7404                                         struct btrfs_key *first_key,
7405                                         struct btrfs_ref_path *ref_path)
7406 {
7407         int ret;
7408
7409         ret = relocate_one_path(trans, root, path, first_key,
7410                                 ref_path, NULL, NULL);
7411         BUG_ON(ret);
7412
7413         return 0;
7414 }
7415
7416 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7417                                     struct btrfs_root *extent_root,
7418                                     struct btrfs_path *path,
7419                                     struct btrfs_key *extent_key)
7420 {
7421         int ret;
7422
7423         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7424         if (ret)
7425                 goto out;
7426         ret = btrfs_del_item(trans, extent_root, path);
7427 out:
7428         btrfs_release_path(extent_root, path);
7429         return ret;
7430 }
7431
7432 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7433                                                 struct btrfs_ref_path *ref_path)
7434 {
7435         struct btrfs_key root_key;
7436
7437         root_key.objectid = ref_path->root_objectid;
7438         root_key.type = BTRFS_ROOT_ITEM_KEY;
7439         if (is_cowonly_root(ref_path->root_objectid))
7440                 root_key.offset = 0;
7441         else
7442                 root_key.offset = (u64)-1;
7443
7444         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7445 }
7446
7447 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7448                                         struct btrfs_path *path,
7449                                         struct btrfs_key *extent_key,
7450                                         struct btrfs_block_group_cache *group,
7451                                         struct inode *reloc_inode, int pass)
7452 {
7453         struct btrfs_trans_handle *trans;
7454         struct btrfs_root *found_root;
7455         struct btrfs_ref_path *ref_path = NULL;
7456         struct disk_extent *new_extents = NULL;
7457         int nr_extents = 0;
7458         int loops;
7459         int ret;
7460         int level;
7461         struct btrfs_key first_key;
7462         u64 prev_block = 0;
7463
7464
7465         trans = btrfs_start_transaction(extent_root, 1);
7466         BUG_ON(!trans);
7467
7468         if (extent_key->objectid == 0) {
7469                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7470                 goto out;
7471         }
7472
7473         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7474         if (!ref_path) {
7475                 ret = -ENOMEM;
7476                 goto out;
7477         }
7478
7479         for (loops = 0; ; loops++) {
7480                 if (loops == 0) {
7481                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7482                                                    extent_key->objectid);
7483                 } else {
7484                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7485                 }
7486                 if (ret < 0)
7487                         goto out;
7488                 if (ret > 0)
7489                         break;
7490
7491                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7492                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7493                         continue;
7494
7495                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7496                 BUG_ON(!found_root);
7497                 /*
7498                  * for reference counted tree, only process reference paths
7499                  * rooted at the latest committed root.
7500                  */
7501                 if (found_root->ref_cows &&
7502                     ref_path->root_generation != found_root->root_key.offset)
7503                         continue;
7504
7505                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7506                         if (pass == 0) {
7507                                 /*
7508                                  * copy data extents to new locations
7509                                  */
7510                                 u64 group_start = group->key.objectid;
7511                                 ret = relocate_data_extent(reloc_inode,
7512                                                            extent_key,
7513                                                            group_start);
7514                                 if (ret < 0)
7515                                         goto out;
7516                                 break;
7517                         }
7518                         level = 0;
7519                 } else {
7520                         level = ref_path->owner_objectid;
7521                 }
7522
7523                 if (prev_block != ref_path->nodes[level]) {
7524                         struct extent_buffer *eb;
7525                         u64 block_start = ref_path->nodes[level];
7526                         u64 block_size = btrfs_level_size(found_root, level);
7527
7528                         eb = read_tree_block(found_root, block_start,
7529                                              block_size, 0);
7530                         btrfs_tree_lock(eb);
7531                         BUG_ON(level != btrfs_header_level(eb));
7532
7533                         if (level == 0)
7534                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7535                         else
7536                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7537
7538                         btrfs_tree_unlock(eb);
7539                         free_extent_buffer(eb);
7540                         prev_block = block_start;
7541                 }
7542
7543                 mutex_lock(&extent_root->fs_info->trans_mutex);
7544                 btrfs_record_root_in_trans(found_root);
7545                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7546                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7547                         /*
7548                          * try to update data extent references while
7549                          * keeping metadata shared between snapshots.
