SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/kthread.h>
  25 #include <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_delalloc +
2899                 data_sinfo->bytes_reserved + data_sinfo->bytes_pinned +
2900                 data_sinfo->bytes_readonly + data_sinfo->bytes_may_use +
2901                 data_sinfo->bytes_super;
2902
2903         if (used + bytes > data_sinfo->total_bytes) {
2904                 struct btrfs_trans_handle *trans;
2905
2906                 /*
2907                  * if we don't have enough free bytes in this space then we need
2908                  * to alloc a new chunk.
2909                  */
2910                 if (!data_sinfo->full) {
2911                         u64 alloc_target;
2912
2913                         data_sinfo->force_alloc = 1;
2914                         spin_unlock(&data_sinfo->lock);
2915 alloc:
2916                         alloc_target = btrfs_get_alloc_profile(root, 1);
2917                         trans = btrfs_join_transaction(root, 1);
2918                         if (IS_ERR(trans))
2919                                 return PTR_ERR(trans);
2920
2921                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2922                                              bytes + 2 * 1024 * 1024,
2923                                              alloc_target, 0);
2924                         btrfs_end_transaction(trans, root);
2925                         if (ret)
2926                                 return ret;
2927
2928                         if (!data_sinfo) {
2929                                 btrfs_set_inode_space_info(root, inode);
2930                                 data_sinfo = BTRFS_I(inode)->space_info;
2931                         }
2932                         goto again;
2933                 }
2934                 spin_unlock(&data_sinfo->lock);
2935
2936                 /* commit the current transaction and try again */
2937                 if (!committed && !root->fs_info->open_ioctl_trans) {
2938                         committed = 1;
2939                         trans = btrfs_join_transaction(root, 1);
2940                         if (IS_ERR(trans))
2941                                 return PTR_ERR(trans);
2942                         ret = btrfs_commit_transaction(trans, root);
2943                         if (ret)
2944                                 return ret;
2945                         goto again;
2946                 }
2947
2948                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
2949                        ", %llu bytes_used, %llu bytes_reserved, "
2950                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
2951                        "%llu total\n", (unsigned long long)bytes,
2952                        (unsigned long long)data_sinfo->bytes_delalloc,
2953                        (unsigned long long)data_sinfo->bytes_used,
2954                        (unsigned long long)data_sinfo->bytes_reserved,
2955                        (unsigned long long)data_sinfo->bytes_pinned,
2956                        (unsigned long long)data_sinfo->bytes_readonly,
2957                        (unsigned long long)data_sinfo->bytes_may_use,
2958                        (unsigned long long)data_sinfo->total_bytes);
2959                 return -ENOSPC;
2960         }
2961         data_sinfo->bytes_may_use += bytes;
2962         BTRFS_I(inode)->reserved_bytes += bytes;
2963         spin_unlock(&data_sinfo->lock);
2964
2965         return 0;
2966 }
2967
2968 /*
2969  * called when we are clearing an delalloc extent from the
2970  * inode's io_tree or there was an error for whatever reason
2971  * after calling btrfs_check_data_free_space
2972  */
2973 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
2974 {
2975         struct btrfs_root *root = BTRFS_I(inode)->root;
2976         struct btrfs_space_info *data_sinfo;
2977
2978         /* make sure bytes are sectorsize aligned */
2979         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
2980
2981         data_sinfo = BTRFS_I(inode)->space_info;
2982         spin_lock(&data_sinfo->lock);
2983         data_sinfo->bytes_may_use -= bytes;
2984         BTRFS_I(inode)->reserved_bytes -= bytes;
2985         spin_unlock(&data_sinfo->lock);
2986 }
2987
2988 static void force_metadata_allocation(struct btrfs_fs_info *info)
2989 {
2990         struct list_head *head = &info->space_info;
2991         struct btrfs_space_info *found;
2992
2993         rcu_read_lock();
2994         list_for_each_entry_rcu(found, head, list) {
2995                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
2996                         found->force_alloc = 1;
2997         }
2998         rcu_read_unlock();
2999 }
3000
3001 static int should_alloc_chunk(struct btrfs_space_info *sinfo,
3002                               u64 alloc_bytes)
3003 {
3004         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3005
3006         if (sinfo->bytes_used + sinfo->bytes_reserved +
3007             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3008                 return 0;
3009
3010         if (sinfo->bytes_used + sinfo->bytes_reserved +
3011             alloc_bytes < div_factor(num_bytes, 8))
3012                 return 0;
3013
3014         return 1;
3015 }
3016
3017 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3018                           struct btrfs_root *extent_root, u64 alloc_bytes,
3019                           u64 flags, int force)
3020 {
3021         struct btrfs_space_info *space_info;
3022         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3023         int ret = 0;
3024
3025         mutex_lock(&fs_info->chunk_mutex);
3026
3027         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3028
3029         space_info = __find_space_info(extent_root->fs_info, flags);
3030         if (!space_info) {
3031                 ret = update_space_info(extent_root->fs_info, flags,
3032                                         0, 0, &space_info);
3033                 BUG_ON(ret);
3034         }
3035         BUG_ON(!space_info);
3036
3037         spin_lock(&space_info->lock);
3038         if (space_info->force_alloc)
3039                 force = 1;
3040         if (space_info->full) {
3041                 spin_unlock(&space_info->lock);
3042                 goto out;
3043         }
3044
3045         if (!force && !should_alloc_chunk(space_info, alloc_bytes)) {
3046                 spin_unlock(&space_info->lock);
3047                 goto out;
3048         }
3049         spin_unlock(&space_info->lock);
3050
3051         /*
3052          * if we're doing a data chunk, go ahead and make sure that
3053          * we keep a reasonable number of metadata chunks allocated in the
3054          * FS as well.
3055          */
3056         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3057                 fs_info->data_chunk_allocations++;
3058                 if (!(fs_info->data_chunk_allocations %
3059                       fs_info->metadata_ratio))
3060                         force_metadata_allocation(fs_info);
3061         }
3062
3063         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3064         spin_lock(&space_info->lock);
3065         if (ret)
3066                 space_info->full = 1;
3067         else
3068                 ret = 1;
3069         space_info->force_alloc = 0;
3070         spin_unlock(&space_info->lock);
3071 out:
3072         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3073         return ret;
3074 }
3075
3076 static int maybe_allocate_chunk(struct btrfs_trans_handle *trans,
3077                                 struct btrfs_root *root,
3078                                 struct btrfs_space_info *sinfo, u64 num_bytes)
3079 {
3080         int ret;
3081         int end_trans = 0;
3082
3083         if (sinfo->full)
3084                 return 0;
3085
3086         spin_lock(&sinfo->lock);
3087         ret = should_alloc_chunk(sinfo, num_bytes + 2 * 1024 * 1024);
3088         spin_unlock(&sinfo->lock);
3089         if (!ret)
3090                 return 0;
3091
3092         if (!trans) {
3093                 trans = btrfs_join_transaction(root, 1);
3094                 BUG_ON(IS_ERR(trans));
3095                 end_trans = 1;
3096         }
3097
3098         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3099                              num_bytes + 2 * 1024 * 1024,
3100                              get_alloc_profile(root, sinfo->flags), 0);
3101
3102         if (end_trans)
3103                 btrfs_end_transaction(trans, root);
3104
3105         return ret == 1 ? 1 : 0;
3106 }
3107
3108 /*
3109  * shrink metadata reservation for delalloc
3110  */
3111 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3112                            struct btrfs_root *root, u64 to_reclaim)
3113 {
3114         struct btrfs_block_rsv *block_rsv;
3115         u64 reserved;
3116         u64 max_reclaim;
3117         u64 reclaimed = 0;
3118         int pause = 1;
3119         int ret;
3120
3121         block_rsv = &root->fs_info->delalloc_block_rsv;
3122         spin_lock(&block_rsv->lock);
3123         reserved = block_rsv->reserved;
3124         spin_unlock(&block_rsv->lock);
3125
3126         if (reserved == 0)
3127                 return 0;
3128
3129         max_reclaim = min(reserved, to_reclaim);
3130
3131         while (1) {
3132                 ret = btrfs_start_one_delalloc_inode(root, trans ? 1 : 0);
3133                 if (!ret) {
3134                         __set_current_state(TASK_INTERRUPTIBLE);
3135                         schedule_timeout(pause);
3136                         pause <<= 1;
3137                         if (pause > HZ / 10)
3138                                 pause = HZ / 10;
3139                 } else {
3140                         pause = 1;
3141                 }
3142
3143                 spin_lock(&block_rsv->lock);
3144                 if (reserved > block_rsv->reserved)
3145                         reclaimed = reserved - block_rsv->reserved;
3146                 reserved = block_rsv->reserved;
3147                 spin_unlock(&block_rsv->lock);
3148
3149                 if (reserved == 0 || reclaimed >= max_reclaim)
3150                         break;
3151
3152                 if (trans && trans->transaction->blocked)
3153                         return -EAGAIN;
3154         }
3155         return reclaimed >= to_reclaim;
3156 }
3157
3158 static int should_retry_reserve(struct btrfs_trans_handle *trans,
3159                                 struct btrfs_root *root,
3160                                 struct btrfs_block_rsv *block_rsv,
3161                                 u64 num_bytes, int *retries)
3162 {
3163         struct btrfs_space_info *space_info = block_rsv->space_info;
3164         int ret;
3165
3166         if ((*retries) > 2)
3167                 return -ENOSPC;
3168
3169         ret = maybe_allocate_chunk(trans, root, space_info, num_bytes);
3170         if (ret)
3171                 return 1;
3172
3173         if (trans && trans->transaction->in_commit)
3174                 return -ENOSPC;
3175
3176         ret = shrink_delalloc(trans, root, num_bytes);
3177         if (ret)
3178                 return ret;
3179
3180         spin_lock(&space_info->lock);
3181         if (space_info->bytes_pinned < num_bytes)
3182                 ret = 1;
3183         spin_unlock(&space_info->lock);
3184         if (ret)
3185                 return -ENOSPC;
3186
3187         (*retries)++;
3188
3189         if (trans)
3190                 return -EAGAIN;
3191
3192         trans = btrfs_join_transaction(root, 1);
3193         BUG_ON(IS_ERR(trans));
3194         ret = btrfs_commit_transaction(trans, root);
3195         BUG_ON(ret);
3196
3197         return 1;
3198 }
3199
3200 static int reserve_metadata_bytes(struct btrfs_block_rsv *block_rsv,
3201                                   u64 num_bytes)
3202 {
3203         struct btrfs_space_info *space_info = block_rsv->space_info;
3204         u64 unused;
3205         int ret = -ENOSPC;
3206
3207         spin_lock(&space_info->lock);
3208         unused = space_info->bytes_used + space_info->bytes_reserved +
3209                  space_info->bytes_pinned + space_info->bytes_readonly;
3210
3211         if (unused < space_info->total_bytes)
3212                 unused = space_info->total_bytes - unused;
3213         else
3214                 unused = 0;
3215
3216         if (unused >= num_bytes) {
3217                 if (block_rsv->priority >= 10) {
3218                         space_info->bytes_reserved += num_bytes;
3219                         ret = 0;
3220                 } else {
3221                         if ((unused + block_rsv->reserved) *
3222                             block_rsv->priority >=
3223                             (num_bytes + block_rsv->reserved) * 10) {
3224                                 space_info->bytes_reserved += num_bytes;
3225                                 ret = 0;
3226                         }
3227                 }
3228         }
3229         spin_unlock(&space_info->lock);
3230
3231         return ret;
3232 }
3233
3234 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3235                                              struct btrfs_root *root)
3236 {
3237         struct btrfs_block_rsv *block_rsv;
3238         if (root->ref_cows)
3239                 block_rsv = trans->block_rsv;
3240         else
3241                 block_rsv = root->block_rsv;
3242
3243         if (!block_rsv)
3244                 block_rsv = &root->fs_info->empty_block_rsv;
3245
3246         return block_rsv;
3247 }
3248
3249 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3250                                u64 num_bytes)
3251 {
3252         int ret = -ENOSPC;
3253         spin_lock(&block_rsv->lock);
3254         if (block_rsv->reserved >= num_bytes) {
3255                 block_rsv->reserved -= num_bytes;
3256                 if (block_rsv->reserved < block_rsv->size)
3257                         block_rsv->full = 0;
3258                 ret = 0;
3259         }
3260         spin_unlock(&block_rsv->lock);
3261         return ret;
3262 }
3263
3264 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3265                                 u64 num_bytes, int update_size)
3266 {
3267         spin_lock(&block_rsv->lock);
3268         block_rsv->reserved += num_bytes;
3269         if (update_size)
3270                 block_rsv->size += num_bytes;
3271         else if (block_rsv->reserved >= block_rsv->size)
3272                 block_rsv->full = 1;
3273         spin_unlock(&block_rsv->lock);
3274 }
3275
3276 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3277                              struct btrfs_block_rsv *dest, u64 num_bytes)
3278 {
3279         struct btrfs_space_info *space_info = block_rsv->space_info;
3280
3281         spin_lock(&block_rsv->lock);
3282         if (num_bytes == (u64)-1)
3283                 num_bytes = block_rsv->size;
3284         block_rsv->size -= num_bytes;
3285         if (block_rsv->reserved >= block_rsv->size) {
3286                 num_bytes = block_rsv->reserved - block_rsv->size;
3287                 block_rsv->reserved = block_rsv->size;
3288                 block_rsv->full = 1;
3289         } else {
3290                 num_bytes = 0;
3291         }
3292         spin_unlock(&block_rsv->lock);
3293
3294         if (num_bytes > 0) {
3295                 if (dest) {
3296                         block_rsv_add_bytes(dest, num_bytes, 0);
3297                 } else {
3298                         spin_lock(&space_info->lock);
3299                         space_info->bytes_reserved -= num_bytes;
3300                         spin_unlock(&space_info->lock);
3301                 }
3302         }
3303 }
3304
3305 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3306                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3307 {
3308         int ret;
3309
3310         ret = block_rsv_use_bytes(src, num_bytes);
3311         if (ret)
3312                 return ret;
3313
3314         block_rsv_add_bytes(dst, num_bytes, 1);
3315         return 0;
3316 }
3317
3318 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3319 {
3320         memset(rsv, 0, sizeof(*rsv));
3321         spin_lock_init(&rsv->lock);
3322         atomic_set(&rsv->usage, 1);
3323         rsv->priority = 6;
3324         INIT_LIST_HEAD(&rsv->list);
3325 }
3326
3327 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3328 {
3329         struct btrfs_block_rsv *block_rsv;
3330         struct btrfs_fs_info *fs_info = root->fs_info;
3331         u64 alloc_target;
3332
3333         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3334         if (!block_rsv)
3335                 return NULL;
3336
3337         btrfs_init_block_rsv(block_rsv);
3338
3339         alloc_target = btrfs_get_alloc_profile(root, 0);
3340         block_rsv->space_info = __find_space_info(fs_info,
3341                                                   BTRFS_BLOCK_GROUP_METADATA);
3342
3343         return block_rsv;
3344 }
3345
3346 void btrfs_free_block_rsv(struct btrfs_root *root,
3347                           struct btrfs_block_rsv *rsv)
3348 {
3349         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3350                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3351                 if (!rsv->durable)
3352                         kfree(rsv);
3353         }
3354 }
3355
3356 /*
3357  * make the block_rsv struct be able to capture freed space.
3358  * the captured space will re-add to the the block_rsv struct
3359  * after transaction commit
3360  */
3361 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3362                                  struct btrfs_block_rsv *block_rsv)
3363 {
3364         block_rsv->durable = 1;
3365         mutex_lock(&fs_info->durable_block_rsv_mutex);
3366         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3367         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3368 }
3369
3370 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3371                         struct btrfs_root *root,
3372                         struct btrfs_block_rsv *block_rsv,
3373                         u64 num_bytes, int *retries)
3374 {
3375         int ret;
3376
3377         if (num_bytes == 0)
3378                 return 0;
3379 again:
3380         ret = reserve_metadata_bytes(block_rsv, num_bytes);
3381         if (!ret) {
3382                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3383                 return 0;
3384         }
3385
3386         ret = should_retry_reserve(trans, root, block_rsv, num_bytes, retries);
3387         if (ret > 0)
3388                 goto again;
3389
3390         return ret;
3391 }
3392
3393 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3394                           struct btrfs_root *root,
3395                           struct btrfs_block_rsv *block_rsv,
3396                           u64 min_reserved, int min_factor)
3397 {
3398         u64 num_bytes = 0;
3399         int commit_trans = 0;
3400         int ret = -ENOSPC;
3401
3402         if (!block_rsv)
3403                 return 0;
3404
3405         spin_lock(&block_rsv->lock);
3406         if (min_factor > 0)
3407                 num_bytes = div_factor(block_rsv->size, min_factor);
3408         if (min_reserved > num_bytes)
3409                 num_bytes = min_reserved;
3410
3411         if (block_rsv->reserved >= num_bytes) {
3412                 ret = 0;
3413         } else {
3414                 num_bytes -= block_rsv->reserved;
3415                 if (block_rsv->durable &&
3416                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3417                         commit_trans = 1;
3418         }
3419         spin_unlock(&block_rsv->lock);
3420         if (!ret)
3421                 return 0;
3422
3423         if (block_rsv->refill_used) {
3424                 ret = reserve_metadata_bytes(block_rsv, num_bytes);
3425                 if (!ret) {
3426                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3427                         return 0;
3428                 }
3429         }
3430
3431         if (commit_trans) {
3432                 if (trans)
3433                         return -EAGAIN;
3434
3435                 trans = btrfs_join_transaction(root, 1);
3436                 BUG_ON(IS_ERR(trans));
3437                 ret = btrfs_commit_transaction(trans, root);
3438                 return 0;
3439         }
3440
3441         WARN_ON(1);
3442         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
3443                 block_rsv->size, block_rsv->reserved,
3444                 block_rsv->freed[0], block_rsv->freed[1]);
3445
3446         return -ENOSPC;
3447 }
3448
3449 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3450                             struct btrfs_block_rsv *dst_rsv,
3451                             u64 num_bytes)
3452 {
3453         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3454 }
3455
3456 void btrfs_block_rsv_release(struct btrfs_root *root,
3457                              struct btrfs_block_rsv *block_rsv,
3458                              u64 num_bytes)
3459 {
3460         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3461         if (global_rsv->full || global_rsv == block_rsv ||
3462             block_rsv->space_info != global_rsv->space_info)
3463                 global_rsv = NULL;
3464         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3465 }
3466
3467 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3468 {
3469         struct btrfs_space_info *space_info;
3470
3471         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3472         fs_info->chunk_block_rsv.space_info = space_info;
3473         fs_info->chunk_block_rsv.priority = 10;
3474
3475         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3476         fs_info->trans_block_rsv.space_info = space_info;
3477         fs_info->empty_block_rsv.space_info = space_info;
3478         fs_info->empty_block_rsv.priority = 10;
3479
3480         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3481 }
3482
3483 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3484 {
3485         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3486                 3 * num_items;
3487 }
3488
3489 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3490                                  struct btrfs_root *root,
3491                                  int num_items, int *retries)
3492 {
3493         u64 num_bytes;
3494         int ret;
3495
3496         if (num_items == 0 || root->fs_info->chunk_root == root)
3497                 return 0;
3498
3499         num_bytes = calc_trans_metadata_size(root, num_items);
3500         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3501                                   num_bytes, retries);
3502         if (!ret) {
3503                 trans->bytes_reserved += num_bytes;
3504                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3505         }
3506         return ret;
3507 }
3508
3509 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3510                                   struct btrfs_root *root)
3511 {
3512         if (!trans->bytes_reserved)
3513                 return;
3514
3515         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3516         btrfs_block_rsv_release(root, trans->block_rsv,
3517                                 trans->bytes_reserved);
3518         trans->bytes_reserved = 0;
3519 }
3520
3521 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3522                                 struct btrfs_pending_snapshot *pending)
3523 {
3524         struct btrfs_root *root = pending->root;
3525         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3526         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3527         /*
3528          * two for root back/forward refs, two for directory entries
3529          * and one for root of the snapshot.
