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