7550                          */
7551                         if (pass == 1) {
7552                                 ret = relocate_one_path(trans, found_root,
7553                                                 path, &first_key, ref_path,
7554                                                 group, reloc_inode);
7555                                 if (ret < 0)
7556                                         goto out;
7557                                 continue;
7558                         }
7559                         /*
7560                          * use fallback method to process the remaining
7561                          * references.
7562                          */
7563                         if (!new_extents) {
7564                                 u64 group_start = group->key.objectid;
7565                                 new_extents = kmalloc(sizeof(*new_extents),
7566                                                       GFP_NOFS);
7567                                 nr_extents = 1;
7568                                 ret = get_new_locations(reloc_inode,
7569                                                         extent_key,
7570                                                         group_start, 1,
7571                                                         &new_extents,
7572                                                         &nr_extents);
7573                                 if (ret)
7574                                         goto out;
7575                         }
7576                         ret = replace_one_extent(trans, found_root,
7577                                                 path, extent_key,
7578                                                 &first_key, ref_path,
7579                                                 new_extents, nr_extents);
7580                 } else {
7581                         ret = relocate_tree_block(trans, found_root, path,
7582                                                   &first_key, ref_path);
7583                 }
7584                 if (ret < 0)
7585                         goto out;
7586         }
7587         ret = 0;
7588 out:
7589         btrfs_end_transaction(trans, extent_root);
7590         kfree(new_extents);
7591         kfree(ref_path);
7592         return ret;
7593 }
7594 #endif
7595
7596 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7597 {
7598         u64 num_devices;
7599         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7600                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7601
7602         num_devices = root->fs_info->fs_devices->rw_devices;
7603         if (num_devices == 1) {
7604                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7605                 stripped = flags & ~stripped;
7606
7607                 /* turn raid0 into single device chunks */
7608                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7609                         return stripped;
7610
7611                 /* turn mirroring into duplication */
7612                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7613                              BTRFS_BLOCK_GROUP_RAID10))
7614                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7615                 return flags;
7616         } else {
7617                 /* they already had raid on here, just return */
7618                 if (flags & stripped)
7619                         return flags;
7620
7621                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7622                 stripped = flags & ~stripped;
7623
7624                 /* switch duplicated blocks with raid1 */
7625                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7626                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7627
7628                 /* turn single device chunks into raid0 */
7629                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7630         }
7631         return flags;
7632 }
7633
7634 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7635 {
7636         struct btrfs_space_info *sinfo = cache->space_info;
7637         u64 num_bytes;
7638         int ret = -ENOSPC;
7639
7640         if (cache->ro)
7641                 return 0;
7642
7643         spin_lock(&sinfo->lock);
7644         spin_lock(&cache->lock);
7645         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7646                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7647
7648         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7649             sinfo->bytes_may_use + sinfo->bytes_readonly +
7650             cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7651                 sinfo->bytes_readonly += num_bytes;
7652                 sinfo->bytes_reserved += cache->reserved_pinned;
7653                 cache->reserved_pinned = 0;
7654                 cache->ro = 1;
7655                 ret = 0;
7656         }
7657         spin_unlock(&cache->lock);
7658         spin_unlock(&sinfo->lock);
7659         return ret;
7660 }
7661
7662 int btrfs_set_block_group_ro(struct btrfs_root *root,
7663                              struct btrfs_block_group_cache *cache)
7664
7665 {
7666         struct btrfs_trans_handle *trans;
7667         u64 alloc_flags;
7668         int ret;
7669
7670         BUG_ON(cache->ro);
7671
7672         trans = btrfs_join_transaction(root, 1);
7673         BUG_ON(IS_ERR(trans));
7674
7675         alloc_flags = update_block_group_flags(root, cache->flags);
7676         if (alloc_flags != cache->flags)
7677                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7678
7679         ret = set_block_group_ro(cache);
7680         if (!ret)
7681                 goto out;
7682         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7683         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7684         if (ret < 0)
7685                 goto out;
7686         ret = set_block_group_ro(cache);
7687 out:
7688         btrfs_end_transaction(trans, root);
7689         return ret;
7690 }
7691
7692 int btrfs_set_block_group_rw(struct btrfs_root *root,
7693                               struct btrfs_block_group_cache *cache)
7694 {
7695         struct btrfs_space_info *sinfo = cache->space_info;
7696         u64 num_bytes;
7697
7698         BUG_ON(!cache->ro);
7699
7700         spin_lock(&sinfo->lock);
7701         spin_lock(&cache->lock);
7702         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7703                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7704         sinfo->bytes_readonly -= num_bytes;
7705         cache->ro = 0;
7706         spin_unlock(&cache->lock);
7707         spin_unlock(&sinfo->lock);
7708         return 0;
7709 }
7710
7711 /*
7712  * checks to see if its even possible to relocate this block group.