3530          */
3531         u64 num_bytes = calc_trans_metadata_size(root, 5);
3532         dst_rsv->space_info = src_rsv->space_info;
3533         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3534 }
3535
3536 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3537 {
3538         return num_bytes >>= 3;
3539 }
3540
3541 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3542 {
3543         struct btrfs_root *root = BTRFS_I(inode)->root;
3544         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3545         u64 to_reserve;
3546         int nr_extents;
3547         int retries = 0;
3548         int ret;
3549
3550         if (btrfs_transaction_in_commit(root->fs_info))
3551                 schedule_timeout(1);
3552
3553         num_bytes = ALIGN(num_bytes, root->sectorsize);
3554 again:
3555         spin_lock(&BTRFS_I(inode)->accounting_lock);
3556         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
3557         if (nr_extents > BTRFS_I(inode)->reserved_extents) {
3558                 nr_extents -= BTRFS_I(inode)->reserved_extents;
3559                 to_reserve = calc_trans_metadata_size(root, nr_extents);
3560         } else {
3561                 nr_extents = 0;
3562                 to_reserve = 0;
3563         }
3564
3565         to_reserve += calc_csum_metadata_size(inode, num_bytes);
3566         ret = reserve_metadata_bytes(block_rsv, to_reserve);
3567         if (ret) {
3568                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
3569                 ret = should_retry_reserve(NULL, root, block_rsv, to_reserve,
3570                                            &retries);
3571                 if (ret > 0)
3572                         goto again;
3573                 return ret;
3574         }
3575
3576         BTRFS_I(inode)->reserved_extents += nr_extents;
3577         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
3578         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3579
3580         block_rsv_add_bytes(block_rsv, to_reserve, 1);
3581
3582         if (block_rsv->size > 512 * 1024 * 1024)
3583                 shrink_delalloc(NULL, root, to_reserve);
3584
3585         return 0;
3586 }
3587
3588 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
3589 {
3590         struct btrfs_root *root = BTRFS_I(inode)->root;
3591         u64 to_free;
3592         int nr_extents;
3593
3594         num_bytes = ALIGN(num_bytes, root->sectorsize);
3595         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
3596
3597         spin_lock(&BTRFS_I(inode)->accounting_lock);
3598         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
3599         if (nr_extents < BTRFS_I(inode)->reserved_extents) {
3600                 nr_extents = BTRFS_I(inode)->reserved_extents - nr_extents;
3601                 BTRFS_I(inode)->reserved_extents -= nr_extents;
3602         } else {
3603                 nr_extents = 0;
3604         }
3605         spin_unlock(&BTRFS_I(inode)->accounting_lock);
3606
3607         to_free = calc_csum_metadata_size(inode, num_bytes);
3608         if (nr_extents > 0)
3609                 to_free += calc_trans_metadata_size(root, nr_extents);
3610
3611         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
3612                                 to_free);
3613 }
3614
3615 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
3616 {
3617         int ret;
3618
3619         ret = btrfs_check_data_free_space(inode, num_bytes);
3620         if (ret)
3621                 return ret;
3622
3623         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
3624         if (ret) {
3625                 btrfs_free_reserved_data_space(inode, num_bytes);
3626                 return ret;
3627         }
3628
3629         return 0;
3630 }
3631
3632 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
3633 {
3634         btrfs_delalloc_release_metadata(inode, num_bytes);
3635         btrfs_free_reserved_data_space(inode, num_bytes);
3636 }
3637
3638 static int update_block_group(struct btrfs_trans_handle *trans,
3639                               struct btrfs_root *root,
3640                               u64 bytenr, u64 num_bytes, int alloc)
3641 {
3642         struct btrfs_block_group_cache *cache;
3643         struct btrfs_fs_info *info = root->fs_info;
3644         int factor;
3645         u64 total = num_bytes;
3646         u64 old_val;
3647         u64 byte_in_group;
3648
3649         /* block accounting for super block */
3650         spin_lock(&info->delalloc_lock);
3651         old_val = btrfs_super_bytes_used(&info->super_copy);
3652         if (alloc)
3653                 old_val += num_bytes;
3654         else
3655                 old_val -= num_bytes;
3656         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3657         spin_unlock(&info->delalloc_lock);
3658
3659         while (total) {
3660                 cache = btrfs_lookup_block_group(info, bytenr);
3661                 if (!cache)
3662                         return -1;
3663                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
3664                                     BTRFS_BLOCK_GROUP_RAID1 |
3665                                     BTRFS_BLOCK_GROUP_RAID10))
3666                         factor = 2;
3667                 else
3668                         factor = 1;
3669                 byte_in_group = bytenr - cache->key.objectid;
3670                 WARN_ON(byte_in_group > cache->key.offset);
3671
3672                 spin_lock(&cache->space_info->lock);
3673                 spin_lock(&cache->lock);
3674                 cache->dirty = 1;
3675                 old_val = btrfs_block_group_used(&cache->item);
3676                 num_bytes = min(total, cache->key.offset - byte_in_group);
3677                 if (alloc) {
3678                         old_val += num_bytes;
3679                         btrfs_set_block_group_used(&cache->item, old_val);
3680                         cache->reserved -= num_bytes;
3681                         cache->space_info->bytes_reserved -= num_bytes;
3682                         cache->space_info->bytes_used += num_bytes;
3683                         cache->space_info->disk_used += num_bytes * factor;
3684                         spin_unlock(&cache->lock);
3685                         spin_unlock(&cache->space_info->lock);
3686                 } else {
3687                         old_val -= num_bytes;
3688                         btrfs_set_block_group_used(&cache->item, old_val);
3689                         cache->pinned += num_bytes;
3690                         cache->space_info->bytes_pinned += num_bytes;
3691                         cache->space_info->bytes_used -= num_bytes;
3692                         cache->space_info->disk_used -= num_bytes * factor;
3693                         spin_unlock(&cache->lock);
3694                         spin_unlock(&cache->space_info->lock);
3695
3696                         set_extent_dirty(info->pinned_extents,
3697                                          bytenr, bytenr + num_bytes - 1,
3698                                          GFP_NOFS | __GFP_NOFAIL);
3699                 }
3700                 btrfs_put_block_group(cache);
3701                 total -= num_bytes;
3702                 bytenr += num_bytes;
3703         }
3704         return 0;
3705 }
3706
3707 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3708 {
3709         struct btrfs_block_group_cache *cache;
3710         u64 bytenr;
3711
3712         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3713         if (!cache)
3714                 return 0;
3715
3716         bytenr = cache->key.objectid;
3717         btrfs_put_block_group(cache);
3718
3719         return bytenr;
3720 }
3721
3722 static int pin_down_extent(struct btrfs_root *root,
3723                            struct btrfs_block_group_cache *cache,
3724                            u64 bytenr, u64 num_bytes, int reserved)
3725 {
3726         spin_lock(&cache->space_info->lock);
3727         spin_lock(&cache->lock);
3728         cache->pinned += num_bytes;
3729         cache->space_info->bytes_pinned += num_bytes;
3730         if (reserved) {
3731                 cache->reserved -= num_bytes;
3732                 cache->space_info->bytes_reserved -= num_bytes;
3733         }
3734         spin_unlock(&cache->lock);
3735         spin_unlock(&cache->space_info->lock);
3736
3737         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
3738                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
3739         return 0;
3740 }
3741
3742 /*
3743  * this function must be called within transaction
3744  */
3745 int btrfs_pin_extent(struct btrfs_root *root,
3746                      u64 bytenr, u64 num_bytes, int reserved)
3747 {
3748         struct btrfs_block_group_cache *cache;
3749
3750         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
3751         BUG_ON(!cache);
3752
3753         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
3754
3755         btrfs_put_block_group(cache);
3756         return 0;
3757 }
3758
3759 /*
3760  * update size of reserved extents. this function may return -EAGAIN
3761  * if 'reserve' is true or 'sinfo' is false.
3762  */
3763 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
3764                                  u64 num_bytes, int reserve, int sinfo)
3765 {
3766         int ret = 0;
3767         if (sinfo) {
3768                 struct btrfs_space_info *space_info = cache->space_info;
3769                 spin_lock(&space_info->lock);
3770                 spin_lock(&cache->lock);
3771                 if (reserve) {
3772                         if (cache->ro) {
3773                                 ret = -EAGAIN;
3774                         } else {
3775                                 cache->reserved += num_bytes;
3776                                 space_info->bytes_reserved += num_bytes;
3777                         }
3778                 } else {
3779                         if (cache->ro)
3780                                 space_info->bytes_readonly += num_bytes;
3781                         cache->reserved -= num_bytes;
3782                         space_info->bytes_reserved -= num_bytes;
3783                 }
3784                 spin_unlock(&cache->lock);
3785                 spin_unlock(&space_info->lock);
3786         } else {
3787                 spin_lock(&cache->lock);
3788                 if (cache->ro) {
3789                         ret = -EAGAIN;
3790                 } else {
3791                         if (reserve)
3792                                 cache->reserved += num_bytes;
3793                         else
3794                                 cache->reserved -= num_bytes;
3795                 }
3796                 spin_unlock(&cache->lock);
3797         }
3798         return ret;
3799 }
3800
3801 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3802                                 struct btrfs_root *root)
3803 {
3804         struct btrfs_fs_info *fs_info = root->fs_info;
3805         struct btrfs_caching_control *next;
3806         struct btrfs_caching_control *caching_ctl;
3807         struct btrfs_block_group_cache *cache;
3808
3809         down_write(&fs_info->extent_commit_sem);
3810
3811         list_for_each_entry_safe(caching_ctl, next,
3812                                  &fs_info->caching_block_groups, list) {
3813                 cache = caching_ctl->block_group;
3814                 if (block_group_cache_done(cache)) {
3815                         cache->last_byte_to_unpin = (u64)-1;
3816                         list_del_init(&caching_ctl->list);
3817                         put_caching_control(caching_ctl);
3818                 } else {
3819                         cache->last_byte_to_unpin = caching_ctl->progress;
3820                 }
3821         }
3822
3823         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3824                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3825         else
3826                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3827
3828         up_write(&fs_info->extent_commit_sem);
3829         return 0;
3830 }
3831
3832 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3833 {
3834         struct btrfs_fs_info *fs_info = root->fs_info;
3835         struct btrfs_block_group_cache *cache = NULL;
3836         u64 len;
3837
3838         while (start <= end) {
3839                 if (!cache ||
3840                     start >= cache->key.objectid + cache->key.offset) {
3841                         if (cache)
3842                                 btrfs_put_block_group(cache);
3843                         cache = btrfs_lookup_block_group(fs_info, start);
3844                         BUG_ON(!cache);
3845                 }
3846
3847                 len = cache->key.objectid + cache->key.offset - start;
3848                 len = min(len, end + 1 - start);
3849
3850                 if (start < cache->last_byte_to_unpin) {
3851                         len = min(len, cache->last_byte_to_unpin - start);
3852                         btrfs_add_free_space(cache, start, len);
3853                 }
3854
3855                 start += len;
3856
3857                 spin_lock(&cache->space_info->lock);
3858                 spin_lock(&cache->lock);
3859                 cache->pinned -= len;
3860                 cache->space_info->bytes_pinned -= len;
3861                 if (cache->ro) {
3862                         cache->space_info->bytes_readonly += len;
3863                 } else if (cache->reserved_pinned > 0) {
3864                         len = min(len, cache->reserved_pinned);
3865                         cache->reserved_pinned -= len;
3866                         cache->space_info->bytes_reserved += len;
3867                 }
3868                 spin_unlock(&cache->lock);
3869                 spin_unlock(&cache->space_info->lock);
3870         }
3871
3872         if (cache)
3873                 btrfs_put_block_group(cache);
3874         return 0;
3875 }
3876
3877 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3878                                struct btrfs_root *root)
3879 {
3880         struct btrfs_fs_info *fs_info = root->fs_info;
3881         struct extent_io_tree *unpin;
3882         struct btrfs_block_rsv *block_rsv;
3883         struct btrfs_block_rsv *next_rsv;
3884         u64 start;
3885         u64 end;
3886         int idx;
3887         int ret;
3888
3889         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3890                 unpin = &fs_info->freed_extents[1];
3891         else
3892                 unpin = &fs_info->freed_extents[0];
3893
3894         while (1) {
3895                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3896                                             EXTENT_DIRTY);
3897                 if (ret)
3898                         break;
3899
3900                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3901
3902                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3903                 unpin_extent_range(root, start, end);
3904                 cond_resched();
3905         }
3906
3907         mutex_lock(&fs_info->durable_block_rsv_mutex);
3908         list_for_each_entry_safe(block_rsv, next_rsv,
3909                                  &fs_info->durable_block_rsv_list, list) {
3910
3911                 idx = trans->transid & 0x1;
3912                 if (block_rsv->freed[idx] > 0) {
3913                         block_rsv_add_bytes(block_rsv,
3914                                             block_rsv->freed[idx], 0);
3915                         block_rsv->freed[idx] = 0;
3916                 }
3917                 if (atomic_read(&block_rsv->usage) == 0) {
3918                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
3919
3920                         if (block_rsv->freed[0] == 0 &&
3921                             block_rsv->freed[1] == 0) {
3922                                 list_del_init(&block_rsv->list);
3923                                 kfree(block_rsv);
3924                         }
3925                 } else {
3926                         btrfs_block_rsv_release(root, block_rsv, 0);
3927                 }
3928         }
3929         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3930
3931         return 0;
3932 }
3933
3934 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3935                                 struct btrfs_root *root,
3936                                 u64 bytenr, u64 num_bytes, u64 parent,
3937                                 u64 root_objectid, u64 owner_objectid,
3938                                 u64 owner_offset, int refs_to_drop,
3939                                 struct btrfs_delayed_extent_op *extent_op)
3940 {
3941         struct btrfs_key key;
3942         struct btrfs_path *path;
3943         struct btrfs_fs_info *info = root->fs_info;
3944         struct btrfs_root *extent_root = info->extent_root;
3945         struct extent_buffer *leaf;
3946         struct btrfs_extent_item *ei;
3947         struct btrfs_extent_inline_ref *iref;
3948         int ret;
3949         int is_data;
3950         int extent_slot = 0;
3951         int found_extent = 0;
3952         int num_to_del = 1;
3953         u32 item_size;
3954         u64 refs;
3955
3956         path = btrfs_alloc_path();
3957         if (!path)
3958                 return -ENOMEM;
3959
3960         path->reada = 1;
3961         path->leave_spinning = 1;
3962
3963         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3964         BUG_ON(!is_data && refs_to_drop != 1);
3965
3966         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3967                                     bytenr, num_bytes, parent,
3968                                     root_objectid, owner_objectid,
3969                                     owner_offset);
3970         if (ret == 0) {
3971                 extent_slot = path->slots[0];
3972                 while (extent_slot >= 0) {
3973                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3974                                               extent_slot);
3975                         if (key.objectid != bytenr)
3976                                 break;
3977                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3978                             key.offset == num_bytes) {
3979                                 found_extent = 1;
3980                                 break;
3981                         }
3982                         if (path->slots[0] - extent_slot > 5)
3983                                 break;
3984                         extent_slot--;
3985                 }
3986 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3987                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3988                 if (found_extent && item_size < sizeof(*ei))
3989                         found_extent = 0;
3990 #endif
3991                 if (!found_extent) {
3992                         BUG_ON(iref);
3993                         ret = remove_extent_backref(trans, extent_root, path,
3994                                                     NULL, refs_to_drop,
3995                                                     is_data);
3996                         BUG_ON(ret);
3997                         btrfs_release_path(extent_root, path);
3998                         path->leave_spinning = 1;
3999
4000                         key.objectid = bytenr;
4001                         key.type = BTRFS_EXTENT_ITEM_KEY;
4002                         key.offset = num_bytes;
4003
4004                         ret = btrfs_search_slot(trans, extent_root,
4005                                                 &key, path, -1, 1);
4006                         if (ret) {
4007                                 printk(KERN_ERR "umm, got %d back from search"
4008                                        ", was looking for %llu\n", ret,
4009                                        (unsigned long long)bytenr);
4010                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4011                         }
4012                         BUG_ON(ret);
4013                         extent_slot = path->slots[0];
4014                 }
4015         } else {
4016                 btrfs_print_leaf(extent_root, path->nodes[0]);
4017                 WARN_ON(1);
4018                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4019                        "parent %llu root %llu  owner %llu offset %llu\n",
4020                        (unsigned long long)bytenr,
4021                        (unsigned long long)parent,
4022                        (unsigned long long)root_objectid,
4023                        (unsigned long long)owner_objectid,
4024                        (unsigned long long)owner_offset);
4025         }
4026
4027         leaf = path->nodes[0];
4028         item_size = btrfs_item_size_nr(leaf, extent_slot);
4029 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4030         if (item_size < sizeof(*ei)) {
4031                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4032                 ret = convert_extent_item_v0(trans, extent_root, path,
4033                                              owner_objectid, 0);
4034                 BUG_ON(ret < 0);
4035
4036                 btrfs_release_path(extent_root, path);
4037                 path->leave_spinning = 1;
4038
4039                 key.objectid = bytenr;
4040                 key.type = BTRFS_EXTENT_ITEM_KEY;
4041                 key.offset = num_bytes;
4042
4043                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4044                                         -1, 1);
4045                 if (ret) {
4046                         printk(KERN_ERR "umm, got %d back from search"
4047                                ", was looking for %llu\n", ret,
4048                                (unsigned long long)bytenr);
4049                         btrfs_print_leaf(extent_root, path->nodes[0]);
4050                 }
4051                 BUG_ON(ret);
4052                 extent_slot = path->slots[0];
4053                 leaf = path->nodes[0];
4054                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4055         }
4056 #endif
4057         BUG_ON(item_size < sizeof(*ei));
4058         ei = btrfs_item_ptr(leaf, extent_slot,
4059                             struct btrfs_extent_item);
4060         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4061                 struct btrfs_tree_block_info *bi;
4062                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4063                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4064                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4065         }
4066
4067         refs = btrfs_extent_refs(leaf, ei);
4068         BUG_ON(refs < refs_to_drop);
4069         refs -= refs_to_drop;
4070
4071         if (refs > 0) {
4072                 if (extent_op)
4073                         __run_delayed_extent_op(extent_op, leaf, ei);
4074                 /*
4075                  * In the case of inline back ref, reference count will
4076                  * be updated by remove_extent_backref
4077                  */
4078                 if (iref) {
4079                         BUG_ON(!found_extent);
4080                 } else {
4081                         btrfs_set_extent_refs(leaf, ei, refs);
4082                         btrfs_mark_buffer_dirty(leaf);
4083                 }
4084                 if (found_extent) {
4085                         ret = remove_extent_backref(trans, extent_root, path,
4086                                                     iref, refs_to_drop,
4087                                                     is_data);
4088                         BUG_ON(ret);
4089                 }
4090         } else {
4091                 if (found_extent) {
4092                         BUG_ON(is_data && refs_to_drop !=
4093                                extent_data_ref_count(root, path, iref));
4094                         if (iref) {
4095                                 BUG_ON(path->slots[0] != extent_slot);
4096                         } else {
4097                                 BUG_ON(path->slots[0] != extent_slot + 1);
4098                                 path->slots[0] = extent_slot;
4099                                 num_to_del = 2;
4100                         }
4101                 }
4102
4103                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4104                                       num_to_del);
4105                 BUG_ON(ret);
4106                 btrfs_release_path(extent_root, path);
4107
4108                 if (is_data) {
4109                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4110                         BUG_ON(ret);
4111                 } else {
4112                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4113                              bytenr >> PAGE_CACHE_SHIFT,
4114                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4115                 }
4116
4117                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4118                 BUG_ON(ret);
4119         }
4120         btrfs_free_path(path);
4121         return ret;
4122 }
4123
4124 /*
4125  * when we free an block, it is possible (and likely) that we free the last
4126  * delayed ref for that extent as well.  This searches the delayed ref tree for
4127  * a given extent, and if there are no other delayed refs to be processed, it
4128  * removes it from the tree.
4129  */
4130 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4131                                       struct btrfs_root *root, u64 bytenr)
4132 {
4133         struct btrfs_delayed_ref_head *head;
4134         struct btrfs_delayed_ref_root *delayed_refs;
4135         struct btrfs_delayed_ref_node *ref;
4136         struct rb_node *node;
4137         int ret = 0;
4138
4139         delayed_refs = &trans->transaction->delayed_refs;
4140         spin_lock(&delayed_refs->lock);
4141         head = btrfs_find_delayed_ref_head(trans, bytenr);
4142         if (!head)
4143                 goto out;
4144
4145         node = rb_prev(&head->node.rb_node);
4146         if (!node)
4147                 goto out;
4148
4149         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4150
4151         /* there are still entries for this ref, we can't drop it */
4152         if (ref->bytenr == bytenr)
4153                 goto out;
4154
4155         if (head->extent_op) {
4156                 if (!head->must_insert_reserved)
4157                         goto out;
4158                 kfree(head->extent_op);
4159                 head->extent_op = NULL;
4160         }
4161
4162         /*
4163          * waiting for the lock here would deadlock.  If someone else has it
4164          * locked they are already in the process of dropping it anyway
4165          */
4166         if (!mutex_trylock(&head->mutex))
4167                 goto out;
4168
4169         /*
4170          * at this point we have a head with no other entries.  Go
4171          * ahead and process it.
