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