7713  *
7714  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7715  * ok to go ahead and try.
7716  */
7717 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7718 {
7719         struct btrfs_block_group_cache *block_group;
7720         struct btrfs_space_info *space_info;
7721         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7722         struct btrfs_device *device;
7723         int full = 0;
7724         int ret = 0;
7725
7726         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7727
7728         /* odd, couldn't find the block group, leave it alone */
7729         if (!block_group)
7730                 return -1;
7731
7732         /* no bytes used, we're good */
7733         if (!btrfs_block_group_used(&block_group->item))
7734                 goto out;
7735
7736         space_info = block_group->space_info;
7737         spin_lock(&space_info->lock);
7738
7739         full = space_info->full;
7740
7741         /*
7742          * if this is the last block group we have in this space, we can't
7743          * relocate it unless we're able to allocate a new chunk below.
7744          *
7745          * Otherwise, we need to make sure we have room in the space to handle
7746          * all of the extents from this block group.  If we can, we're good
7747          */
7748         if ((space_info->total_bytes != block_group->key.offset) &&
7749            (space_info->bytes_used + space_info->bytes_reserved +
7750             space_info->bytes_pinned + space_info->bytes_readonly +
7751             btrfs_block_group_used(&block_group->item) <
7752             space_info->total_bytes)) {
7753                 spin_unlock(&space_info->lock);
7754                 goto out;
7755         }
7756         spin_unlock(&space_info->lock);
7757
7758         /*
7759          * ok we don't have enough space, but maybe we have free space on our
7760          * devices to allocate new chunks for relocation, so loop through our
7761          * alloc devices and guess if we have enough space.  However, if we
7762          * were marked as full, then we know there aren't enough chunks, and we
7763          * can just return.
7764          */
7765         ret = -1;
7766         if (full)
7767                 goto out;
7768
7769         mutex_lock(&root->fs_info->chunk_mutex);
7770         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7771                 u64 min_free = btrfs_block_group_used(&block_group->item);
7772                 u64 dev_offset, max_avail;
7773
7774                 /*
7775                  * check to make sure we can actually find a chunk with enough
7776                  * space to fit our block group in.
7777                  */
7778                 if (device->total_bytes > device->bytes_used + min_free) {
7779                         ret = find_free_dev_extent(NULL, device, min_free,
7780                                                    &dev_offset, &max_avail);
7781                         if (!ret)
7782                                 break;
7783                         ret = -1;
7784                 }
7785         }
7786         mutex_unlock(&root->fs_info->chunk_mutex);
7787 out:
7788         btrfs_put_block_group(block_group);
7789         return ret;
7790 }
7791
7792 static int find_first_block_group(struct btrfs_root *root,
7793                 struct btrfs_path *path, struct btrfs_key *key)
7794 {
7795         int ret = 0;
7796         struct btrfs_key found_key;
7797         struct extent_buffer *leaf;
7798         int slot;
7799
7800         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7801         if (ret < 0)
7802                 goto out;
7803
7804         while (1) {
7805                 slot = path->slots[0];
7806                 leaf = path->nodes[0];
7807                 if (slot >= btrfs_header_nritems(leaf)) {
7808                         ret = btrfs_next_leaf(root, path);
7809                         if (ret == 0)
7810                                 continue;
7811                         if (ret < 0)
7812                                 goto out;
7813                         break;
7814                 }
7815                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7816
7817                 if (found_key.objectid >= key->objectid &&
7818                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7819                         ret = 0;
7820                         goto out;
7821                 }
7822                 path->slots[0]++;
7823         }
7824 out:
7825         return ret;
7826 }
7827
7828 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7829 {
7830         struct btrfs_block_group_cache *block_group;
7831         struct btrfs_space_info *space_info;
7832         struct btrfs_caching_control *caching_ctl;
7833         struct rb_node *n;
7834
7835         down_write(&info->extent_commit_sem);
7836         while (!list_empty(&info->caching_block_groups)) {
7837                 caching_ctl = list_entry(info->caching_block_groups.next,
7838                                          struct btrfs_caching_control, list);
7839                 list_del(&caching_ctl->list);
7840                 put_caching_control(caching_ctl);
7841         }
7842         up_write(&info->extent_commit_sem);
7843
7844         spin_lock(&info->block_group_cache_lock);