4172          */
4173         head->node.in_tree = 0;
4174         rb_erase(&head->node.rb_node, &delayed_refs->root);
4175
4176         delayed_refs->num_entries--;
4177
4178         /*
4179          * we don't take a ref on the node because we're removing it from the
4180          * tree, so we just steal the ref the tree was holding.
4181          */
4182         delayed_refs->num_heads--;
4183         if (list_empty(&head->cluster))
4184                 delayed_refs->num_heads_ready--;
4185
4186         list_del_init(&head->cluster);
4187         spin_unlock(&delayed_refs->lock);
4188
4189         BUG_ON(head->extent_op);
4190         if (head->must_insert_reserved)
4191                 ret = 1;
4192
4193         mutex_unlock(&head->mutex);
4194         btrfs_put_delayed_ref(&head->node);
4195         return ret;
4196 out:
4197         spin_unlock(&delayed_refs->lock);
4198         return 0;
4199 }
4200
4201 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4202                            struct btrfs_root *root,
4203                            struct extent_buffer *buf,
4204                            u64 parent, int last_ref)
4205 {
4206         struct btrfs_block_rsv *block_rsv;
4207         struct btrfs_block_group_cache *cache = NULL;
4208         int ret;
4209
4210         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4211                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4212                                                 parent, root->root_key.objectid,
4213                                                 btrfs_header_level(buf),
4214                                                 BTRFS_DROP_DELAYED_REF, NULL);
4215                 BUG_ON(ret);
4216         }
4217
4218         if (!last_ref)
4219                 return;
4220
4221         block_rsv = get_block_rsv(trans, root);
4222         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4223         BUG_ON(block_rsv->space_info != cache->space_info);
4224
4225         if (btrfs_header_generation(buf) == trans->transid) {
4226                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4227                         ret = check_ref_cleanup(trans, root, buf->start);
4228                         if (!ret)
4229                                 goto pin;
4230                 }
4231
4232                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4233                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4234                         goto pin;
4235                 }
4236
4237                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4238
4239                 btrfs_add_free_space(cache, buf->start, buf->len);
4240                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4241                 if (ret == -EAGAIN) {
4242                         /* block group became read-only */
4243                         update_reserved_bytes(cache, buf->len, 0, 1);
4244                         goto out;
4245                 }
4246
4247                 ret = 1;
4248                 spin_lock(&block_rsv->lock);
4249                 if (block_rsv->reserved < block_rsv->size) {
4250                         block_rsv->reserved += buf->len;
4251                         ret = 0;
4252                 }
4253                 spin_unlock(&block_rsv->lock);
4254
4255                 if (ret) {
4256                         spin_lock(&cache->space_info->lock);
4257                         cache->space_info->bytes_reserved -= buf->len;
4258                         spin_unlock(&cache->space_info->lock);
4259                 }
4260                 goto out;
4261         }
4262 pin:
4263         if (block_rsv->durable && !cache->ro) {
4264                 ret = 0;
4265                 spin_lock(&cache->lock);
4266                 if (!cache->ro) {
4267                         cache->reserved_pinned += buf->len;
4268                         ret = 1;
4269                 }
4270                 spin_unlock(&cache->lock);
4271
4272                 if (ret) {
4273                         spin_lock(&block_rsv->lock);
4274                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4275                         spin_unlock(&block_rsv->lock);
4276                 }
4277         }
4278 out:
4279         btrfs_put_block_group(cache);
4280 }
4281
4282 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4283                       struct btrfs_root *root,
4284                       u64 bytenr, u64 num_bytes, u64 parent,
4285                       u64 root_objectid, u64 owner, u64 offset)
4286 {
4287         int ret;
4288
4289         /*
4290          * tree log blocks never actually go into the extent allocation
4291          * tree, just update pinning info and exit early.
4292          */
4293         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4294                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4295                 /* unlocks the pinned mutex */
4296                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4297                 ret = 0;
4298         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4299                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4300                                         parent, root_objectid, (int)owner,
4301                                         BTRFS_DROP_DELAYED_REF, NULL);
4302                 BUG_ON(ret);
4303         } else {
4304                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4305                                         parent, root_objectid, owner,
4306                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4307                 BUG_ON(ret);
4308         }
4309         return ret;
4310 }
4311
4312 static u64 stripe_align(struct btrfs_root *root, u64 val)
4313 {
4314         u64 mask = ((u64)root->stripesize - 1);
4315         u64 ret = (val + mask) & ~mask;
4316         return ret;
4317 }
4318
4319 /*
4320  * when we wait for progress in the block group caching, its because
4321  * our allocation attempt failed at least once.  So, we must sleep
4322  * and let some progress happen before we try again.
4323  *
4324  * This function will sleep at least once waiting for new free space to
4325  * show up, and then it will check the block group free space numbers
4326  * for our min num_bytes.  Another option is to have it go ahead
4327  * and look in the rbtree for a free extent of a given size, but this
4328  * is a good start.
4329  */
4330 static noinline int
4331 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4332                                 u64 num_bytes)
4333 {
4334         struct btrfs_caching_control *caching_ctl;
4335         DEFINE_WAIT(wait);
4336
4337         caching_ctl = get_caching_control(cache);
4338         if (!caching_ctl)
4339                 return 0;
4340
4341         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4342                    (cache->free_space >= num_bytes));
4343
4344         put_caching_control(caching_ctl);
4345         return 0;
4346 }
4347
4348 static noinline int
4349 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4350 {
4351         struct btrfs_caching_control *caching_ctl;
4352         DEFINE_WAIT(wait);
4353
4354         caching_ctl = get_caching_control(cache);
4355         if (!caching_ctl)
4356                 return 0;
4357
4358         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4359
4360         put_caching_control(caching_ctl);
4361         return 0;
4362 }
4363
4364 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4365 {
4366         int index;
4367         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4368                 index = 0;
4369         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4370                 index = 1;
4371         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4372                 index = 2;
4373         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4374                 index = 3;
4375         else
4376                 index = 4;
4377         return index;
4378 }
4379
4380 enum btrfs_loop_type {
4381         LOOP_FIND_IDEAL = 0,
4382         LOOP_CACHING_NOWAIT = 1,
4383         LOOP_CACHING_WAIT = 2,
4384         LOOP_ALLOC_CHUNK = 3,
4385         LOOP_NO_EMPTY_SIZE = 4,
4386 };
4387
4388 /*
4389  * walks the btree of allocated extents and find a hole of a given size.
4390  * The key ins is changed to record the hole:
4391  * ins->objectid == block start
4392  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4393  * ins->offset == number of blocks
4394  * Any available blocks before search_start are skipped.
4395  */
4396 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4397                                      struct btrfs_root *orig_root,
4398                                      u64 num_bytes, u64 empty_size,
4399                                      u64 search_start, u64 search_end,
4400                                      u64 hint_byte, struct btrfs_key *ins,
4401                                      int data)
4402 {
4403         int ret = 0;
4404         struct btrfs_root *root = orig_root->fs_info->extent_root;
4405         struct btrfs_free_cluster *last_ptr = NULL;
4406         struct btrfs_block_group_cache *block_group = NULL;
4407         int empty_cluster = 2 * 1024 * 1024;
4408         int allowed_chunk_alloc = 0;
4409         int done_chunk_alloc = 0;
4410         struct btrfs_space_info *space_info;
4411         int last_ptr_loop = 0;
4412         int loop = 0;
4413         int index = 0;
4414         bool found_uncached_bg = false;
4415         bool failed_cluster_refill = false;
4416         bool failed_alloc = false;
4417         u64 ideal_cache_percent = 0;
4418         u64 ideal_cache_offset = 0;
4419
4420         WARN_ON(num_bytes < root->sectorsize);
4421         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4422         ins->objectid = 0;
4423         ins->offset = 0;
4424
4425         space_info = __find_space_info(root->fs_info, data);
4426         if (!space_info) {
4427                 printk(KERN_ERR "No space info for %d\n", data);
4428                 return -ENOSPC;
4429         }
4430
4431         if (orig_root->ref_cows || empty_size)
4432                 allowed_chunk_alloc = 1;
4433
4434         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4435                 last_ptr = &root->fs_info->meta_alloc_cluster;
4436                 if (!btrfs_test_opt(root, SSD))
4437                         empty_cluster = 64 * 1024;
4438         }
4439
4440         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4441                 last_ptr = &root->fs_info->data_alloc_cluster;
4442         }
4443
4444         if (last_ptr) {
4445                 spin_lock(&last_ptr->lock);
4446                 if (last_ptr->block_group)
4447                         hint_byte = last_ptr->window_start;
4448                 spin_unlock(&last_ptr->lock);
4449         }
4450
4451         search_start = max(search_start, first_logical_byte(root, 0));
4452         search_start = max(search_start, hint_byte);
4453
4454         if (!last_ptr)
4455                 empty_cluster = 0;
4456
4457         if (search_start == hint_byte) {
4458 ideal_cache:
4459                 block_group = btrfs_lookup_block_group(root->fs_info,
4460                                                        search_start);
4461                 /*
4462                  * we don't want to use the block group if it doesn't match our
4463                  * allocation bits, or if its not cached.
4464                  *
4465                  * However if we are re-searching with an ideal block group
4466                  * picked out then we don't care that the block group is cached.
4467                  */
4468                 if (block_group && block_group_bits(block_group, data) &&
4469                     (block_group->cached != BTRFS_CACHE_NO ||
4470                      search_start == ideal_cache_offset)) {
4471                         down_read(&space_info->groups_sem);
4472                         if (list_empty(&block_group->list) ||
4473                             block_group->ro) {
4474                                 /*
4475                                  * someone is removing this block group,
4476                                  * we can't jump into the have_block_group
4477                                  * target because our list pointers are not
4478                                  * valid
4479                                  */
4480                                 btrfs_put_block_group(block_group);
4481                                 up_read(&space_info->groups_sem);
4482                         } else {
4483                                 index = get_block_group_index(block_group);
4484                                 goto have_block_group;
4485                         }
4486                 } else if (block_group) {
4487                         btrfs_put_block_group(block_group);
4488                 }
4489         }
4490 search:
4491         down_read(&space_info->groups_sem);
4492         list_for_each_entry(block_group, &space_info->block_groups[index],
4493                             list) {
4494                 u64 offset;
4495                 int cached;
4496
4497                 btrfs_get_block_group(block_group);
4498                 search_start = block_group->key.objectid;
4499
4500 have_block_group:
4501                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4502                         u64 free_percent;
4503
4504                         free_percent = btrfs_block_group_used(&block_group->item);
4505                         free_percent *= 100;
4506                         free_percent = div64_u64(free_percent,
4507                                                  block_group->key.offset);
4508                         free_percent = 100 - free_percent;
4509                         if (free_percent > ideal_cache_percent &&
4510                             likely(!block_group->ro)) {
4511                                 ideal_cache_offset = block_group->key.objectid;
4512                                 ideal_cache_percent = free_percent;
4513                         }
4514
4515                         /*
4516                          * We only want to start kthread caching if we are at
4517                          * the point where we will wait for caching to make
4518                          * progress, or if our ideal search is over and we've
4519                          * found somebody to start caching.
4520                          */
4521                         if (loop > LOOP_CACHING_NOWAIT ||
4522                             (loop > LOOP_FIND_IDEAL &&
4523                              atomic_read(&space_info->caching_threads) < 2)) {
4524                                 ret = cache_block_group(block_group);
4525                                 BUG_ON(ret);
4526                         }
4527                         found_uncached_bg = true;
4528
4529                         /*
4530                          * If loop is set for cached only, try the next block
4531                          * group.
4532                          */
4533                         if (loop == LOOP_FIND_IDEAL)
4534                                 goto loop;
4535                 }
4536
4537                 cached = block_group_cache_done(block_group);
4538                 if (unlikely(!cached))
4539                         found_uncached_bg = true;
4540
4541                 if (unlikely(block_group->ro))
4542                         goto loop;
4543
4544                 /*
4545                  * Ok we want to try and use the cluster allocator, so lets look
4546                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4547                  * have tried the cluster allocator plenty of times at this
4548                  * point and not have found anything, so we are likely way too
4549                  * fragmented for the clustering stuff to find anything, so lets
4550                  * just skip it and let the allocator find whatever block it can
4551                  * find
4552                  */
4553                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4554                         /*
4555                          * the refill lock keeps out other
4556                          * people trying to start a new cluster
4557                          */
4558                         spin_lock(&last_ptr->refill_lock);
4559                         if (last_ptr->block_group &&
4560                             (last_ptr->block_group->ro ||
4561                             !block_group_bits(last_ptr->block_group, data))) {
4562                                 offset = 0;
4563                                 goto refill_cluster;
4564                         }
4565
4566                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4567                                                  num_bytes, search_start);
4568                         if (offset) {
4569                                 /* we have a block, we're done */
4570                                 spin_unlock(&last_ptr->refill_lock);
4571                                 goto checks;
4572                         }
4573
4574                         spin_lock(&last_ptr->lock);
4575                         /*
4576                          * whoops, this cluster doesn't actually point to
4577                          * this block group.  Get a ref on the block
4578                          * group is does point to and try again
4579                          */
4580                         if (!last_ptr_loop && last_ptr->block_group &&
4581                             last_ptr->block_group != block_group) {
4582
4583                                 btrfs_put_block_group(block_group);
4584                                 block_group = last_ptr->block_group;
4585                                 btrfs_get_block_group(block_group);
4586                                 spin_unlock(&last_ptr->lock);
4587                                 spin_unlock(&last_ptr->refill_lock);
4588
4589                                 last_ptr_loop = 1;
4590                                 search_start = block_group->key.objectid;
4591                                 /*
4592                                  * we know this block group is properly
4593                                  * in the list because
4594                                  * btrfs_remove_block_group, drops the
4595                                  * cluster before it removes the block
4596                                  * group from the list
4597                                  */
4598                                 goto have_block_group;
4599                         }
4600                         spin_unlock(&last_ptr->lock);
4601 refill_cluster:
4602                         /*
4603                          * this cluster didn't work out, free it and
4604                          * start over
4605                          */
4606                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4607
4608                         last_ptr_loop = 0;
4609
4610                         /* allocate a cluster in this block group */
4611                         ret = btrfs_find_space_cluster(trans, root,
4612                                                block_group, last_ptr,
4613                                                offset, num_bytes,
4614                                                empty_cluster + empty_size);
4615                         if (ret == 0) {
4616                                 /*
4617                                  * now pull our allocation out of this
4618                                  * cluster
4619                                  */
4620                                 offset = btrfs_alloc_from_cluster(block_group,
4621                                                   last_ptr, num_bytes,
4622                                                   search_start);
4623                                 if (offset) {
4624                                         /* we found one, proceed */
4625                                         spin_unlock(&last_ptr->refill_lock);
4626                                         goto checks;
4627                                 }
4628                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4629                                    && !failed_cluster_refill) {
4630                                 spin_unlock(&last_ptr->refill_lock);
4631
4632                                 failed_cluster_refill = true;
4633                                 wait_block_group_cache_progress(block_group,
4634                                        num_bytes + empty_cluster + empty_size);
4635                                 goto have_block_group;
4636                         }
4637
4638                         /*
4639                          * at this point we either didn't find a cluster
4640                          * or we weren't able to allocate a block from our
4641                          * cluster.  Free the cluster we've been trying
4642                          * to use, and go to the next block group
4643                          */
4644                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4645                         spin_unlock(&last_ptr->refill_lock);
4646                         goto loop;
4647                 }
4648
4649                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4650                                                     num_bytes, empty_size);
4651                 /*
4652                  * If we didn't find a chunk, and we haven't failed on this
4653                  * block group before, and this block group is in the middle of
4654                  * caching and we are ok with waiting, then go ahead and wait
4655                  * for progress to be made, and set failed_alloc to true.
4656                  *
4657                  * If failed_alloc is true then we've already waited on this
4658                  * block group once and should move on to the next block group.
4659                  */
4660                 if (!offset && !failed_alloc && !cached &&
4661                     loop > LOOP_CACHING_NOWAIT) {
4662                         wait_block_group_cache_progress(block_group,
4663                                                 num_bytes + empty_size);
4664                         failed_alloc = true;
4665                         goto have_block_group;
4666                 } else if (!offset) {
4667                         goto loop;
4668                 }
4669 checks:
4670                 search_start = stripe_align(root, offset);
4671                 /* move on to the next group */
4672                 if (search_start + num_bytes >= search_end) {
4673                         btrfs_add_free_space(block_group, offset, num_bytes);
4674                         goto loop;
4675                 }
4676
4677                 /* move on to the next group */
4678                 if (search_start + num_bytes >
4679                     block_group->key.objectid + block_group->key.offset) {
4680                         btrfs_add_free_space(block_group, offset, num_bytes);
4681                         goto loop;
4682                 }
4683
4684                 ins->objectid = search_start;
4685                 ins->offset = num_bytes;
4686
4687                 if (offset < search_start)
4688                         btrfs_add_free_space(block_group, offset,
4689                                              search_start - offset);
4690                 BUG_ON(offset > search_start);
4691
4692                 ret = update_reserved_bytes(block_group, num_bytes, 1,
4693                                             (data & BTRFS_BLOCK_GROUP_DATA));
4694                 if (ret == -EAGAIN) {
4695                         btrfs_add_free_space(block_group, offset, num_bytes);
4696                         goto loop;
4697                 }
4698
4699                 /* we are all good, lets return */
4700                 ins->objectid = search_start;
4701                 ins->offset = num_bytes;
4702
4703                 if (offset < search_start)
4704                         btrfs_add_free_space(block_group, offset,
4705                                              search_start - offset);
4706                 BUG_ON(offset > search_start);
4707                 break;
4708 loop:
4709                 failed_cluster_refill = false;
4710                 failed_alloc = false;
4711                 BUG_ON(index != get_block_group_index(block_group));
4712                 btrfs_put_block_group(block_group);
4713         }
4714         up_read(&space_info->groups_sem);
4715
4716         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
4717                 goto search;
4718
4719         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4720          *                      for them to make caching progress.  Also
4721          *                      determine the best possible bg to cache
4722          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4723          *                      caching kthreads as we move along
4724          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4725          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4726          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4727          *                      again
4728          */
4729         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4730             (found_uncached_bg || empty_size || empty_cluster ||
4731              allowed_chunk_alloc)) {
4732                 index = 0;
4733                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4734                         found_uncached_bg = false;
4735                         loop++;
4736                         if (!ideal_cache_percent &&
4737                             atomic_read(&space_info->caching_threads))
4738                                 goto search;
4739
4740                         /*
4741                          * 1 of the following 2 things have happened so far
4742                          *
4743                          * 1) We found an ideal block group for caching that
4744                          * is mostly full and will cache quickly, so we might
4745                          * as well wait for it.
4746                          *
4747                          * 2) We searched for cached only and we didn't find
4748                          * anything, and we didn't start any caching kthreads
4749                          * either, so chances are we will loop through and
4750                          * start a couple caching kthreads, and then come back
4751                          * around and just wait for them.  This will be slower
4752                          * because we will have 2 caching kthreads reading at
4753                          * the same time when we could have just started one
4754                          * and waited for it to get far enough to give us an
4755                          * allocation, so go ahead and go to the wait caching
4756                          * loop.
4757                          */
4758                         loop = LOOP_CACHING_WAIT;
4759                         search_start = ideal_cache_offset;
4760                         ideal_cache_percent = 0;
4761                         goto ideal_cache;
4762                 } else if (loop == LOOP_FIND_IDEAL) {
4763                         /*
4764                          * Didn't find a uncached bg, wait on anything we find
4765                          * next.
