fs/btrfs/extent-tree.c

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