7845         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7846                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7847                                        cache_node);
7848                 rb_erase(&block_group->cache_node,
7849                          &info->block_group_cache_tree);
7850                 spin_unlock(&info->block_group_cache_lock);
7851
7852                 down_write(&block_group->space_info->groups_sem);
7853                 list_del(&block_group->list);
7854                 up_write(&block_group->space_info->groups_sem);
7855
7856                 if (block_group->cached == BTRFS_CACHE_STARTED)
7857                         wait_block_group_cache_done(block_group);
7858
7859                 btrfs_remove_free_space_cache(block_group);
7860                 btrfs_put_block_group(block_group);
7861
7862                 spin_lock(&info->block_group_cache_lock);
7863         }
7864         spin_unlock(&info->block_group_cache_lock);
7865
7866         /* now that all the block groups are freed, go through and
7867          * free all the space_info structs.  This is only called during
7868          * the final stages of unmount, and so we know nobody is
7869          * using them.  We call synchronize_rcu() once before we start,
7870          * just to be on the safe side.
7871          */
7872         synchronize_rcu();
7873
7874         release_global_block_rsv(info);
7875
7876         while(!list_empty(&info->space_info)) {
7877                 space_info = list_entry(info->space_info.next,
7878                                         struct btrfs_space_info,
7879                                         list);
7880                 if (space_info->bytes_pinned > 0 ||
7881                     space_info->bytes_reserved > 0) {
7882                         WARN_ON(1);
7883                         dump_space_info(space_info, 0, 0);
7884                 }
7885                 list_del(&space_info->list);
7886                 kfree(space_info);
7887         }
7888         return 0;
7889 }
7890
7891 static void __link_block_group(struct btrfs_space_info *space_info,
7892                                struct btrfs_block_group_cache *cache)
7893 {
7894         int index = get_block_group_index(cache);
7895
7896         down_write(&space_info->groups_sem);
7897         list_add_tail(&cache->list, &space_info->block_groups[index]);
7898         up_write(&space_info->groups_sem);
7899 }
7900
7901 int btrfs_read_block_groups(struct btrfs_root *root)
7902 {
7903         struct btrfs_path *path;
7904         int ret;
7905         struct btrfs_block_group_cache *cache;
7906         struct btrfs_fs_info *info = root->fs_info;
7907         struct btrfs_space_info *space_info;
7908         struct btrfs_key key;
7909         struct btrfs_key found_key;
7910         struct extent_buffer *leaf;
7911
7912         root = info->extent_root;
7913         key.objectid = 0;
7914         key.offset = 0;
7915         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7916         path = btrfs_alloc_path();
7917         if (!path)
7918                 return -ENOMEM;
7919
7920         while (1) {
7921                 ret = find_first_block_group(root, path, &key);
7922                 if (ret > 0)
7923                         break;
7924                 if (ret != 0)
7925                         goto error;
7926
7927                 leaf = path->nodes[0];
7928                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7929                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7930                 if (!cache) {
7931                         ret = -ENOMEM;
7932                         goto error;
7933                 }
7934
7935                 atomic_set(&cache->count, 1);
7936                 spin_lock_init(&cache->lock);
7937                 spin_lock_init(&cache->tree_lock);
7938                 cache->fs_info = info;
7939                 INIT_LIST_HEAD(&cache->list);
7940                 INIT_LIST_HEAD(&cache->cluster_list);
7941
7942                 /*
7943                  * we only want to have 32k of ram per block group for keeping
7944                  * track of free space, and if we pass 1/2 of that we want to
7945                  * start converting things over to using bitmaps
7946                  */
7947                 cache->extents_thresh = ((1024 * 32) / 2) /
7948                         sizeof(struct btrfs_free_space);
7949
7950                 read_extent_buffer(leaf, &cache->item,
7951                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7952                                    sizeof(cache->item));
7953                 memcpy(&cache->key, &found_key, sizeof(found_key));
7954
7955                 key.objectid = found_key.objectid + found_key.offset;
7956                 btrfs_release_path(root, path);
7957                 cache->flags = btrfs_block_group_flags(&cache->item);
7958                 cache->sectorsize = root->sectorsize;
7959
7960                 /*
7961                  * check for two cases, either we are full, and therefore
7962                  * don't need to bother with the caching work since we won't
7963                  * find any space, or we are empty, and we can just add all
7964                  * the space in and be done with it.  This saves us _alot_ of
7965                  * time, particularly in the full case.