4766                          */
4767                         loop = LOOP_CACHING_WAIT;
4768                         goto search;
4769                 }
4770
4771                 if (loop < LOOP_CACHING_WAIT) {
4772                         loop++;
4773                         goto search;
4774                 }
4775
4776                 if (loop == LOOP_ALLOC_CHUNK) {
4777                         empty_size = 0;
4778                         empty_cluster = 0;
4779                 }
4780
4781                 if (allowed_chunk_alloc) {
4782                         ret = do_chunk_alloc(trans, root, num_bytes +
4783                                              2 * 1024 * 1024, data, 1);
4784                         allowed_chunk_alloc = 0;
4785                         done_chunk_alloc = 1;
4786                 } else if (!done_chunk_alloc) {
4787                         space_info->force_alloc = 1;
4788                 }
4789
4790                 if (loop < LOOP_NO_EMPTY_SIZE) {
4791                         loop++;
4792                         goto search;
4793                 }
4794                 ret = -ENOSPC;
4795         } else if (!ins->objectid) {
4796                 ret = -ENOSPC;
4797         }
4798
4799         /* we found what we needed */
4800         if (ins->objectid) {
4801                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4802                         trans->block_group = block_group->key.objectid;
4803
4804                 btrfs_put_block_group(block_group);
4805                 ret = 0;
4806         }
4807
4808         return ret;
4809 }
4810
4811 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4812                             int dump_block_groups)
4813 {
4814         struct btrfs_block_group_cache *cache;
4815         int index = 0;
4816
4817         spin_lock(&info->lock);
4818         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4819                (unsigned long long)(info->total_bytes - info->bytes_used -
4820                                     info->bytes_pinned - info->bytes_reserved -
4821                                     info->bytes_super),
4822                (info->full) ? "" : "not ");
4823         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4824                " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4825                "\n",
4826                (unsigned long long)info->total_bytes,
4827                (unsigned long long)info->bytes_pinned,
4828                (unsigned long long)info->bytes_delalloc,
4829                (unsigned long long)info->bytes_may_use,
4830                (unsigned long long)info->bytes_used,
4831                (unsigned long long)info->bytes_root,
4832                (unsigned long long)info->bytes_super,
4833                (unsigned long long)info->bytes_reserved);
4834         spin_unlock(&info->lock);
4835
4836         if (!dump_block_groups)
4837                 return;
4838
4839         down_read(&info->groups_sem);
4840 again:
4841         list_for_each_entry(cache, &info->block_groups[index], list) {
4842                 spin_lock(&cache->lock);
4843                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4844                        "%llu pinned %llu reserved\n",
4845                        (unsigned long long)cache->key.objectid,
4846                        (unsigned long long)cache->key.offset,
4847                        (unsigned long long)btrfs_block_group_used(&cache->item),
4848                        (unsigned long long)cache->pinned,
4849                        (unsigned long long)cache->reserved);
4850                 btrfs_dump_free_space(cache, bytes);
4851                 spin_unlock(&cache->lock);
4852         }
4853         if (++index < BTRFS_NR_RAID_TYPES)
4854                 goto again;
4855         up_read(&info->groups_sem);
4856 }
4857
4858 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4859                          struct btrfs_root *root,
4860                          u64 num_bytes, u64 min_alloc_size,
4861                          u64 empty_size, u64 hint_byte,
4862                          u64 search_end, struct btrfs_key *ins,
4863                          u64 data)
4864 {
4865         int ret;
4866         u64 search_start = 0;
4867
4868         data = btrfs_get_alloc_profile(root, data);
4869 again:
4870         /*
4871          * the only place that sets empty_size is btrfs_realloc_node, which
4872          * is not called recursively on allocations
4873          */
4874         if (empty_size || root->ref_cows)
4875                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4876                                      num_bytes + 2 * 1024 * 1024, data, 0);
4877
4878         WARN_ON(num_bytes < root->sectorsize);
4879         ret = find_free_extent(trans, root, num_bytes, empty_size,
4880                                search_start, search_end, hint_byte,
4881                                ins, data);
4882
4883         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4884                 num_bytes = num_bytes >> 1;
4885                 num_bytes = num_bytes & ~(root->sectorsize - 1);
4886                 num_bytes = max(num_bytes, min_alloc_size);
4887                 do_chunk_alloc(trans, root->fs_info->extent_root,
4888                                num_bytes, data, 1);
4889                 goto again;
4890         }
4891         if (ret == -ENOSPC) {
4892                 struct btrfs_space_info *sinfo;
4893
4894                 sinfo = __find_space_info(root->fs_info, data);
4895                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4896                        "wanted %llu\n", (unsigned long long)data,
4897                        (unsigned long long)num_bytes);
4898                 dump_space_info(sinfo, num_bytes, 1);
4899         }
4900
4901         return ret;
4902 }
4903
4904 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4905 {
4906         struct btrfs_block_group_cache *cache;
4907         int ret = 0;
4908
4909         cache = btrfs_lookup_block_group(root->fs_info, start);
4910         if (!cache) {
4911                 printk(KERN_ERR "Unable to find block group for %llu\n",
4912                        (unsigned long long)start);
4913                 return -ENOSPC;
4914         }
4915
4916         ret = btrfs_discard_extent(root, start, len);
4917
4918         btrfs_add_free_space(cache, start, len);
4919         update_reserved_bytes(cache, len, 0, 1);
4920         btrfs_put_block_group(cache);
4921
4922         return ret;
4923 }
4924
4925 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4926                                       struct btrfs_root *root,
4927                                       u64 parent, u64 root_objectid,
4928                                       u64 flags, u64 owner, u64 offset,
4929                                       struct btrfs_key *ins, int ref_mod)
4930 {
4931         int ret;
4932         struct btrfs_fs_info *fs_info = root->fs_info;
4933         struct btrfs_extent_item *extent_item;
4934         struct btrfs_extent_inline_ref *iref;
4935         struct btrfs_path *path;
4936         struct extent_buffer *leaf;
4937         int type;
4938         u32 size;
4939
4940         if (parent > 0)
4941                 type = BTRFS_SHARED_DATA_REF_KEY;
4942         else
4943                 type = BTRFS_EXTENT_DATA_REF_KEY;
4944
4945         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4946
4947         path = btrfs_alloc_path();
4948         BUG_ON(!path);
4949
4950         path->leave_spinning = 1;
4951         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4952                                       ins, size);
4953         BUG_ON(ret);
4954
4955         leaf = path->nodes[0];
4956         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4957                                      struct btrfs_extent_item);
4958         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4959         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4960         btrfs_set_extent_flags(leaf, extent_item,
4961                                flags | BTRFS_EXTENT_FLAG_DATA);
4962
4963         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4964         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4965         if (parent > 0) {
4966                 struct btrfs_shared_data_ref *ref;
4967                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4968                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4969                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4970         } else {
4971                 struct btrfs_extent_data_ref *ref;
4972                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4973                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4974                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4975                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4976                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4977         }
4978
4979         btrfs_mark_buffer_dirty(path->nodes[0]);
4980         btrfs_free_path(path);
4981
4982         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
4983         if (ret) {
4984                 printk(KERN_ERR "btrfs update block group failed for %llu "
4985                        "%llu\n", (unsigned long long)ins->objectid,
4986                        (unsigned long long)ins->offset);
4987                 BUG();
4988         }
4989         return ret;
4990 }
4991
4992 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4993                                      struct btrfs_root *root,
4994                                      u64 parent, u64 root_objectid,
4995                                      u64 flags, struct btrfs_disk_key *key,
4996                                      int level, struct btrfs_key *ins)
4997 {
4998         int ret;
4999         struct btrfs_fs_info *fs_info = root->fs_info;
5000         struct btrfs_extent_item *extent_item;
5001         struct btrfs_tree_block_info *block_info;
5002         struct btrfs_extent_inline_ref *iref;
5003         struct btrfs_path *path;
5004         struct extent_buffer *leaf;
5005         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5006
5007         path = btrfs_alloc_path();
5008         BUG_ON(!path);
5009
5010         path->leave_spinning = 1;
5011         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5012                                       ins, size);
5013         BUG_ON(ret);
5014
5015         leaf = path->nodes[0];
5016         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5017                                      struct btrfs_extent_item);
5018         btrfs_set_extent_refs(leaf, extent_item, 1);
5019         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5020         btrfs_set_extent_flags(leaf, extent_item,
5021                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5022         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5023
5024         btrfs_set_tree_block_key(leaf, block_info, key);
5025         btrfs_set_tree_block_level(leaf, block_info, level);
5026
5027         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5028         if (parent > 0) {
5029                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5030                 btrfs_set_extent_inline_ref_type(leaf, iref,
5031                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5032                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5033         } else {
5034                 btrfs_set_extent_inline_ref_type(leaf, iref,
5035                                                  BTRFS_TREE_BLOCK_REF_KEY);
5036                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5037         }
5038
5039         btrfs_mark_buffer_dirty(leaf);
5040         btrfs_free_path(path);
5041
5042         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5043         if (ret) {
5044                 printk(KERN_ERR "btrfs update block group failed for %llu "
5045                        "%llu\n", (unsigned long long)ins->objectid,
5046                        (unsigned long long)ins->offset);
5047                 BUG();
5048         }
5049         return ret;
5050 }
5051
5052 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5053                                      struct btrfs_root *root,
5054                                      u64 root_objectid, u64 owner,
5055                                      u64 offset, struct btrfs_key *ins)
5056 {
5057         int ret;
5058
5059         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5060
5061         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5062                                          0, root_objectid, owner, offset,
5063                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5064         return ret;
5065 }
5066
5067 /*
5068  * this is used by the tree logging recovery code.  It records that
5069  * an extent has been allocated and makes sure to clear the free
5070  * space cache bits as well
5071  */
5072 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5073                                    struct btrfs_root *root,
5074                                    u64 root_objectid, u64 owner, u64 offset,
5075                                    struct btrfs_key *ins)
5076 {
5077         int ret;
5078         struct btrfs_block_group_cache *block_group;
5079         struct btrfs_caching_control *caching_ctl;
5080         u64 start = ins->objectid;
5081         u64 num_bytes = ins->offset;
5082
5083         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5084         cache_block_group(block_group);
5085         caching_ctl = get_caching_control(block_group);
5086
5087         if (!caching_ctl) {
5088                 BUG_ON(!block_group_cache_done(block_group));
5089                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5090                 BUG_ON(ret);
5091         } else {
5092                 mutex_lock(&caching_ctl->mutex);
5093
5094                 if (start >= caching_ctl->progress) {
5095                         ret = add_excluded_extent(root, start, num_bytes);
5096                         BUG_ON(ret);
5097                 } else if (start + num_bytes <= caching_ctl->progress) {
5098                         ret = btrfs_remove_free_space(block_group,
5099                                                       start, num_bytes);
5100                         BUG_ON(ret);
5101                 } else {
5102                         num_bytes = caching_ctl->progress - start;
5103                         ret = btrfs_remove_free_space(block_group,
5104                                                       start, num_bytes);
5105                         BUG_ON(ret);
5106
5107                         start = caching_ctl->progress;
5108                         num_bytes = ins->objectid + ins->offset -
5109                                     caching_ctl->progress;
5110                         ret = add_excluded_extent(root, start, num_bytes);
5111                         BUG_ON(ret);
5112                 }
5113
5114                 mutex_unlock(&caching_ctl->mutex);
5115                 put_caching_control(caching_ctl);
5116         }
5117
5118         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5119         BUG_ON(ret);
5120         btrfs_put_block_group(block_group);
5121         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5122                                          0, owner, offset, ins, 1);
5123         return ret;
5124 }
5125
5126 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5127                                             struct btrfs_root *root,
5128                                             u64 bytenr, u32 blocksize,
5129                                             int level)
5130 {
5131         struct extent_buffer *buf;
5132
5133         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5134         if (!buf)
5135                 return ERR_PTR(-ENOMEM);
5136         btrfs_set_header_generation(buf, trans->transid);
5137         btrfs_set_buffer_lockdep_class(buf, level);
5138         btrfs_tree_lock(buf);
5139         clean_tree_block(trans, root, buf);
5140
5141         btrfs_set_lock_blocking(buf);
5142         btrfs_set_buffer_uptodate(buf);
5143
5144         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5145                 /*
5146                  * we allow two log transactions at a time, use different
5147                  * EXENT bit to differentiate dirty pages.
5148                  */
5149                 if (root->log_transid % 2 == 0)
5150                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5151                                         buf->start + buf->len - 1, GFP_NOFS);
5152                 else
5153                         set_extent_new(&root->dirty_log_pages, buf->start,
5154                                         buf->start + buf->len - 1, GFP_NOFS);
5155         } else {
5156                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5157                          buf->start + buf->len - 1, GFP_NOFS);
5158         }
5159         trans->blocks_used++;
5160         /* this returns a buffer locked for blocking */
5161         return buf;
5162 }
5163
5164 static struct btrfs_block_rsv *
5165 use_block_rsv(struct btrfs_trans_handle *trans,
5166               struct btrfs_root *root, u32 blocksize)
5167 {
5168         struct btrfs_block_rsv *block_rsv;
5169         int ret;
5170
5171         block_rsv = get_block_rsv(trans, root);
5172
5173         if (block_rsv->size == 0) {
5174                 ret = reserve_metadata_bytes(block_rsv, blocksize);
5175                 if (ret)
5176                         return ERR_PTR(ret);
5177                 return block_rsv;
5178         }
5179
5180         ret = block_rsv_use_bytes(block_rsv, blocksize);
5181         if (!ret)
5182                 return block_rsv;
5183
5184         WARN_ON(1);
5185         printk(KERN_INFO"block_rsv size %llu reserved %llu freed %llu %llu\n",
5186                 block_rsv->size, block_rsv->reserved,
5187                 block_rsv->freed[0], block_rsv->freed[1]);
5188
5189         return ERR_PTR(-ENOSPC);
5190 }
5191
5192 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5193 {
5194         block_rsv_add_bytes(block_rsv, blocksize, 0);
5195         block_rsv_release_bytes(block_rsv, NULL, 0);
5196 }
5197
5198 /*
5199  * finds a free extent and does all the dirty work required for allocation
5200  * returns the key for the extent through ins, and a tree buffer for
5201  * the first block of the extent through buf.
5202  *
5203  * returns the tree buffer or NULL.
5204  */
5205 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5206                                         struct btrfs_root *root, u32 blocksize,
5207                                         u64 parent, u64 root_objectid,
5208                                         struct btrfs_disk_key *key, int level,
5209                                         u64 hint, u64 empty_size)
5210 {
5211         struct btrfs_key ins;
5212         struct btrfs_block_rsv *block_rsv;
5213         struct extent_buffer *buf;
5214         u64 flags = 0;
5215         int ret;
5216
5217
5218         block_rsv = use_block_rsv(trans, root, blocksize);
5219         if (IS_ERR(block_rsv))
5220                 return ERR_CAST(block_rsv);
5221
5222         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5223                                    empty_size, hint, (u64)-1, &ins, 0);
5224         if (ret) {
5225                 unuse_block_rsv(block_rsv, blocksize);
5226                 return ERR_PTR(ret);
5227         }
5228
5229         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5230                                     blocksize, level);
5231         BUG_ON(IS_ERR(buf));
5232
5233         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5234                 if (parent == 0)
5235                         parent = ins.objectid;
5236                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5237         } else
5238                 BUG_ON(parent > 0);
5239
5240         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5241                 struct btrfs_delayed_extent_op *extent_op;
5242                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5243                 BUG_ON(!extent_op);
5244                 if (key)
5245                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5246                 else
5247                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5248                 extent_op->flags_to_set = flags;
5249                 extent_op->update_key = 1;
5250                 extent_op->update_flags = 1;
5251                 extent_op->is_data = 0;
5252
5253                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5254                                         ins.offset, parent, root_objectid,
5255                                         level, BTRFS_ADD_DELAYED_EXTENT,
5256                                         extent_op);
5257                 BUG_ON(ret);
5258         }
5259         return buf;
5260 }
5261
5262 struct walk_control {
5263         u64 refs[BTRFS_MAX_LEVEL];
5264         u64 flags[BTRFS_MAX_LEVEL];
5265         struct btrfs_key update_progress;
5266         int stage;
5267         int level;
5268         int shared_level;
5269         int update_ref;
5270         int keep_locks;
5271         int reada_slot;
5272         int reada_count;
5273 };
5274
5275 #define DROP_REFERENCE  1
5276 #define UPDATE_BACKREF  2
5277
5278 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5279                                      struct btrfs_root *root,
5280                                      struct walk_control *wc,
5281                                      struct btrfs_path *path)
5282 {
5283         u64 bytenr;
5284         u64 generation;
5285         u64 refs;
5286         u64 flags;
5287         u64 last = 0;
5288         u32 nritems;
5289         u32 blocksize;
5290         struct btrfs_key key;
5291         struct extent_buffer *eb;
5292         int ret;
5293         int slot;
5294         int nread = 0;
5295
5296         if (path->slots[wc->level] < wc->reada_slot) {
5297                 wc->reada_count = wc->reada_count * 2 / 3;
5298                 wc->reada_count = max(wc->reada_count, 2);
5299         } else {
5300                 wc->reada_count = wc->reada_count * 3 / 2;
5301                 wc->reada_count = min_t(int, wc->reada_count,
5302                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5303         }
5304
5305         eb = path->nodes[wc->level];
5306         nritems = btrfs_header_nritems(eb);
5307         blocksize = btrfs_level_size(root, wc->level - 1);
5308
5309         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5310                 if (nread >= wc->reada_count)
5311                         break;
5312
5313                 cond_resched();
5314                 bytenr = btrfs_node_blockptr(eb, slot);
5315                 generation = btrfs_node_ptr_generation(eb, slot);
5316
5317                 if (slot == path->slots[wc->level])
5318                         goto reada;
5319
5320                 if (wc->stage == UPDATE_BACKREF &&
5321                     generation <= root->root_key.offset)
5322                         continue;
5323
5324                 /* We don't lock the tree block, it's OK to be racy here */
5325                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5326                                                &refs, &flags);
5327                 BUG_ON(ret);
5328                 BUG_ON(refs == 0);
5329
5330                 if (wc->stage == DROP_REFERENCE) {
5331                         if (refs == 1)
5332                                 goto reada;
5333
5334                         if (wc->level == 1 &&
5335                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5336                                 continue;
5337                         if (!wc->update_ref ||
5338                             generation <= root->root_key.offset)
5339                                 continue;
5340                         btrfs_node_key_to_cpu(eb, &key, slot);
5341                         ret = btrfs_comp_cpu_keys(&key,
5342                                                   &wc->update_progress);
5343                         if (ret < 0)
5344                                 continue;
5345                 } else {
5346                         if (wc->level == 1 &&
5347                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5348                                 continue;
5349                 }
5350 reada:
5351                 ret = readahead_tree_block(root, bytenr, blocksize,
5352                                            generation);
5353                 if (ret)
5354                         break;
5355                 last = bytenr + blocksize;
5356                 nread++;
5357         }
5358         wc->reada_slot = slot;
5359 }
5360
5361 /*
5362  * hepler to process tree block while walking down the tree.
5363  *
5364  * when wc->stage == UPDATE_BACKREF, this function updates
5365  * back refs for pointers in the block.
5366  *
5367  * NOTE: return value 1 means we should stop walking down.
5368  */
5369 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5370                                    struct btrfs_root *root,
5371                                    struct btrfs_path *path,
5372                                    struct walk_control *wc, int lookup_info)
5373 {
5374         int level = wc->level;
5375         struct extent_buffer *eb = path->nodes[level];
5376         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5377         int ret;
5378
5379         if (wc->stage == UPDATE_BACKREF &&
5380             btrfs_header_owner(eb) != root->root_key.objectid)
5381                 return 1;
5382
5383         /*
5384          * when reference count of tree block is 1, it won't increase
5385          * again. once full backref flag is set, we never clear it.
5386          */
5387         if (lookup_info &&
5388             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5389              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5390                 BUG_ON(!path->locks[level]);
5391                 ret = btrfs_lookup_extent_info(trans, root,
5392                                                eb->start, eb->len,
5393                                                &wc->refs[level],
5394                                                &wc->flags[level]);
5395                 BUG_ON(ret);
5396                 BUG_ON(wc->refs[level] == 0);
5397         }
5398
5399         if (wc->stage == DROP_REFERENCE) {
5400                 if (wc->refs[level] > 1)
5401                         return 1;
5402
5403                 if (path->locks[level] && !wc->keep_locks) {
5404                         btrfs_tree_unlock(eb);
5405                         path->locks[level] = 0;
5406                 }
5407                 return 0;
5408         }
5409
5410         /* wc->stage == UPDATE_BACKREF */
5411         if (!(wc->flags[level] & flag)) {
5412                 BUG_ON(!path->locks[level]);
5413                 ret = btrfs_inc_ref(trans, root, eb, 1);
5414                 BUG_ON(ret);
5415                 ret = btrfs_dec_ref(trans, root, eb, 0);
5416                 BUG_ON(ret);
5417                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5418                                                   eb->len, flag, 0);
5419                 BUG_ON(ret);
5420                 wc->flags[level] |= flag;
5421         }
5422
5423         /*
5424          * the block is shared by multiple trees, so it's not good to
5425          * keep the tree lock
5426          */
5427         if (path->locks[level] && level > 0) {
5428                 btrfs_tree_unlock(eb);
5429                 path->locks[level] = 0;
5430         }
5431         return 0;
5432 }
5433
5434 /*
5435  * hepler to process tree block pointer.