7966                  */
7967                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7968                         exclude_super_stripes(root, cache);
7969                         cache->last_byte_to_unpin = (u64)-1;
7970                         cache->cached = BTRFS_CACHE_FINISHED;
7971                         free_excluded_extents(root, cache);
7972                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7973                         exclude_super_stripes(root, cache);
7974                         cache->last_byte_to_unpin = (u64)-1;
7975                         cache->cached = BTRFS_CACHE_FINISHED;
7976                         add_new_free_space(cache, root->fs_info,
7977                                            found_key.objectid,
7978                                            found_key.objectid +
7979                                            found_key.offset);
7980                         free_excluded_extents(root, cache);
7981                 }
7982
7983                 ret = update_space_info(info, cache->flags, found_key.offset,
7984                                         btrfs_block_group_used(&cache->item),
7985                                         &space_info);
7986                 BUG_ON(ret);
7987                 cache->space_info = space_info;
7988                 spin_lock(&cache->space_info->lock);
7989                 cache->space_info->bytes_readonly += cache->bytes_super;
7990                 spin_unlock(&cache->space_info->lock);
7991
7992                 __link_block_group(space_info, cache);
7993
7994                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7995                 BUG_ON(ret);
7996
7997                 set_avail_alloc_bits(root->fs_info, cache->flags);
7998                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7999                         set_block_group_ro(cache);
8000         }
8001
8002         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8003                 if (!(get_alloc_profile(root, space_info->flags) &
8004                       (BTRFS_BLOCK_GROUP_RAID10 |
8005                        BTRFS_BLOCK_GROUP_RAID1 |
8006                        BTRFS_BLOCK_GROUP_DUP)))
8007                         continue;
8008                 /*
8009                  * avoid allocating from un-mirrored block group if there are
8010                  * mirrored block groups.
8011                  */
8012                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8013                         set_block_group_ro(cache);
8014                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8015                         set_block_group_ro(cache);
8016         }
8017
8018         init_global_block_rsv(info);
8019         ret = 0;
8020 error:
8021         btrfs_free_path(path);
8022         return ret;
8023 }
8024
8025 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8026                            struct btrfs_root *root, u64 bytes_used,
8027                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8028                            u64 size)
8029 {
8030         int ret;
8031         struct btrfs_root *extent_root;
8032         struct btrfs_block_group_cache *cache;
8033
8034         extent_root = root->fs_info->extent_root;
8035
8036         root->fs_info->last_trans_log_full_commit = trans->transid;
8037
8038         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8039         if (!cache)
8040                 return -ENOMEM;
8041
8042         cache->key.objectid = chunk_offset;
8043         cache->key.offset = size;
8044         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8045         cache->sectorsize = root->sectorsize;
8046
8047         /*
8048          * we only want to have 32k of ram per block group for keeping track
8049          * of free space, and if we pass 1/2 of that we want to start
8050          * converting things over to using bitmaps
8051          */
8052         cache->extents_thresh = ((1024 * 32) / 2) /
8053                 sizeof(struct btrfs_free_space);
8054         atomic_set(&cache->count, 1);
8055         spin_lock_init(&cache->lock);
8056         spin_lock_init(&cache->tree_lock);
8057         INIT_LIST_HEAD(&cache->list);
8058         INIT_LIST_HEAD(&cache->cluster_list);
8059
8060         btrfs_set_block_group_used(&cache->item, bytes_used);
8061         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8062         cache->flags = type;