5436  *
5437  * when wc->stage == DROP_REFERENCE, this function checks
5438  * reference count of the block pointed to. if the block
5439  * is shared and we need update back refs for the subtree
5440  * rooted at the block, this function changes wc->stage to
5441  * UPDATE_BACKREF. if the block is shared and there is no
5442  * need to update back, this function drops the reference
5443  * to the block.
5444  *
5445  * NOTE: return value 1 means we should stop walking down.
5446  */
5447 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5448                                  struct btrfs_root *root,
5449                                  struct btrfs_path *path,
5450                                  struct walk_control *wc, int *lookup_info)
5451 {
5452         u64 bytenr;
5453         u64 generation;
5454         u64 parent;
5455         u32 blocksize;
5456         struct btrfs_key key;
5457         struct extent_buffer *next;
5458         int level = wc->level;
5459         int reada = 0;
5460         int ret = 0;
5461
5462         generation = btrfs_node_ptr_generation(path->nodes[level],
5463                                                path->slots[level]);
5464         /*
5465          * if the lower level block was created before the snapshot
5466          * was created, we know there is no need to update back refs
5467          * for the subtree
5468          */
5469         if (wc->stage == UPDATE_BACKREF &&
5470             generation <= root->root_key.offset) {
5471                 *lookup_info = 1;
5472                 return 1;
5473         }
5474
5475         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5476         blocksize = btrfs_level_size(root, level - 1);
5477
5478         next = btrfs_find_tree_block(root, bytenr, blocksize);
5479         if (!next) {
5480                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5481                 if (!next)
5482                         return -ENOMEM;
5483                 reada = 1;
5484         }
5485         btrfs_tree_lock(next);
5486         btrfs_set_lock_blocking(next);
5487
5488         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5489                                        &wc->refs[level - 1],
5490                                        &wc->flags[level - 1]);
5491         BUG_ON(ret);
5492         BUG_ON(wc->refs[level - 1] == 0);
5493         *lookup_info = 0;
5494
5495         if (wc->stage == DROP_REFERENCE) {
5496                 if (wc->refs[level - 1] > 1) {
5497                         if (level == 1 &&
5498                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5499                                 goto skip;
5500
5501                         if (!wc->update_ref ||
5502                             generation <= root->root_key.offset)
5503                                 goto skip;
5504
5505                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5506                                               path->slots[level]);
5507                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5508                         if (ret < 0)
5509                                 goto skip;
5510
5511                         wc->stage = UPDATE_BACKREF;
5512                         wc->shared_level = level - 1;
5513                 }
5514         } else {
5515                 if (level == 1 &&
5516                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5517                         goto skip;
5518         }
5519
5520         if (!btrfs_buffer_uptodate(next, generation)) {
5521                 btrfs_tree_unlock(next);
5522                 free_extent_buffer(next);
5523                 next = NULL;
5524                 *lookup_info = 1;
5525         }
5526
5527         if (!next) {
5528                 if (reada && level == 1)
5529                         reada_walk_down(trans, root, wc, path);
5530                 next = read_tree_block(root, bytenr, blocksize, generation);
5531                 btrfs_tree_lock(next);
5532                 btrfs_set_lock_blocking(next);
5533         }
5534
5535         level--;
5536         BUG_ON(level != btrfs_header_level(next));
5537         path->nodes[level] = next;
5538         path->slots[level] = 0;
5539         path->locks[level] = 1;
5540         wc->level = level;
5541         if (wc->level == 1)
5542                 wc->reada_slot = 0;
5543         return 0;
5544 skip:
5545         wc->refs[level - 1] = 0;
5546         wc->flags[level - 1] = 0;
5547         if (wc->stage == DROP_REFERENCE) {
5548                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5549                         parent = path->nodes[level]->start;
5550                 } else {
5551                         BUG_ON(root->root_key.objectid !=
5552                                btrfs_header_owner(path->nodes[level]));
5553                         parent = 0;
5554                 }
5555
5556                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5557                                         root->root_key.objectid, level - 1, 0);
5558                 BUG_ON(ret);
5559         }
5560         btrfs_tree_unlock(next);
5561         free_extent_buffer(next);
5562         *lookup_info = 1;
5563         return 1;
5564 }
5565
5566 /*
5567  * hepler to process tree block while walking up the tree.
5568  *
5569  * when wc->stage == DROP_REFERENCE, this function drops
5570  * reference count on the block.
5571  *
5572  * when wc->stage == UPDATE_BACKREF, this function changes
5573  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5574  * to UPDATE_BACKREF previously while processing the block.
5575  *
5576  * NOTE: return value 1 means we should stop walking up.
5577  */
5578 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5579                                  struct btrfs_root *root,
5580                                  struct btrfs_path *path,
5581                                  struct walk_control *wc)
5582 {
5583         int ret;
5584         int level = wc->level;
5585         struct extent_buffer *eb = path->nodes[level];
5586         u64 parent = 0;
5587
5588         if (wc->stage == UPDATE_BACKREF) {
5589                 BUG_ON(wc->shared_level < level);
5590                 if (level < wc->shared_level)
5591                         goto out;
5592
5593                 ret = find_next_key(path, level + 1, &wc->update_progress);
5594                 if (ret > 0)
5595                         wc->update_ref = 0;
5596
5597                 wc->stage = DROP_REFERENCE;
5598                 wc->shared_level = -1;
5599                 path->slots[level] = 0;
5600
5601                 /*
5602                  * check reference count again if the block isn't locked.
5603                  * we should start walking down the tree again if reference
5604                  * count is one.
5605                  */
5606                 if (!path->locks[level]) {
5607                         BUG_ON(level == 0);
5608                         btrfs_tree_lock(eb);
5609                         btrfs_set_lock_blocking(eb);
5610                         path->locks[level] = 1;
5611
5612                         ret = btrfs_lookup_extent_info(trans, root,
5613                                                        eb->start, eb->len,
5614                                                        &wc->refs[level],
5615                                                        &wc->flags[level]);
5616                         BUG_ON(ret);
5617                         BUG_ON(wc->refs[level] == 0);
5618                         if (wc->refs[level] == 1) {
5619                                 btrfs_tree_unlock(eb);
5620                                 path->locks[level] = 0;
5621                                 return 1;
5622                         }
5623                 }
5624         }
5625
5626         /* wc->stage == DROP_REFERENCE */
5627         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5628
5629         if (wc->refs[level] == 1) {
5630                 if (level == 0) {
5631                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5632                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5633                         else
5634                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5635                         BUG_ON(ret);
5636                 }
5637                 /* make block locked assertion in clean_tree_block happy */
5638                 if (!path->locks[level] &&
5639                     btrfs_header_generation(eb) == trans->transid) {
5640                         btrfs_tree_lock(eb);
5641                         btrfs_set_lock_blocking(eb);
5642                         path->locks[level] = 1;
5643                 }
5644                 clean_tree_block(trans, root, eb);
5645         }
5646
5647         if (eb == root->node) {
5648                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5649                         parent = eb->start;
5650                 else
5651                         BUG_ON(root->root_key.objectid !=
5652                                btrfs_header_owner(eb));
5653         } else {
5654                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5655                         parent = path->nodes[level + 1]->start;
5656                 else
5657                         BUG_ON(root->root_key.objectid !=
5658                                btrfs_header_owner(path->nodes[level + 1]));
5659         }
5660
5661         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
5662 out:
5663         wc->refs[level] = 0;
5664         wc->flags[level] = 0;
5665         return 0;
5666 }
5667
5668 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5669                                    struct btrfs_root *root,
5670                                    struct btrfs_path *path,
5671                                    struct walk_control *wc)
5672 {
5673         int level = wc->level;
5674         int lookup_info = 1;
5675         int ret;
5676
5677         while (level >= 0) {
5678                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5679                 if (ret > 0)
5680                         break;
5681
5682                 if (level == 0)
5683                         break;
5684
5685                 if (path->slots[level] >=
5686                     btrfs_header_nritems(path->nodes[level]))
5687                         break;
5688
5689                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5690                 if (ret > 0) {
5691                         path->slots[level]++;
5692                         continue;
5693                 } else if (ret < 0)
5694                         return ret;
5695                 level = wc->level;
5696         }
5697         return 0;
5698 }
5699
5700 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5701                                  struct btrfs_root *root,
5702                                  struct btrfs_path *path,
5703                                  struct walk_control *wc, int max_level)
5704 {
5705         int level = wc->level;
5706         int ret;
5707
5708         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5709         while (level < max_level && path->nodes[level]) {
5710                 wc->level = level;
5711                 if (path->slots[level] + 1 <
5712                     btrfs_header_nritems(path->nodes[level])) {
5713                         path->slots[level]++;
5714                         return 0;
5715                 } else {
5716                         ret = walk_up_proc(trans, root, path, wc);
5717                         if (ret > 0)
5718                                 return 0;
5719
5720                         if (path->locks[level]) {
5721                                 btrfs_tree_unlock(path->nodes[level]);
5722                                 path->locks[level] = 0;
5723                         }
5724                         free_extent_buffer(path->nodes[level]);
5725                         path->nodes[level] = NULL;
5726                         level++;
5727                 }
5728         }
5729         return 1;
5730 }
5731
5732 /*
5733  * drop a subvolume tree.
5734  *
5735  * this function traverses the tree freeing any blocks that only
5736  * referenced by the tree.
5737  *
5738  * when a shared tree block is found. this function decreases its
5739  * reference count by one. if update_ref is true, this function
5740  * also make sure backrefs for the shared block and all lower level
5741  * blocks are properly updated.
5742  */
5743 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5744 {
5745         struct btrfs_path *path;
5746         struct btrfs_trans_handle *trans;
5747         struct btrfs_root *tree_root = root->fs_info->tree_root;
5748         struct btrfs_root_item *root_item = &root->root_item;
5749         struct walk_control *wc;
5750         struct btrfs_key key;
5751         int err = 0;
5752         int ret;
5753         int level;
5754
5755         path = btrfs_alloc_path();
5756         BUG_ON(!path);
5757
5758         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5759         BUG_ON(!wc);
5760
5761         trans = btrfs_start_transaction(tree_root, 0);
5762
5763         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5764                 level = btrfs_header_level(root->node);
5765                 path->nodes[level] = btrfs_lock_root_node(root);
5766                 btrfs_set_lock_blocking(path->nodes[level]);
5767                 path->slots[level] = 0;
5768                 path->locks[level] = 1;
5769                 memset(&wc->update_progress, 0,
5770                        sizeof(wc->update_progress));
5771         } else {
5772                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5773                 memcpy(&wc->update_progress, &key,
5774                        sizeof(wc->update_progress));
5775
5776                 level = root_item->drop_level;
5777                 BUG_ON(level == 0);
5778                 path->lowest_level = level;
5779                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5780                 path->lowest_level = 0;
5781                 if (ret < 0) {
5782                         err = ret;
5783                         goto out;
5784                 }
5785                 WARN_ON(ret > 0);
5786
5787                 /*
5788                  * unlock our path, this is safe because only this
5789                  * function is allowed to delete this snapshot
5790                  */
5791                 btrfs_unlock_up_safe(path, 0);
5792
5793                 level = btrfs_header_level(root->node);
5794                 while (1) {
5795                         btrfs_tree_lock(path->nodes[level]);
5796                         btrfs_set_lock_blocking(path->nodes[level]);
5797
5798                         ret = btrfs_lookup_extent_info(trans, root,
5799                                                 path->nodes[level]->start,
5800                                                 path->nodes[level]->len,
5801                                                 &wc->refs[level],
5802                                                 &wc->flags[level]);
5803                         BUG_ON(ret);
5804                         BUG_ON(wc->refs[level] == 0);
5805
5806                         if (level == root_item->drop_level)
5807                                 break;
5808
5809                         btrfs_tree_unlock(path->nodes[level]);
5810                         WARN_ON(wc->refs[level] != 1);
5811                         level--;
5812                 }
5813         }
5814
5815         wc->level = level;
5816         wc->shared_level = -1;
5817         wc->stage = DROP_REFERENCE;
5818         wc->update_ref = update_ref;
5819         wc->keep_locks = 0;
5820         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5821
5822         while (1) {
5823                 ret = walk_down_tree(trans, root, path, wc);
5824                 if (ret < 0) {
5825                         err = ret;
5826                         break;
5827                 }
5828
5829                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5830                 if (ret < 0) {
5831                         err = ret;
5832                         break;
5833                 }
5834
5835                 if (ret > 0) {
5836                         BUG_ON(wc->stage != DROP_REFERENCE);
5837                         break;
5838                 }
5839
5840                 if (wc->stage == DROP_REFERENCE) {
5841                         level = wc->level;
5842                         btrfs_node_key(path->nodes[level],
5843                                        &root_item->drop_progress,
5844                                        path->slots[level]);
5845                         root_item->drop_level = level;
5846                 }
5847
5848                 BUG_ON(wc->level == 0);
5849                 if (trans->transaction->in_commit ||
5850                     trans->transaction->delayed_refs.flushing) {
5851                         ret = btrfs_update_root(trans, tree_root,
5852                                                 &root->root_key,
5853                                                 root_item);
5854                         BUG_ON(ret);
5855
5856                         btrfs_end_transaction(trans, tree_root);
5857                         trans = btrfs_start_transaction(tree_root, 0);
5858                         if (IS_ERR(trans))
5859                                 return PTR_ERR(trans);
5860                 } else {
5861                         unsigned long update;
5862                         update = trans->delayed_ref_updates;
5863                         trans->delayed_ref_updates = 0;
5864                         if (update)
5865                                 btrfs_run_delayed_refs(trans, tree_root,
5866                                                        update);
5867                 }
5868         }
5869         btrfs_release_path(root, path);
5870         BUG_ON(err);
5871
5872         ret = btrfs_del_root(trans, tree_root, &root->root_key);
5873         BUG_ON(ret);
5874
5875         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5876                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5877                                            NULL, NULL);
5878                 BUG_ON(ret < 0);
5879                 if (ret > 0) {
5880                         ret = btrfs_del_orphan_item(trans, tree_root,
5881                                                     root->root_key.objectid);
5882                         BUG_ON(ret);
5883                 }
5884         }
5885
5886         if (root->in_radix) {
5887                 btrfs_free_fs_root(tree_root->fs_info, root);
5888         } else {
5889                 free_extent_buffer(root->node);
5890                 free_extent_buffer(root->commit_root);
5891                 kfree(root);
5892         }
5893 out:
5894         btrfs_end_transaction(trans, tree_root);
5895         kfree(wc);
5896         btrfs_free_path(path);
5897         return err;
5898 }
5899
5900 /*
5901  * drop subtree rooted at tree block 'node'.
5902  *
5903  * NOTE: this function will unlock and release tree block 'node'
5904  */
5905 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5906                         struct btrfs_root *root,
5907                         struct extent_buffer *node,
5908                         struct extent_buffer *parent)
5909 {
5910         struct btrfs_path *path;
5911         struct walk_control *wc;
5912         int level;
5913         int parent_level;
5914         int ret = 0;
5915         int wret;
5916
5917         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5918
5919         path = btrfs_alloc_path();
5920         BUG_ON(!path);
5921
5922         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5923         BUG_ON(!wc);
5924
5925         btrfs_assert_tree_locked(parent);
5926         parent_level = btrfs_header_level(parent);
5927         extent_buffer_get(parent);
5928         path->nodes[parent_level] = parent;
5929         path->slots[parent_level] = btrfs_header_nritems(parent);
5930
5931         btrfs_assert_tree_locked(node);
5932         level = btrfs_header_level(node);
5933         path->nodes[level] = node;
5934         path->slots[level] = 0;
5935         path->locks[level] = 1;
5936
5937         wc->refs[parent_level] = 1;
5938         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5939         wc->level = level;
5940         wc->shared_level = -1;
5941         wc->stage = DROP_REFERENCE;
5942         wc->update_ref = 0;
5943         wc->keep_locks = 1;
5944         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5945
5946         while (1) {
5947                 wret = walk_down_tree(trans, root, path, wc);
5948                 if (wret < 0) {
5949                         ret = wret;
5950                         break;
5951                 }
5952
5953                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5954                 if (wret < 0)
5955                         ret = wret;
5956                 if (wret != 0)
5957                         break;
5958         }
5959
5960         kfree(wc);
5961         btrfs_free_path(path);
5962         return ret;
5963 }
5964
5965 #if 0
5966 static unsigned long calc_ra(unsigned long start, unsigned long last,
5967                              unsigned long nr)
5968 {
5969         return min(last, start + nr - 1);
5970 }
5971
5972 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5973                                          u64 len)
5974 {
5975         u64 page_start;
5976         u64 page_end;
5977         unsigned long first_index;
5978         unsigned long last_index;
5979         unsigned long i;
5980         struct page *page;
5981         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5982         struct file_ra_state *ra;
5983         struct btrfs_ordered_extent *ordered;
5984         unsigned int total_read = 0;
5985         unsigned int total_dirty = 0;
5986         int ret = 0;
5987
5988         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5989
5990         mutex_lock(&inode->i_mutex);
5991         first_index = start >> PAGE_CACHE_SHIFT;
5992         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5993
5994         /* make sure the dirty trick played by the caller work */
5995         ret = invalidate_inode_pages2_range(inode->i_mapping,
5996                                             first_index, last_index);
5997         if (ret)
5998                 goto out_unlock;
5999
6000         file_ra_state_init(ra, inode->i_mapping);
6001
6002         for (i = first_index ; i <= last_index; i++) {
6003                 if (total_read % ra->ra_pages == 0) {
6004                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6005                                        calc_ra(i, last_index, ra->ra_pages));
6006                 }
6007                 total_read++;
6008 again:
6009                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6010                         BUG_ON(1);
6011                 page = grab_cache_page(inode->i_mapping, i);
6012                 if (!page) {
6013                         ret = -ENOMEM;
6014                         goto out_unlock;
6015                 }
6016                 if (!PageUptodate(page)) {
6017                         btrfs_readpage(NULL, page);
6018                         lock_page(page);
6019                         if (!PageUptodate(page)) {
6020                                 unlock_page(page);
6021                                 page_cache_release(page);
6022                                 ret = -EIO;
6023                                 goto out_unlock;
6024                         }
6025                 }
6026                 wait_on_page_writeback(page);
6027
6028                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6029                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6030                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6031
6032                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6033                 if (ordered) {
6034                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6035                         unlock_page(page);
6036                         page_cache_release(page);
6037                         btrfs_start_ordered_extent(inode, ordered, 1);
6038                         btrfs_put_ordered_extent(ordered);
6039                         goto again;
6040                 }
6041                 set_page_extent_mapped(page);
6042
6043                 if (i == first_index)
6044                         set_extent_bits(io_tree, page_start, page_end,
6045                                         EXTENT_BOUNDARY, GFP_NOFS);
6046                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6047
6048                 set_page_dirty(page);
6049                 total_dirty++;
6050
6051                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6052                 unlock_page(page);
6053                 page_cache_release(page);
6054         }
6055
6056 out_unlock:
6057         kfree(ra);
6058         mutex_unlock(&inode->i_mutex);
6059         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6060         return ret;
6061 }
6062
6063 static noinline int relocate_data_extent(struct inode *reloc_inode,
6064                                          struct btrfs_key *extent_key,
6065                                          u64 offset)
6066 {
6067         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6068         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6069         struct extent_map *em;
6070         u64 start = extent_key->objectid - offset;
6071         u64 end = start + extent_key->offset - 1;
6072
6073         em = alloc_extent_map(GFP_NOFS);
6074         BUG_ON(!em || IS_ERR(em));
6075
6076         em->start = start;
6077         em->len = extent_key->offset;
6078         em->block_len = extent_key->offset;
6079         em->block_start = extent_key->objectid;
6080         em->bdev = root->fs_info->fs_devices->latest_bdev;
6081         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6082
6083         /* setup extent map to cheat btrfs_readpage */
6084         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6085         while (1) {
6086                 int ret;
6087                 write_lock(&em_tree->lock);
6088                 ret = add_extent_mapping(em_tree, em);
6089                 write_unlock(&em_tree->lock);
6090                 if (ret != -EEXIST) {
6091                         free_extent_map(em);
6092                         break;
6093                 }
6094                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6095         }
6096         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6097
6098         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6099 }
6100
6101 struct btrfs_ref_path {
6102         u64 extent_start;
6103         u64 nodes[BTRFS_MAX_LEVEL];
6104         u64 root_objectid;
6105         u64 root_generation;
6106         u64 owner_objectid;
6107         u32 num_refs;
6108         int lowest_level;
6109         int current_level;
6110         int shared_level;
6111
6112         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6113         u64 new_nodes[BTRFS_MAX_LEVEL];
6114 };
6115
6116 struct disk_extent {
6117         u64 ram_bytes;
6118         u64 disk_bytenr;
6119         u64 disk_num_bytes;
6120         u64 offset;
6121         u64 num_bytes;
6122         u8 compression;
6123         u8 encryption;
6124         u16 other_encoding;
6125 };
6126
6127 static int is_cowonly_root(u64 root_objectid)
6128 {
6129         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6130             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6131             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6132             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6133             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6134             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6135                 return 1;
6136         return 0;
6137 }
6138
6139 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6140                                     struct btrfs_root *extent_root,
6141                                     struct btrfs_ref_path *ref_path,
6142                                     int first_time)
6143 {
6144         struct extent_buffer *leaf;
6145         struct btrfs_path *path;
6146         struct btrfs_extent_ref *ref;
6147         struct btrfs_key key;
6148         struct btrfs_key found_key;
6149         u64 bytenr;
6150         u32 nritems;
6151         int level;
6152         int ret = 1;
6153
6154         path = btrfs_alloc_path();
6155         if (!path)
6156                 return -ENOMEM;
6157
6158         if (first_time) {
6159                 ref_path->lowest_level = -1;
6160                 ref_path->current_level = -1;
6161                 ref_path->shared_level = -1;
6162                 goto walk_up;
6163         }
6164 walk_down:
6165         level = ref_path->current_level - 1;
6166         while (level >= -1) {
6167                 u64 parent;
6168                 if (level < ref_path->lowest_level)
6169                         break;
6170
6171                 if (level >= 0)
6172                         bytenr = ref_path->nodes[level];
6173                 else
6174                         bytenr = ref_path->extent_start;
6175                 BUG_ON(bytenr == 0);
6176
6177                 parent = ref_path->nodes[level + 1];
6178                 ref_path->nodes[level + 1] = 0;
6179                 ref_path->current_level = level;
6180                 BUG_ON(parent == 0);
6181
6182                 key.objectid = bytenr;
6183                 key.offset = parent + 1;
6184                 key.type = BTRFS_EXTENT_REF_KEY;
6185
6186                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6187                 if (ret < 0)
6188                         goto out;
6189                 BUG_ON(ret == 0);
6190
6191                 leaf = path->nodes[0];
6192                 nritems = btrfs_header_nritems(leaf);
6193                 if (path->slots[0] >= nritems) {
6194                         ret = btrfs_next_leaf(extent_root, path);
6195                         if (ret < 0)
6196                                 goto out;
6197                         if (ret > 0)
6198                                 goto next;
6199                         leaf = path->nodes[0];
6200                 }
6201
6202                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6203                 if (found_key.objectid == bytenr &&
6204                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6205                         if (level < ref_path->shared_level)
6206                                 ref_path->shared_level = level;
6207                         goto found;
6208                 }
6209 next:
6210                 level--;
6211                 btrfs_release_path(extent_root, path);
6212                 cond_resched();
6213         }
6214         /* reached lowest level */
6215         ret = 1;
6216         goto out;
6217 walk_up:
6218         level = ref_path->current_level;
6219         while (level < BTRFS_MAX_LEVEL - 1) {
6220                 u64 ref_objectid;
6221
6222                 if (level >= 0)
6223                         bytenr = ref_path->nodes[level];
6224                 else
6225                         bytenr = ref_path->extent_start;
6226
6227                 BUG_ON(bytenr == 0);
6228
6229                 key.objectid = bytenr;
6230                 key.offset = 0;
6231                 key.type = BTRFS_EXTENT_REF_KEY;
6232
6233                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6234                 if (ret < 0)
6235                         goto out;
6236
6237                 leaf = path->nodes[0];
6238                 nritems = btrfs_header_nritems(leaf);
6239                 if (path->slots[0] >= nritems) {
6240                         ret = btrfs_next_leaf(extent_root, path);
6241                         if (ret < 0)
6242                                 goto out;
6243                         if (ret > 0) {
6244                                 /* the extent was freed by someone */
6245                                 if (ref_path->lowest_level == level)
6246                                         goto out;
6247                                 btrfs_release_path(extent_root, path);
6248                                 goto walk_down;
6249                         }
6250                         leaf = path->nodes[0];
6251                 }
6252
6253                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6254                 if (found_key.objectid != bytenr ||
6255                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6256                         /* the extent was freed by someone */
6257                         if (ref_path->lowest_level == level) {
6258                                 ret = 1;
6259                                 goto out;
6260                         }
6261                         btrfs_release_path(extent_root, path);
6262                         goto walk_down;
6263                 }
6264 found:
6265                 ref = btrfs_item_ptr(leaf, path->slots[0],
6266                                 struct btrfs_extent_ref);
6267                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6268                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6269                         if (first_time) {
6270                                 level = (int)ref_objectid;
6271                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6272                                 ref_path->lowest_level = level;
6273                                 ref_path->current_level = level;
6274                                 ref_path->nodes[level] = bytenr;
6275                         } else {
6276                                 WARN_ON(ref_objectid != level);
6277                         }
6278                 } else {
6279                         WARN_ON(level != -1);
6280                 }
6281                 first_time = 0;
6282
6283                 if (ref_path->lowest_level == level) {
6284                         ref_path->owner_objectid = ref_objectid;
6285                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6286                 }
6287
6288                 /*
6289                  * the block is tree root or the block isn't in reference
6290                  * counted tree.
6291                  */
6292                 if (found_key.objectid == found_key.offset ||
6293                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6294                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6295                         ref_path->root_generation =
6296                                 btrfs_ref_generation(leaf, ref);
6297                         if (level < 0) {
6298                                 /* special reference from the tree log */
6299                                 ref_path->nodes[0] = found_key.offset;
6300                                 ref_path->current_level = 0;
6301                         }
6302                         ret = 0;
6303                         goto out;
6304                 }
6305
6306                 level++;
6307                 BUG_ON(ref_path->nodes[level] != 0);
6308                 ref_path->nodes[level] = found_key.offset;
6309                 ref_path->current_level = level;
6310
6311                 /*
6312                  * the reference was created in the running transaction,
6313                  * no need to continue walking up.
6314                  */
6315                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6316                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6317                         ref_path->root_generation =
6318                                 btrfs_ref_generation(leaf, ref);
6319                         ret = 0;
6320                         goto out;
6321                 }
6322
6323                 btrfs_release_path(extent_root, path);
6324                 cond_resched();
6325         }
6326         /* reached max tree level, but no tree root found. */
6327         BUG();
6328 out:
6329         btrfs_free_path(path);
6330         return ret;
6331 }
6332
6333 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6334                                 struct btrfs_root *extent_root,
6335                                 struct btrfs_ref_path *ref_path,
6336                                 u64 extent_start)
6337 {
6338         memset(ref_path, 0, sizeof(*ref_path));
6339         ref_path->extent_start = extent_start;
6340
6341         return __next_ref_path(trans, extent_root, ref_path, 1);
6342 }
6343
6344 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6345                                struct btrfs_root *extent_root,
6346                                struct btrfs_ref_path *ref_path)
6347 {
6348         return __next_ref_path(trans, extent_root, ref_path, 0);
6349 }
6350
6351 static noinline int get_new_locations(struct inode *reloc_inode,
6352                                       struct btrfs_key *extent_key,
6353                                       u64 offset, int no_fragment,
6354                                       struct disk_extent **extents,
6355                                       int *nr_extents)
6356 {
6357         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6358         struct btrfs_path *path;
6359         struct btrfs_file_extent_item *fi;
6360         struct extent_buffer *leaf;
6361         struct disk_extent *exts = *extents;
6362         struct btrfs_key found_key;
6363         u64 cur_pos;
6364         u64 last_byte;
6365         u32 nritems;
6366         int nr = 0;
6367         int max = *nr_extents;
6368         int ret;
6369
6370         WARN_ON(!no_fragment && *extents);
6371         if (!exts) {
6372                 max = 1;
6373                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6374                 if (!exts)
6375                         return -ENOMEM;
6376         }
6377
6378         path = btrfs_alloc_path();
6379         BUG_ON(!path);
6380
6381         cur_pos = extent_key->objectid - offset;
6382         last_byte = extent_key->objectid + extent_key->offset;
6383         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6384                                        cur_pos, 0);
6385         if (ret < 0)
6386                 goto out;
6387         if (ret > 0) {
6388                 ret = -ENOENT;
6389                 goto out;
6390         }
6391
6392         while (1) {
6393                 leaf = path->nodes[0];
6394                 nritems = btrfs_header_nritems(leaf);
6395                 if (path->slots[0] >= nritems) {
6396                         ret = btrfs_next_leaf(root, path);
6397                         if (ret < 0)
6398                                 goto out;
6399                         if (ret > 0)
6400                                 break;
6401                         leaf = path->nodes[0];
6402                 }
6403
6404                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6405                 if (found_key.offset != cur_pos ||
6406                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6407                     found_key.objectid != reloc_inode->i_ino)
6408                         break;
6409
6410                 fi = btrfs_item_ptr(leaf, path->slots[0],
6411                                     struct btrfs_file_extent_item);
6412                 if (btrfs_file_extent_type(leaf, fi) !=
6413                     BTRFS_FILE_EXTENT_REG ||
6414                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6415                         break;
6416
6417                 if (nr == max) {
6418                         struct disk_extent *old = exts;
6419                         max *= 2;
6420                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6421                         memcpy(exts, old, sizeof(*exts) * nr);
6422                         if (old != *extents)
6423                                 kfree(old);
6424                 }
6425
6426                 exts[nr].disk_bytenr =
6427                         btrfs_file_extent_disk_bytenr(leaf, fi);
6428                 exts[nr].disk_num_bytes =
6429                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6430                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6431                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6432                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6433                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6434                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6435                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6436                                                                            fi);
6437                 BUG_ON(exts[nr].offset > 0);
6438                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6439                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6440
6441                 cur_pos += exts[nr].num_bytes;
6442                 nr++;
6443
6444                 if (cur_pos + offset >= last_byte)
6445                         break;
6446
6447                 if (no_fragment) {
6448                         ret = 1;
6449                         goto out;
6450                 }
6451                 path->slots[0]++;
6452         }
6453
6454         BUG_ON(cur_pos + offset > last_byte);
6455         if (cur_pos + offset < last_byte) {
6456                 ret = -ENOENT;
6457                 goto out;
6458         }
6459         ret = 0;
6460 out:
6461         btrfs_free_path(path);
6462         if (ret) {
6463                 if (exts != *extents)
6464                         kfree(exts);
6465         } else {
6466                 *extents = exts;
6467                 *nr_extents = nr;
6468         }
6469         return ret;
6470 }
6471
6472 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6473                                         struct btrfs_root *root,
6474                                         struct btrfs_path *path,
6475                                         struct btrfs_key *extent_key,
6476                                         struct btrfs_key *leaf_key,
6477                                         struct btrfs_ref_path *ref_path,
6478                                         struct disk_extent *new_extents,
6479                                         int nr_extents)
6480 {
6481         struct extent_buffer *leaf;
6482         struct btrfs_file_extent_item *fi;
6483         struct inode *inode = NULL;
6484         struct btrfs_key key;
6485         u64 lock_start = 0;
6486         u64 lock_end = 0;
6487         u64 num_bytes;
6488         u64 ext_offset;
6489         u64 search_end = (u64)-1;
6490         u32 nritems;
6491         int nr_scaned = 0;
6492         int extent_locked = 0;
6493         int extent_type;
6494         int ret;
6495
6496         memcpy(&key, leaf_key, sizeof(key));
6497         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6498                 if (key.objectid < ref_path->owner_objectid ||
6499                     (key.objectid == ref_path->owner_objectid &&
6500                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6501                         key.objectid = ref_path->owner_objectid;
6502                         key.type = BTRFS_EXTENT_DATA_KEY;
6503                         key.offset = 0;
6504                 }
6505         }
6506
6507         while (1) {
6508                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6509                 if (ret < 0)
6510                         goto out;
6511
6512                 leaf = path->nodes[0];
6513                 nritems = btrfs_header_nritems(leaf);
6514 next:
6515                 if (extent_locked && ret > 0) {
6516                         /*
6517                          * the file extent item was modified by someone
6518                          * before the extent got locked.
6519                          */
6520                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6521                                       lock_end, GFP_NOFS);
6522                         extent_locked = 0;
6523                 }
6524
6525                 if (path->slots[0] >= nritems) {
6526                         if (++nr_scaned > 2)
6527                                 break;
6528
6529                         BUG_ON(extent_locked);
6530                         ret = btrfs_next_leaf(root, path);
6531                         if (ret < 0)
6532                                 goto out;
6533                         if (ret > 0)
6534                                 break;
6535                         leaf = path->nodes[0];
6536                         nritems = btrfs_header_nritems(leaf);
6537                 }
6538
6539                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6540
6541                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6542                         if ((key.objectid > ref_path->owner_objectid) ||
6543                             (key.objectid == ref_path->owner_objectid &&
6544                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6545                             key.offset >= search_end)
6546                                 break;
6547                 }
6548
6549                 if (inode && key.objectid != inode->i_ino) {
6550                         BUG_ON(extent_locked);
6551                         btrfs_release_path(root, path);
6552                         mutex_unlock(&inode->i_mutex);
6553                         iput(inode);
6554                         inode = NULL;
6555                         continue;
6556                 }
6557
6558                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6559                         path->slots[0]++;
6560                         ret = 1;
6561                         goto next;
6562                 }
6563                 fi = btrfs_item_ptr(leaf, path->slots[0],
6564                                     struct btrfs_file_extent_item);
6565                 extent_type = btrfs_file_extent_type(leaf, fi);
6566                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6567                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6568                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6569                      extent_key->objectid)) {
6570                         path->slots[0]++;
6571                         ret = 1;
6572                         goto next;
6573                 }
6574
6575                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6576                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6577
6578                 if (search_end == (u64)-1) {
6579                         search_end = key.offset - ext_offset +
6580                                 btrfs_file_extent_ram_bytes(leaf, fi);
6581                 }
6582
6583                 if (!extent_locked) {
6584                         lock_start = key.offset;
6585                         lock_end = lock_start + num_bytes - 1;
6586                 } else {
6587                         if (lock_start > key.offset ||
6588                             lock_end + 1 < key.offset + num_bytes) {
6589                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6590                                               lock_start, lock_end, GFP_NOFS);
6591                                 extent_locked = 0;
6592                         }
6593                 }
6594
6595                 if (!inode) {
6596                         btrfs_release_path(root, path);
6597
6598                         inode = btrfs_iget_locked(root->fs_info->sb,
6599                                                   key.objectid, root);
6600                         if (inode->i_state & I_NEW) {
6601                                 BTRFS_I(inode)->root = root;
6602                                 BTRFS_I(inode)->location.objectid =
6603                                         key.objectid;
6604                                 BTRFS_I(inode)->location.type =
6605                                         BTRFS_INODE_ITEM_KEY;
6606                                 BTRFS_I(inode)->location.offset = 0;
6607                                 btrfs_read_locked_inode(inode);
6608                                 unlock_new_inode(inode);
6609                         }
6610                         /*
6611                          * some code call btrfs_commit_transaction while
6612                          * holding the i_mutex, so we can't use mutex_lock
6613                          * here.