8063         btrfs_set_block_group_flags(&cache->item, type);
8064
8065         cache->last_byte_to_unpin = (u64)-1;
8066         cache->cached = BTRFS_CACHE_FINISHED;
8067         exclude_super_stripes(root, cache);
8068
8069         add_new_free_space(cache, root->fs_info, chunk_offset,
8070                            chunk_offset + size);
8071
8072         free_excluded_extents(root, cache);
8073
8074         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8075                                 &cache->space_info);
8076         BUG_ON(ret);
8077
8078         spin_lock(&cache->space_info->lock);
8079         cache->space_info->bytes_readonly += cache->bytes_super;
8080         spin_unlock(&cache->space_info->lock);
8081
8082         __link_block_group(cache->space_info, cache);
8083
8084         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8085         BUG_ON(ret);
8086
8087         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8088                                 sizeof(cache->item));
8089         BUG_ON(ret);
8090
8091         set_avail_alloc_bits(extent_root->fs_info, type);
8092
8093         return 0;
8094 }
8095
8096 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8097                              struct btrfs_root *root, u64 group_start)
8098 {
8099         struct btrfs_path *path;
8100         struct btrfs_block_group_cache *block_group;
8101         struct btrfs_free_cluster *cluster;
8102         struct btrfs_key key;
8103         int ret;
8104         int factor;
8105
8106         root = root->fs_info->extent_root;
8107
8108         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8109         BUG_ON(!block_group);
8110         BUG_ON(!block_group->ro);
8111
8112         memcpy(&key, &block_group->key, sizeof(key));
8113         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8114                                   BTRFS_BLOCK_GROUP_RAID1 |
8115                                   BTRFS_BLOCK_GROUP_RAID10))
8116                 factor = 2;
8117         else
8118                 factor = 1;
8119
8120         /* make sure this block group isn't part of an allocation cluster */
8121         cluster = &root->fs_info->data_alloc_cluster;
8122         spin_lock(&cluster->refill_lock);
8123         btrfs_return_cluster_to_free_space(block_group, cluster);
8124         spin_unlock(&cluster->refill_lock);
8125
8126         /*
8127          * make sure this block group isn't part of a metadata
8128          * allocation cluster
8129          */
8130         cluster = &root->fs_info->meta_alloc_cluster;
8131         spin_lock(&cluster->refill_lock);
8132         btrfs_return_cluster_to_free_space(block_group, cluster);
8133         spin_unlock(&cluster->refill_lock);
8134
8135         path = btrfs_alloc_path();
8136         BUG_ON(!path);
8137
8138         spin_lock(&root->fs_info->block_group_cache_lock);
8139         rb_erase(&block_group->cache_node,
8140                  &root->fs_info->block_group_cache_tree);
8141         spin_unlock(&root->fs_info->block_group_cache_lock);
8142
8143         down_write(&block_group->space_info->groups_sem);
8144         /*
8145          * we must use list_del_init so people can check to see if they
8146          * are still on the list after taking the semaphore
8147          */
8148         list_del_init(&block_group->list);
8149         up_write(&block_group->space_info->groups_sem);
8150
8151         if (block_group->cached == BTRFS_CACHE_STARTED)
8152                 wait_block_group_cache_done(block_group);
8153
8154         btrfs_remove_free_space_cache(block_group);
8155
8156         spin_lock(&block_group->space_info->lock);
8157         block_group->space_info->total_bytes -= block_group->key.offset;
8158         block_group->space_info->bytes_readonly -= block_group->key.offset;
8159         block_group->space_info->disk_total -= block_group->key.offset * factor;
8160         spin_unlock(&block_group->space_info->lock);
8161
8162         btrfs_clear_space_info_full(root->fs_info);
8163
8164         btrfs_put_block_group(block_group);
8165         btrfs_put_block_group(block_group);
8166
8167         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8168         if (ret > 0)
8169                 ret = -EIO;
8170         if (ret < 0)
8171                 goto out;
8172
8173         ret = btrfs_del_item(trans, root, path);
8174 out:
8175         btrfs_free_path(path);
8176         return ret;
8177 }