6614                          */
6615                         if (is_bad_inode(inode) ||
6616                             !mutex_trylock(&inode->i_mutex)) {
6617                                 iput(inode);
6618                                 inode = NULL;
6619                                 key.offset = (u64)-1;
6620                                 goto skip;
6621                         }
6622                 }
6623
6624                 if (!extent_locked) {
6625                         struct btrfs_ordered_extent *ordered;
6626
6627                         btrfs_release_path(root, path);
6628
6629                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6630                                     lock_end, GFP_NOFS);
6631                         ordered = btrfs_lookup_first_ordered_extent(inode,
6632                                                                     lock_end);
6633                         if (ordered &&
6634                             ordered->file_offset <= lock_end &&
6635                             ordered->file_offset + ordered->len > lock_start) {
6636                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6637                                               lock_start, lock_end, GFP_NOFS);
6638                                 btrfs_start_ordered_extent(inode, ordered, 1);
6639                                 btrfs_put_ordered_extent(ordered);
6640                                 key.offset += num_bytes;
6641                                 goto skip;
6642                         }
6643                         if (ordered)
6644                                 btrfs_put_ordered_extent(ordered);
6645
6646                         extent_locked = 1;
6647                         continue;
6648                 }
6649
6650                 if (nr_extents == 1) {
6651                         /* update extent pointer in place */
6652                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6653                                                 new_extents[0].disk_bytenr);
6654                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6655                                                 new_extents[0].disk_num_bytes);
6656                         btrfs_mark_buffer_dirty(leaf);
6657
6658                         btrfs_drop_extent_cache(inode, key.offset,
6659                                                 key.offset + num_bytes - 1, 0);
6660
6661                         ret = btrfs_inc_extent_ref(trans, root,
6662                                                 new_extents[0].disk_bytenr,
6663                                                 new_extents[0].disk_num_bytes,
6664                                                 leaf->start,
6665                                                 root->root_key.objectid,
6666                                                 trans->transid,
6667                                                 key.objectid);
6668                         BUG_ON(ret);
6669
6670                         ret = btrfs_free_extent(trans, root,
6671                                                 extent_key->objectid,
6672                                                 extent_key->offset,
6673                                                 leaf->start,
6674                                                 btrfs_header_owner(leaf),
6675                                                 btrfs_header_generation(leaf),
6676                                                 key.objectid, 0);
6677                         BUG_ON(ret);
6678
6679                         btrfs_release_path(root, path);
6680                         key.offset += num_bytes;
6681                 } else {
6682                         BUG_ON(1);
6683 #if 0
6684                         u64 alloc_hint;
6685                         u64 extent_len;
6686                         int i;
6687                         /*
6688                          * drop old extent pointer at first, then insert the
6689                          * new pointers one bye one
6690                          */
6691                         btrfs_release_path(root, path);
6692                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
6693                                                  key.offset + num_bytes,
6694                                                  key.offset, &alloc_hint);
6695                         BUG_ON(ret);
6696
6697                         for (i = 0; i < nr_extents; i++) {
6698                                 if (ext_offset >= new_extents[i].num_bytes) {
6699                                         ext_offset -= new_extents[i].num_bytes;
6700                                         continue;
6701                                 }
6702                                 extent_len = min(new_extents[i].num_bytes -
6703                                                  ext_offset, num_bytes);
6704
6705                                 ret = btrfs_insert_empty_item(trans, root,
6706                                                               path, &key,
6707                                                               sizeof(*fi));
6708                                 BUG_ON(ret);
6709
6710                                 leaf = path->nodes[0];
6711                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6712                                                 struct btrfs_file_extent_item);
6713                                 btrfs_set_file_extent_generation(leaf, fi,
6714                                                         trans->transid);
6715                                 btrfs_set_file_extent_type(leaf, fi,
6716                                                         BTRFS_FILE_EXTENT_REG);
6717                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6718                                                 new_extents[i].disk_bytenr);
6719                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6720                                                 new_extents[i].disk_num_bytes);
6721                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6722                                                 new_extents[i].ram_bytes);
6723
6724                                 btrfs_set_file_extent_compression(leaf, fi,
6725                                                 new_extents[i].compression);
6726                                 btrfs_set_file_extent_encryption(leaf, fi,
6727                                                 new_extents[i].encryption);
6728                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6729                                                 new_extents[i].other_encoding);
6730
6731                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6732                                                         extent_len);
6733                                 ext_offset += new_extents[i].offset;
6734                                 btrfs_set_file_extent_offset(leaf, fi,
6735                                                         ext_offset);
6736                                 btrfs_mark_buffer_dirty(leaf);
6737
6738                                 btrfs_drop_extent_cache(inode, key.offset,
6739                                                 key.offset + extent_len - 1, 0);
6740
6741                                 ret = btrfs_inc_extent_ref(trans, root,
6742                                                 new_extents[i].disk_bytenr,
6743                                                 new_extents[i].disk_num_bytes,
6744                                                 leaf->start,
6745                                                 root->root_key.objectid,
6746                                                 trans->transid, key.objectid);
6747                                 BUG_ON(ret);
6748                                 btrfs_release_path(root, path);
6749
6750                                 inode_add_bytes(inode, extent_len);
6751
6752                                 ext_offset = 0;
6753                                 num_bytes -= extent_len;
6754                                 key.offset += extent_len;
6755
6756                                 if (num_bytes == 0)
6757                                         break;
6758                         }
6759                         BUG_ON(i >= nr_extents);
6760 #endif
6761                 }
6762
6763                 if (extent_locked) {
6764                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6765                                       lock_end, GFP_NOFS);
6766                         extent_locked = 0;
6767                 }
6768 skip:
6769                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6770                     key.offset >= search_end)
6771                         break;
6772
6773                 cond_resched();
6774         }
6775         ret = 0;
6776 out:
6777         btrfs_release_path(root, path);
6778         if (inode) {
6779                 mutex_unlock(&inode->i_mutex);
6780                 if (extent_locked) {
6781                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6782                                       lock_end, GFP_NOFS);
6783                 }
6784                 iput(inode);
6785         }
6786         return ret;
6787 }
6788
6789 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6790                                struct btrfs_root *root,
6791                                struct extent_buffer *buf, u64 orig_start)
6792 {
6793         int level;
6794         int ret;
6795
6796         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6797         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6798
6799         level = btrfs_header_level(buf);
6800         if (level == 0) {
6801                 struct btrfs_leaf_ref *ref;
6802                 struct btrfs_leaf_ref *orig_ref;
6803
6804                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6805                 if (!orig_ref)
6806                         return -ENOENT;
6807
6808                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6809                 if (!ref) {
6810                         btrfs_free_leaf_ref(root, orig_ref);
6811                         return -ENOMEM;
6812                 }
6813
6814                 ref->nritems = orig_ref->nritems;
6815                 memcpy(ref->extents, orig_ref->extents,
6816                         sizeof(ref->extents[0]) * ref->nritems);
6817
6818                 btrfs_free_leaf_ref(root, orig_ref);
6819
6820                 ref->root_gen = trans->transid;
6821                 ref->bytenr = buf->start;
6822                 ref->owner = btrfs_header_owner(buf);
6823                 ref->generation = btrfs_header_generation(buf);
6824
6825                 ret = btrfs_add_leaf_ref(root, ref, 0);
6826                 WARN_ON(ret);
6827                 btrfs_free_leaf_ref(root, ref);
6828         }
6829         return 0;
6830 }
6831
6832 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6833                                         struct extent_buffer *leaf,
6834                                         struct btrfs_block_group_cache *group,
6835                                         struct btrfs_root *target_root)
6836 {
6837         struct btrfs_key key;
6838         struct inode *inode = NULL;
6839         struct btrfs_file_extent_item *fi;
6840         struct extent_state *cached_state = NULL;
6841         u64 num_bytes;
6842         u64 skip_objectid = 0;
6843         u32 nritems;
6844         u32 i;
6845
6846         nritems = btrfs_header_nritems(leaf);
6847         for (i = 0; i < nritems; i++) {
6848                 btrfs_item_key_to_cpu(leaf, &key, i);
6849                 if (key.objectid == skip_objectid ||
6850                     key.type != BTRFS_EXTENT_DATA_KEY)
6851                         continue;
6852                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6853                 if (btrfs_file_extent_type(leaf, fi) ==
6854                     BTRFS_FILE_EXTENT_INLINE)
6855                         continue;
6856                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6857                         continue;
6858                 if (!inode || inode->i_ino != key.objectid) {
6859                         iput(inode);
6860                         inode = btrfs_ilookup(target_root->fs_info->sb,
6861                                               key.objectid, target_root, 1);
6862                 }
6863                 if (!inode) {
6864                         skip_objectid = key.objectid;
6865                         continue;
6866                 }
6867                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6868
6869                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
6870                                  key.offset + num_bytes - 1, 0, &cached_state,
6871                                  GFP_NOFS);
6872                 btrfs_drop_extent_cache(inode, key.offset,
6873                                         key.offset + num_bytes - 1, 1);
6874                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
6875                                      key.offset + num_bytes - 1, &cached_state,
6876                                      GFP_NOFS);
6877                 cond_resched();
6878         }
6879         iput(inode);
6880         return 0;
6881 }
6882
6883 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6884                                         struct btrfs_root *root,
6885                                         struct extent_buffer *leaf,
6886                                         struct btrfs_block_group_cache *group,
6887                                         struct inode *reloc_inode)
6888 {
6889         struct btrfs_key key;
6890         struct btrfs_key extent_key;
6891         struct btrfs_file_extent_item *fi;
6892         struct btrfs_leaf_ref *ref;
6893         struct disk_extent *new_extent;
6894         u64 bytenr;
6895         u64 num_bytes;
6896         u32 nritems;
6897         u32 i;
6898         int ext_index;
6899         int nr_extent;
6900         int ret;
6901
6902         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6903         BUG_ON(!new_extent);
6904
6905         ref = btrfs_lookup_leaf_ref(root, leaf->start);
6906         BUG_ON(!ref);
6907
6908         ext_index = -1;
6909         nritems = btrfs_header_nritems(leaf);
6910         for (i = 0; i < nritems; i++) {
6911                 btrfs_item_key_to_cpu(leaf, &key, i);
6912                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6913                         continue;
6914                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6915                 if (btrfs_file_extent_type(leaf, fi) ==
6916                     BTRFS_FILE_EXTENT_INLINE)
6917                         continue;
6918                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6919                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6920                 if (bytenr == 0)
6921                         continue;
6922
6923                 ext_index++;
6924                 if (bytenr >= group->key.objectid + group->key.offset ||
6925                     bytenr + num_bytes <= group->key.objectid)
6926                         continue;
6927
6928                 extent_key.objectid = bytenr;
6929                 extent_key.offset = num_bytes;
6930                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6931                 nr_extent = 1;
6932                 ret = get_new_locations(reloc_inode, &extent_key,
6933                                         group->key.objectid, 1,
6934                                         &new_extent, &nr_extent);
6935                 if (ret > 0)
6936                         continue;
6937                 BUG_ON(ret < 0);
6938
6939                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6940                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6941                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6942                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6943
6944                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6945                                                 new_extent->disk_bytenr);
6946                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6947                                                 new_extent->disk_num_bytes);
6948                 btrfs_mark_buffer_dirty(leaf);
6949
6950                 ret = btrfs_inc_extent_ref(trans, root,
6951                                         new_extent->disk_bytenr,
6952                                         new_extent->disk_num_bytes,
6953                                         leaf->start,
6954                                         root->root_key.objectid,
6955                                         trans->transid, key.objectid);
6956                 BUG_ON(ret);
6957
6958                 ret = btrfs_free_extent(trans, root,
6959                                         bytenr, num_bytes, leaf->start,
6960                                         btrfs_header_owner(leaf),
6961                                         btrfs_header_generation(leaf),
6962                                         key.objectid, 0);
6963                 BUG_ON(ret);
6964                 cond_resched();
6965         }
6966         kfree(new_extent);
6967         BUG_ON(ext_index + 1 != ref->nritems);
6968         btrfs_free_leaf_ref(root, ref);
6969         return 0;
6970 }
6971
6972 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6973                           struct btrfs_root *root)
6974 {
6975         struct btrfs_root *reloc_root;
6976         int ret;
6977
6978         if (root->reloc_root) {
6979                 reloc_root = root->reloc_root;
6980                 root->reloc_root = NULL;
6981                 list_add(&reloc_root->dead_list,
6982                          &root->fs_info->dead_reloc_roots);
6983
6984                 btrfs_set_root_bytenr(&reloc_root->root_item,
6985                                       reloc_root->node->start);
6986                 btrfs_set_root_level(&root->root_item,
6987                                      btrfs_header_level(reloc_root->node));
6988                 memset(&reloc_root->root_item.drop_progress, 0,
6989                         sizeof(struct btrfs_disk_key));
6990                 reloc_root->root_item.drop_level = 0;
6991
6992                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6993                                         &reloc_root->root_key,
6994                                         &reloc_root->root_item);
6995                 BUG_ON(ret);
6996         }
6997         return 0;
6998 }
6999
7000 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7001 {
7002         struct btrfs_trans_handle *trans;
7003         struct btrfs_root *reloc_root;
7004         struct btrfs_root *prev_root = NULL;
7005         struct list_head dead_roots;
7006         int ret;
7007         unsigned long nr;
7008
7009         INIT_LIST_HEAD(&dead_roots);
7010         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7011
7012         while (!list_empty(&dead_roots)) {
7013                 reloc_root = list_entry(dead_roots.prev,
7014                                         struct btrfs_root, dead_list);
7015                 list_del_init(&reloc_root->dead_list);
7016
7017                 BUG_ON(reloc_root->commit_root != NULL);
7018                 while (1) {
7019                         trans = btrfs_join_transaction(root, 1);
7020                         BUG_ON(!trans);
7021
7022                         mutex_lock(&root->fs_info->drop_mutex);
7023                         ret = btrfs_drop_snapshot(trans, reloc_root);
7024                         if (ret != -EAGAIN)
7025                                 break;
7026                         mutex_unlock(&root->fs_info->drop_mutex);
7027
7028                         nr = trans->blocks_used;
7029                         ret = btrfs_end_transaction(trans, root);
7030                         BUG_ON(ret);
7031                         btrfs_btree_balance_dirty(root, nr);
7032                 }
7033
7034                 free_extent_buffer(reloc_root->node);
7035
7036                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7037                                      &reloc_root->root_key);
7038                 BUG_ON(ret);
7039                 mutex_unlock(&root->fs_info->drop_mutex);
7040
7041                 nr = trans->blocks_used;
7042                 ret = btrfs_end_transaction(trans, root);
7043                 BUG_ON(ret);
7044                 btrfs_btree_balance_dirty(root, nr);
7045
7046                 kfree(prev_root);
7047                 prev_root = reloc_root;
7048         }
7049         if (prev_root) {
7050                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7051                 kfree(prev_root);
7052         }
7053         return 0;
7054 }
7055
7056 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7057 {
7058         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7059         return 0;
7060 }
7061
7062 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7063 {
7064         struct btrfs_root *reloc_root;
7065         struct btrfs_trans_handle *trans;
7066         struct btrfs_key location;
7067         int found;
7068         int ret;
7069
7070         mutex_lock(&root->fs_info->tree_reloc_mutex);
7071         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7072         BUG_ON(ret);
7073         found = !list_empty(&root->fs_info->dead_reloc_roots);
7074         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7075
7076         if (found) {
7077                 trans = btrfs_start_transaction(root, 1);
7078                 BUG_ON(!trans);
7079                 ret = btrfs_commit_transaction(trans, root);
7080                 BUG_ON(ret);
7081         }
7082
7083         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7084         location.offset = (u64)-1;
7085         location.type = BTRFS_ROOT_ITEM_KEY;
7086
7087         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7088         BUG_ON(!reloc_root);
7089         btrfs_orphan_cleanup(reloc_root);
7090         return 0;
7091 }
7092
7093 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7094                                     struct btrfs_root *root)
7095 {
7096         struct btrfs_root *reloc_root;
7097         struct extent_buffer *eb;
7098         struct btrfs_root_item *root_item;
7099         struct btrfs_key root_key;
7100         int ret;
7101
7102         BUG_ON(!root->ref_cows);
7103         if (root->reloc_root)
7104                 return 0;
7105
7106         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7107         BUG_ON(!root_item);
7108
7109         ret = btrfs_copy_root(trans, root, root->commit_root,
7110                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7111         BUG_ON(ret);
7112
7113         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7114         root_key.offset = root->root_key.objectid;
7115         root_key.type = BTRFS_ROOT_ITEM_KEY;
7116
7117         memcpy(root_item, &root->root_item, sizeof(root_item));
7118         btrfs_set_root_refs(root_item, 0);
7119         btrfs_set_root_bytenr(root_item, eb->start);
7120         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7121         btrfs_set_root_generation(root_item, trans->transid);
7122
7123         btrfs_tree_unlock(eb);
7124         free_extent_buffer(eb);
7125
7126         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7127                                 &root_key, root_item);
7128         BUG_ON(ret);
7129         kfree(root_item);
7130
7131         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7132                                                  &root_key);
7133         BUG_ON(!reloc_root);
7134         reloc_root->last_trans = trans->transid;
7135         reloc_root->commit_root = NULL;
7136         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7137
7138         root->reloc_root = reloc_root;
7139         return 0;
7140 }
7141
7142 /*
7143  * Core function of space balance.
7144  *
7145  * The idea is using reloc trees to relocate tree blocks in reference
7146  * counted roots. There is one reloc tree for each subvol, and all
7147  * reloc trees share same root key objectid. Reloc trees are snapshots
7148  * of the latest committed roots of subvols (root->commit_root).
7149  *
7150  * To relocate a tree block referenced by a subvol, there are two steps.
7151  * COW the block through subvol's reloc tree, then update block pointer
7152  * in the subvol to point to the new block. Since all reloc trees share
7153  * same root key objectid, doing special handing for tree blocks owned
7154  * by them is easy. Once a tree block has been COWed in one reloc tree,
7155  * we can use the resulting new block directly when the same block is
7156  * required to COW again through other reloc trees. By this way, relocated
7157  * tree blocks are shared between reloc trees, so they are also shared
7158  * between subvols.
7159  */
7160 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7161                                       struct btrfs_root *root,
7162                                       struct btrfs_path *path,
7163                                       struct btrfs_key *first_key,
7164                                       struct btrfs_ref_path *ref_path,
7165                                       struct btrfs_block_group_cache *group,
7166                                       struct inode *reloc_inode)
7167 {
7168         struct btrfs_root *reloc_root;
7169         struct extent_buffer *eb = NULL;
7170         struct btrfs_key *keys;
7171         u64 *nodes;
7172         int level;
7173         int shared_level;
7174         int lowest_level = 0;
7175         int ret;
7176
7177         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7178                 lowest_level = ref_path->owner_objectid;
7179
7180         if (!root->ref_cows) {
7181                 path->lowest_level = lowest_level;
7182                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7183                 BUG_ON(ret < 0);
7184                 path->lowest_level = 0;
7185                 btrfs_release_path(root, path);
7186                 return 0;
7187         }
7188
7189         mutex_lock(&root->fs_info->tree_reloc_mutex);
7190         ret = init_reloc_tree(trans, root);
7191         BUG_ON(ret);
7192         reloc_root = root->reloc_root;
7193
7194         shared_level = ref_path->shared_level;
7195         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7196
7197         keys = ref_path->node_keys;
7198         nodes = ref_path->new_nodes;
7199         memset(&keys[shared_level + 1], 0,
7200                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7201         memset(&nodes[shared_level + 1], 0,
7202                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7203
7204         if (nodes[lowest_level] == 0) {
7205                 path->lowest_level = lowest_level;
7206                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7207                                         0, 1);
7208                 BUG_ON(ret);
7209                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7210                         eb = path->nodes[level];
7211                         if (!eb || eb == reloc_root->node)
7212                                 break;
7213                         nodes[level] = eb->start;
7214                         if (level == 0)
7215                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7216                         else
7217                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7218                 }
7219                 if (nodes[0] &&
7220                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7221                         eb = path->nodes[0];
7222                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7223                                                       group, reloc_inode);
7224                         BUG_ON(ret);
7225                 }
7226                 btrfs_release_path(reloc_root, path);
7227         } else {
7228                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7229                                        lowest_level);
7230                 BUG_ON(ret);
7231         }
7232
7233         /*
7234          * replace tree blocks in the fs tree with tree blocks in
7235          * the reloc tree.
7236          */
7237         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7238         BUG_ON(ret < 0);
7239
7240         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7241                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7242                                         0, 0);
7243                 BUG_ON(ret);
7244                 extent_buffer_get(path->nodes[0]);
7245                 eb = path->nodes[0];
7246                 btrfs_release_path(reloc_root, path);
7247                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7248                 BUG_ON(ret);
7249                 free_extent_buffer(eb);
7250         }
7251
7252         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7253         path->lowest_level = 0;
7254         return 0;
7255 }
7256
7257 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7258                                         struct btrfs_root *root,
7259                                         struct btrfs_path *path,
7260                                         struct btrfs_key *first_key,
7261                                         struct btrfs_ref_path *ref_path)
7262 {
7263         int ret;
7264
7265         ret = relocate_one_path(trans, root, path, first_key,
7266                                 ref_path, NULL, NULL);
7267         BUG_ON(ret);
7268
7269         return 0;
7270 }
7271
7272 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7273                                     struct btrfs_root *extent_root,
7274                                     struct btrfs_path *path,
7275                                     struct btrfs_key *extent_key)
7276 {
7277         int ret;
7278
7279         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7280         if (ret)
7281                 goto out;
7282         ret = btrfs_del_item(trans, extent_root, path);
7283 out:
7284         btrfs_release_path(extent_root, path);
7285         return ret;
7286 }
7287
7288 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7289                                                 struct btrfs_ref_path *ref_path)
7290 {
7291         struct btrfs_key root_key;
7292
7293         root_key.objectid = ref_path->root_objectid;
7294         root_key.type = BTRFS_ROOT_ITEM_KEY;
7295         if (is_cowonly_root(ref_path->root_objectid))
7296                 root_key.offset = 0;
7297         else
7298                 root_key.offset = (u64)-1;
7299
7300         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7301 }
7302
7303 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7304                                         struct btrfs_path *path,
7305                                         struct btrfs_key *extent_key,
7306                                         struct btrfs_block_group_cache *group,
7307                                         struct inode *reloc_inode, int pass)
7308 {
7309         struct btrfs_trans_handle *trans;
7310         struct btrfs_root *found_root;
7311         struct btrfs_ref_path *ref_path = NULL;
7312         struct disk_extent *new_extents = NULL;
7313         int nr_extents = 0;
7314         int loops;
7315         int ret;
7316         int level;
7317         struct btrfs_key first_key;
7318         u64 prev_block = 0;
7319
7320
7321         trans = btrfs_start_transaction(extent_root, 1);
7322         BUG_ON(!trans);
7323
7324         if (extent_key->objectid == 0) {
7325                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7326                 goto out;
7327         }
7328
7329         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7330         if (!ref_path) {
7331                 ret = -ENOMEM;
7332                 goto out;
7333         }
7334
7335         for (loops = 0; ; loops++) {
7336                 if (loops == 0) {
7337                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7338                                                    extent_key->objectid);
7339                 } else {
7340                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7341                 }
7342                 if (ret < 0)
7343                         goto out;
7344                 if (ret > 0)
7345                         break;
7346
7347                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7348                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7349                         continue;
7350
7351                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7352                 BUG_ON(!found_root);
7353                 /*
7354                  * for reference counted tree, only process reference paths
7355                  * rooted at the latest committed root.
7356                  */
7357                 if (found_root->ref_cows &&
7358                     ref_path->root_generation != found_root->root_key.offset)
7359                         continue;
7360
7361                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7362                         if (pass == 0) {
7363                                 /*
7364                                  * copy data extents to new locations
7365                                  */
7366                                 u64 group_start = group->key.objectid;
7367                                 ret = relocate_data_extent(reloc_inode,
7368                                                            extent_key,
7369                                                            group_start);
7370                                 if (ret < 0)
7371                                         goto out;
7372                                 break;
7373                         }
7374                         level = 0;
7375                 } else {
7376                         level = ref_path->owner_objectid;
7377                 }
7378
7379                 if (prev_block != ref_path->nodes[level]) {
7380                         struct extent_buffer *eb;
7381                         u64 block_start = ref_path->nodes[level];
7382                         u64 block_size = btrfs_level_size(found_root, level);
7383
7384                         eb = read_tree_block(found_root, block_start,
7385                                              block_size, 0);
7386                         btrfs_tree_lock(eb);
7387                         BUG_ON(level != btrfs_header_level(eb));
7388
7389                         if (level == 0)
7390                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7391                         else
7392                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7393
7394                         btrfs_tree_unlock(eb);
7395                         free_extent_buffer(eb);
7396                         prev_block = block_start;
7397                 }
7398
7399                 mutex_lock(&extent_root->fs_info->trans_mutex);
7400                 btrfs_record_root_in_trans(found_root);
7401                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7402                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7403                         /*
7404                          * try to update data extent references while
7405                          * keeping metadata shared between snapshots.
7406                          */
7407                         if (pass == 1) {
7408                                 ret = relocate_one_path(trans, found_root,
7409                                                 path, &first_key, ref_path,
7410                                                 group, reloc_inode);
7411                                 if (ret < 0)
7412                                         goto out;
7413                                 continue;
7414                         }
7415                         /*
7416                          * use fallback method to process the remaining
7417                          * references.
7418                          */
7419                         if (!new_extents) {
7420                                 u64 group_start = group->key.objectid;
7421                                 new_extents = kmalloc(sizeof(*new_extents),
7422                                                       GFP_NOFS);
7423                                 nr_extents = 1;
7424                                 ret = get_new_locations(reloc_inode,
7425                                                         extent_key,
7426                                                         group_start, 1,
7427                                                         &new_extents,
7428                                                         &nr_extents);
7429                                 if (ret)
7430                                         goto out;
7431                         }
7432                         ret = replace_one_extent(trans, found_root,
7433                                                 path, extent_key,
7434                                                 &first_key, ref_path,
7435                                                 new_extents, nr_extents);
7436                 } else {
7437                         ret = relocate_tree_block(trans, found_root, path,
7438                                                   &first_key, ref_path);
7439                 }
7440                 if (ret < 0)
7441                         goto out;
7442         }
7443         ret = 0;
7444 out:
7445         btrfs_end_transaction(trans, extent_root);
7446         kfree(new_extents);
7447         kfree(ref_path);
7448         return ret;
7449 }
7450 #endif
7451
7452 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7453 {
7454         u64 num_devices;
7455         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7456                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7457
7458         num_devices = root->fs_info->fs_devices->rw_devices;
7459         if (num_devices == 1) {
7460                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7461                 stripped = flags & ~stripped;
7462
7463                 /* turn raid0 into single device chunks */
7464                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7465                         return stripped;
7466
7467                 /* turn mirroring into duplication */
7468                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7469                              BTRFS_BLOCK_GROUP_RAID10))
7470                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7471                 return flags;
7472         } else {
7473                 /* they already had raid on here, just return */
7474                 if (flags & stripped)
7475                         return flags;
7476
7477                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7478                 stripped = flags & ~stripped;
7479
7480                 /* switch duplicated blocks with raid1 */
7481                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7482                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7483
7484                 /* turn single device chunks into raid0 */
7485                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7486         }
7487         return flags;
7488 }
7489
7490 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
7491 {
7492         struct btrfs_space_info *sinfo = cache->space_info;
7493         u64 num_bytes;
7494         int ret = -ENOSPC;
7495
7496         if (cache->ro)
7497                 return 0;
7498
7499         spin_lock(&sinfo->lock);
7500         spin_lock(&cache->lock);
7501         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7502                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7503
7504         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
7505             sinfo->bytes_may_use + sinfo->bytes_readonly +
7506             cache->reserved_pinned + num_bytes < sinfo->total_bytes) {
7507                 sinfo->bytes_readonly += num_bytes;
7508                 sinfo->bytes_reserved += cache->reserved_pinned;
7509                 cache->reserved_pinned = 0;
7510                 cache->ro = 1;
7511                 ret = 0;
7512         }
7513         spin_unlock(&cache->lock);
7514         spin_unlock(&sinfo->lock);
7515         return ret;
7516 }
7517
7518 int btrfs_set_block_group_ro(struct btrfs_root *root,
7519                              struct btrfs_block_group_cache *cache)
7520
7521 {
7522         struct btrfs_trans_handle *trans;
7523         u64 alloc_flags;
7524         int ret;
7525
7526         BUG_ON(cache->ro);
7527
7528         trans = btrfs_join_transaction(root, 1);
7529         BUG_ON(IS_ERR(trans));
7530
7531         alloc_flags = update_block_group_flags(root, cache->flags);
7532         if (alloc_flags != cache->flags)
7533                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7534
7535         ret = set_block_group_ro(cache);
7536         if (!ret)
7537                 goto out;
7538         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
7539         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
7540         if (ret < 0)
7541                 goto out;
7542         ret = set_block_group_ro(cache);
7543 out:
7544         btrfs_end_transaction(trans, root);
7545         return ret;
7546 }
7547
7548 int btrfs_set_block_group_rw(struct btrfs_root *root,
7549                               struct btrfs_block_group_cache *cache)
7550 {
7551         struct btrfs_space_info *sinfo = cache->space_info;
7552         u64 num_bytes;
7553
7554         BUG_ON(!cache->ro);
7555
7556         spin_lock(&sinfo->lock);
7557         spin_lock(&cache->lock);
7558         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
7559                     cache->bytes_super - btrfs_block_group_used(&cache->item);
7560         sinfo->bytes_readonly -= num_bytes;
7561         cache->ro = 0;
7562         spin_unlock(&cache->lock);
7563         spin_unlock(&sinfo->lock);
7564         return 0;
7565 }
7566
7567 /*
7568  * checks to see if its even possible to relocate this block group.
7569  *
7570  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7571  * ok to go ahead and try.
7572  */
7573 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7574 {
7575         struct btrfs_block_group_cache *block_group;
7576         struct btrfs_space_info *space_info;
7577         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7578         struct btrfs_device *device;
7579         int full = 0;
7580         int ret = 0;
7581
7582         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7583
7584         /* odd, couldn't find the block group, leave it alone */
7585         if (!block_group)
7586                 return -1;
7587
7588         /* no bytes used, we're good */
7589         if (!btrfs_block_group_used(&block_group->item))
7590                 goto out;
7591
7592         space_info = block_group->space_info;
7593         spin_lock(&space_info->lock);
7594
7595         full = space_info->full;
7596
7597         /*
7598          * if this is the last block group we have in this space, we can't
7599          * relocate it unless we're able to allocate a new chunk below.
7600          *
7601          * Otherwise, we need to make sure we have room in the space to handle
7602          * all of the extents from this block group.  If we can, we're good
7603          */
7604         if ((space_info->total_bytes != block_group->key.offset) &&
7605            (space_info->bytes_used + space_info->bytes_reserved +
7606             space_info->bytes_pinned + space_info->bytes_readonly +
7607             btrfs_block_group_used(&block_group->item) <
7608             space_info->total_bytes)) {
7609                 spin_unlock(&space_info->lock);
7610                 goto out;
7611         }
7612         spin_unlock(&space_info->lock);
7613
7614         /*
7615          * ok we don't have enough space, but maybe we have free space on our
7616          * devices to allocate new chunks for relocation, so loop through our
7617          * alloc devices and guess if we have enough space.  However, if we
7618          * were marked as full, then we know there aren't enough chunks, and we
7619          * can just return.
7620          */
7621         ret = -1;
7622         if (full)
7623                 goto out;
7624
7625         mutex_lock(&root->fs_info->chunk_mutex);
7626         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7627                 u64 min_free = btrfs_block_group_used(&block_group->item);
7628                 u64 dev_offset, max_avail;
7629
7630                 /*
7631                  * check to make sure we can actually find a chunk with enough
7632                  * space to fit our block group in.
7633                  */
7634                 if (device->total_bytes > device->bytes_used + min_free) {
7635                         ret = find_free_dev_extent(NULL, device, min_free,
7636                                                    &dev_offset, &max_avail);
7637                         if (!ret)
7638                                 break;
7639                         ret = -1;
7640                 }
7641         }
7642         mutex_unlock(&root->fs_info->chunk_mutex);
7643 out:
7644         btrfs_put_block_group(block_group);
7645         return ret;
7646 }
7647
7648 static int find_first_block_group(struct btrfs_root *root,
7649                 struct btrfs_path *path, struct btrfs_key *key)
7650 {
7651         int ret = 0;
7652         struct btrfs_key found_key;
7653         struct extent_buffer *leaf;
7654         int slot;
7655
7656         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7657         if (ret < 0)
7658                 goto out;
7659
7660         while (1) {
7661                 slot = path->slots[0];
7662                 leaf = path->nodes[0];
7663                 if (slot >= btrfs_header_nritems(leaf)) {
7664                         ret = btrfs_next_leaf(root, path);
7665                         if (ret == 0)
7666                                 continue;
7667                         if (ret < 0)
7668                                 goto out;
7669                         break;
7670                 }
7671                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7672
7673                 if (found_key.objectid >= key->objectid &&
7674                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7675                         ret = 0;
7676                         goto out;
7677                 }
7678                 path->slots[0]++;
7679         }
7680 out:
7681         return ret;
7682 }
7683
7684 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7685 {
7686         struct btrfs_block_group_cache *block_group;
7687         struct btrfs_space_info *space_info;
7688         struct btrfs_caching_control *caching_ctl;
7689         struct rb_node *n;
7690
7691         down_write(&info->extent_commit_sem);
7692         while (!list_empty(&info->caching_block_groups)) {
7693                 caching_ctl = list_entry(info->caching_block_groups.next,
7694                                          struct btrfs_caching_control, list);
7695                 list_del(&caching_ctl->list);
7696                 put_caching_control(caching_ctl);
7697         }
7698         up_write(&info->extent_commit_sem);
7699
7700         spin_lock(&info->block_group_cache_lock);
7701         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7702                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7703                                        cache_node);
7704                 rb_erase(&block_group->cache_node,
7705                          &info->block_group_cache_tree);
7706                 spin_unlock(&info->block_group_cache_lock);
7707
7708                 down_write(&block_group->space_info->groups_sem);
7709                 list_del(&block_group->list);
7710                 up_write(&block_group->space_info->groups_sem);
7711
7712                 if (block_group->cached == BTRFS_CACHE_STARTED)
7713                         wait_block_group_cache_done(block_group);
7714
7715                 btrfs_remove_free_space_cache(block_group);
7716                 btrfs_put_block_group(block_group);
7717
7718                 spin_lock(&info->block_group_cache_lock);
7719         }
7720         spin_unlock(&info->block_group_cache_lock);
7721
7722         /* now that all the block groups are freed, go through and
7723          * free all the space_info structs.  This is only called during
7724          * the final stages of unmount, and so we know nobody is
7725          * using them.  We call synchronize_rcu() once before we start,
7726          * just to be on the safe side.
7727          */
7728         synchronize_rcu();
7729
7730         while(!list_empty(&info->space_info)) {
7731                 space_info = list_entry(info->space_info.next,
7732                                         struct btrfs_space_info,
7733                                         list);
7734                 if (space_info->bytes_pinned > 0 ||
7735                     space_info->bytes_reserved > 0) {
7736                         WARN_ON(1);
7737                         dump_space_info(space_info, 0, 0);
7738                 }
7739                 list_del(&space_info->list);
7740                 kfree(space_info);
7741         }
7742         return 0;
7743 }
7744
7745 static void __link_block_group(struct btrfs_space_info *space_info,
7746                                struct btrfs_block_group_cache *cache)
7747 {
7748         int index = get_block_group_index(cache);
7749
7750         down_write(&space_info->groups_sem);
7751         list_add_tail(&cache->list, &space_info->block_groups[index]);
7752         up_write(&space_info->groups_sem);
7753 }
7754
7755 int btrfs_read_block_groups(struct btrfs_root *root)
7756 {
7757         struct btrfs_path *path;
7758         int ret;
7759         struct btrfs_block_group_cache *cache;
7760         struct btrfs_fs_info *info = root->fs_info;
7761         struct btrfs_space_info *space_info;
7762         struct btrfs_key key;
7763         struct btrfs_key found_key;
7764         struct extent_buffer *leaf;
7765
7766         root = info->extent_root;
7767         key.objectid = 0;
7768         key.offset = 0;
7769         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7770         path = btrfs_alloc_path();
7771         if (!path)
7772                 return -ENOMEM;
7773
7774         while (1) {
7775                 ret = find_first_block_group(root, path, &key);
7776                 if (ret > 0)
7777                         break;
7778                 if (ret != 0)
7779                         goto error;
7780
7781                 leaf = path->nodes[0];
7782                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7783                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7784                 if (!cache) {
7785                         ret = -ENOMEM;
7786                         goto error;
7787                 }
7788
7789                 atomic_set(&cache->count, 1);
7790                 spin_lock_init(&cache->lock);
7791                 spin_lock_init(&cache->tree_lock);
7792                 cache->fs_info = info;
7793                 INIT_LIST_HEAD(&cache->list);
7794                 INIT_LIST_HEAD(&cache->cluster_list);
7795
7796                 /*
7797                  * we only want to have 32k of ram per block group for keeping
7798                  * track of free space, and if we pass 1/2 of that we want to
7799                  * start converting things over to using bitmaps
7800                  */
7801                 cache->extents_thresh = ((1024 * 32) / 2) /
7802                         sizeof(struct btrfs_free_space);
7803
7804                 read_extent_buffer(leaf, &cache->item,
7805                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7806                                    sizeof(cache->item));
7807                 memcpy(&cache->key, &found_key, sizeof(found_key));
7808
7809                 key.objectid = found_key.objectid + found_key.offset;
7810                 btrfs_release_path(root, path);
7811                 cache->flags = btrfs_block_group_flags(&cache->item);
7812                 cache->sectorsize = root->sectorsize;
7813
7814                 /*
7815                  * check for two cases, either we are full, and therefore
7816                  * don't need to bother with the caching work since we won't
7817                  * find any space, or we are empty, and we can just add all
7818                  * the space in and be done with it.  This saves us _alot_ of
7819                  * time, particularly in the full case.
7820                  */
7821                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7822                         exclude_super_stripes(root, cache);
7823                         cache->last_byte_to_unpin = (u64)-1;
7824                         cache->cached = BTRFS_CACHE_FINISHED;
7825                         free_excluded_extents(root, cache);
7826                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7827                         exclude_super_stripes(root, cache);
7828                         cache->last_byte_to_unpin = (u64)-1;
7829                         cache->cached = BTRFS_CACHE_FINISHED;
7830                         add_new_free_space(cache, root->fs_info,
7831                                            found_key.objectid,
7832                                            found_key.objectid +
7833                                            found_key.offset);
7834                         free_excluded_extents(root, cache);
7835                 }
7836
7837                 ret = update_space_info(info, cache->flags, found_key.offset,
7838                                         btrfs_block_group_used(&cache->item),
7839                                         &space_info);
7840                 BUG_ON(ret);
7841                 cache->space_info = space_info;
7842                 spin_lock(&cache->space_info->lock);
7843                 cache->space_info->bytes_readonly += cache->bytes_super;
7844                 spin_unlock(&cache->space_info->lock);
7845
7846                 __link_block_group(space_info, cache);
7847
7848                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7849                 BUG_ON(ret);
7850
7851                 set_avail_alloc_bits(root->fs_info, cache->flags);
7852                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7853                         set_block_group_ro(cache);
7854         }
7855
7856         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
7857                 if (!(get_alloc_profile(root, space_info->flags) &
7858                       (BTRFS_BLOCK_GROUP_RAID10 |
7859                        BTRFS_BLOCK_GROUP_RAID1 |
7860                        BTRFS_BLOCK_GROUP_DUP)))
7861                         continue;
7862                 /*
7863                  * avoid allocating from un-mirrored block group if there are
7864                  * mirrored block groups.
7865                  */
7866                 list_for_each_entry(cache, &space_info->block_groups[3], list)
7867                         set_block_group_ro(cache);
7868                 list_for_each_entry(cache, &space_info->block_groups[4], list)
7869                         set_block_group_ro(cache);
7870         }
7871
7872         init_global_block_rsv(info);
7873         ret = 0;
7874 error:
7875         btrfs_free_path(path);
7876         return ret;
7877 }
7878
7879 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7880                            struct btrfs_root *root, u64 bytes_used,
7881                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7882                            u64 size)
7883 {
7884         int ret;
7885         struct btrfs_root *extent_root;
7886         struct btrfs_block_group_cache *cache;
7887
7888         extent_root = root->fs_info->extent_root;
7889
7890         root->fs_info->last_trans_log_full_commit = trans->transid;
7891
7892         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7893         if (!cache)
7894                 return -ENOMEM;
7895
7896         cache->key.objectid = chunk_offset;
7897         cache->key.offset = size;
7898         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7899         cache->sectorsize = root->sectorsize;
7900
7901         /*
7902          * we only want to have 32k of ram per block group for keeping track
7903          * of free space, and if we pass 1/2 of that we want to start
7904          * converting things over to using bitmaps
7905          */
7906         cache->extents_thresh = ((1024 * 32) / 2) /
7907                 sizeof(struct btrfs_free_space);
7908         atomic_set(&cache->count, 1);
7909         spin_lock_init(&cache->lock);
7910         spin_lock_init(&cache->tree_lock);
7911         INIT_LIST_HEAD(&cache->list);
7912         INIT_LIST_HEAD(&cache->cluster_list);
7913
7914         btrfs_set_block_group_used(&cache->item, bytes_used);
7915         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7916         cache->flags = type;
7917         btrfs_set_block_group_flags(&cache->item, type);
7918
7919         cache->last_byte_to_unpin = (u64)-1;
7920         cache->cached = BTRFS_CACHE_FINISHED;
7921         exclude_super_stripes(root, cache);
7922
7923         add_new_free_space(cache, root->fs_info, chunk_offset,
7924                            chunk_offset + size);
7925
7926         free_excluded_extents(root, cache);
7927
7928         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7929                                 &cache->space_info);
7930         BUG_ON(ret);
7931
7932         spin_lock(&cache->space_info->lock);
7933         cache->space_info->bytes_readonly += cache->bytes_super;
7934         spin_unlock(&cache->space_info->lock);
7935
7936         __link_block_group(cache->space_info, cache);
7937
7938         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7939         BUG_ON(ret);
7940
7941         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7942                                 sizeof(cache->item));
7943         BUG_ON(ret);
7944
7945         set_avail_alloc_bits(extent_root->fs_info, type);
7946
7947         return 0;
7948 }
7949
7950 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7951                              struct btrfs_root *root, u64 group_start)
7952 {
7953         struct btrfs_path *path;
7954         struct btrfs_block_group_cache *block_group;
7955         struct btrfs_free_cluster *cluster;
7956         struct btrfs_key key;
7957         int ret;
7958
7959         root = root->fs_info->extent_root;
7960
7961         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7962         BUG_ON(!block_group);
7963         BUG_ON(!block_group->ro);
7964
7965         memcpy(&key, &block_group->key, sizeof(key));
7966
7967         /* make sure this block group isn't part of an allocation cluster */
7968         cluster = &root->fs_info->data_alloc_cluster;
7969         spin_lock(&cluster->refill_lock);
7970         btrfs_return_cluster_to_free_space(block_group, cluster);
7971         spin_unlock(&cluster->refill_lock);
7972
7973         /*
7974          * make sure this block group isn't part of a metadata
7975          * allocation cluster
7976          */
7977         cluster = &root->fs_info->meta_alloc_cluster;
7978         spin_lock(&cluster->refill_lock);
7979         btrfs_return_cluster_to_free_space(block_group, cluster);
7980         spin_unlock(&cluster->refill_lock);
7981
7982         path = btrfs_alloc_path();
7983         BUG_ON(!path);
7984
7985         spin_lock(&root->fs_info->block_group_cache_lock);
7986         rb_erase(&block_group->cache_node,
7987                  &root->fs_info->block_group_cache_tree);
7988         spin_unlock(&root->fs_info->block_group_cache_lock);
7989
7990         down_write(&block_group->space_info->groups_sem);
7991         /*
7992          * we must use list_del_init so people can check to see if they
7993          * are still on the list after taking the semaphore
7994          */
7995         list_del_init(&block_group->list);
7996         up_write(&block_group->space_info->groups_sem);
7997
7998         if (block_group->cached == BTRFS_CACHE_STARTED)
7999                 wait_block_group_cache_done(block_group);
8000
8001         btrfs_remove_free_space_cache(block_group);
8002
8003         spin_lock(&block_group->space_info->lock);
8004         block_group->space_info->total_bytes -= block_group->key.offset;
8005         block_group->space_info->bytes_readonly -= block_group->key.offset;
8006         spin_unlock(&block_group->space_info->lock);
8007
8008         btrfs_clear_space_info_full(root->fs_info);
8009
8010         btrfs_put_block_group(block_group);
8011         btrfs_put_block_group(block_group);
8012
8013         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8014         if (ret > 0)
8015                 ret = -EIO;
8016         if (ret < 0)
8017                 goto out;
8018
8019         ret = btrfs_del_item(trans, root, path);
8020 out:
8021         btrfs_free_path(path);
8022         return ret;
8023 }