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         /* We're loading it the fast way, so we don't have a caching_ctl. */
 246         if (!cache->caching_ctl) {
 247                 spin_unlock(&cache->lock);
 248                 return NULL;
 249         }
 250
 251         ctl = cache->caching_ctl;
 252         atomic_inc(&ctl->count);
 253         spin_unlock(&cache->lock);
 254         return ctl;
 255 }
 256
 257 static void put_caching_control(struct btrfs_caching_control *ctl)
 258 {
 259         if (atomic_dec_and_test(&ctl->count))
 260                 kfree(ctl);
 261 }
 262
 263 /*
 264  * this is only called by cache_block_group, since we could have freed extents
 265  * we need to check the pinned_extents for any extents that can't be used yet
 266  * since their free space will be released as soon as the transaction commits.
 267  */
 268 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
 269                               struct btrfs_fs_info *info, u64 start, u64 end)
 270 {
 271         u64 extent_start, extent_end, size, total_added = 0;
 272         int ret;
 273
 274         while (start < end) {
 275                 ret = find_first_extent_bit(info->pinned_extents, start,
 276                                             &extent_start, &extent_end,
 277                                             EXTENT_DIRTY | EXTENT_UPTODATE);
 278                 if (ret)
 279                         break;
 280
 281                 if (extent_start <= start) {
 282                         start = extent_end + 1;
 283                 } else if (extent_start > start && extent_start < end) {
 284                         size = extent_start - start;
 285                         total_added += size;
 286                         ret = btrfs_add_free_space(block_group, start,
 287                                                    size);
 288                         BUG_ON(ret);
 289                         start = extent_end + 1;
 290                 } else {
 291                         break;
 292                 }
 293         }
 294
 295         if (start < end) {
 296                 size = end - start;
 297                 total_added += size;
 298                 ret = btrfs_add_free_space(block_group, start, size);
 299                 BUG_ON(ret);
 300         }
 301
 302         return total_added;
 303 }
 304
 305 static int caching_kthread(void *data)
 306 {
 307         struct btrfs_block_group_cache *block_group = data;
 308         struct btrfs_fs_info *fs_info = block_group->fs_info;
 309         struct btrfs_caching_control *caching_ctl = block_group->caching_ctl;
 310         struct btrfs_root *extent_root = fs_info->extent_root;
 311         struct btrfs_path *path;
 312         struct extent_buffer *leaf;
 313         struct btrfs_key key;
 314         u64 total_found = 0;
 315         u64 last = 0;
 316         u32 nritems;
 317         int ret = 0;
 318
 319         path = btrfs_alloc_path();
 320         if (!path)
 321                 return -ENOMEM;
 322
 323         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 324
 325         /*
 326          * We don't want to deadlock with somebody trying to allocate a new
 327          * extent for the extent root while also trying to search the extent
 328          * root to add free space.  So we skip locking and search the commit
 329          * root, since its read-only
 330          */
 331         path->skip_locking = 1;
 332         path->search_commit_root = 1;
 333         path->reada = 2;
 334
 335         key.objectid = last;
 336         key.offset = 0;
 337         key.type = BTRFS_EXTENT_ITEM_KEY;
 338 again:
 339         mutex_lock(&caching_ctl->mutex);
 340         /* need to make sure the commit_root doesn't disappear */
 341         down_read(&fs_info->extent_commit_sem);
 342
 343         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 344         if (ret < 0)
 345                 goto err;
 346
 347         leaf = path->nodes[0];
 348         nritems = btrfs_header_nritems(leaf);
 349
 350         while (1) {
 351                 smp_mb();
 352                 if (fs_info->closing > 1) {
 353                         last = (u64)-1;
 354                         break;
 355                 }
 356
 357                 if (path->slots[0] < nritems) {
 358                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 359                 } else {
 360                         ret = find_next_key(path, 0, &key);
 361                         if (ret)
 362                                 break;
 363
 364                         caching_ctl->progress = last;
 365                         btrfs_release_path(extent_root, path);
 366                         up_read(&fs_info->extent_commit_sem);
 367                         mutex_unlock(&caching_ctl->mutex);
 368                         if (btrfs_transaction_in_commit(fs_info))
 369                                 schedule_timeout(1);
 370                         else
 371                                 cond_resched();
 372                         goto again;
 373                 }
 374
 375                 if (key.objectid < block_group->key.objectid) {
 376                         path->slots[0]++;
 377                         continue;
 378                 }
 379
 380                 if (key.objectid >= block_group->key.objectid +
 381                     block_group->key.offset)
 382                         break;
 383
 384                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 385                         total_found += add_new_free_space(block_group,
 386                                                           fs_info, last,
 387                                                           key.objectid);
 388                         last = key.objectid + key.offset;
 389
 390                         if (total_found > (1024 * 1024 * 2)) {
 391                                 total_found = 0;
 392                                 wake_up(&caching_ctl->wait);
 393                         }
 394                 }
 395                 path->slots[0]++;
 396         }
 397         ret = 0;
 398
 399         total_found += add_new_free_space(block_group, fs_info, last,
 400                                           block_group->key.objectid +
 401                                           block_group->key.offset);
 402         caching_ctl->progress = (u64)-1;
 403
 404         spin_lock(&block_group->lock);
 405         block_group->caching_ctl = NULL;
 406         block_group->cached = BTRFS_CACHE_FINISHED;
 407         spin_unlock(&block_group->lock);
 408
 409 err:
 410         btrfs_free_path(path);
 411         up_read(&fs_info->extent_commit_sem);
 412
 413         free_excluded_extents(extent_root, block_group);
 414
 415         mutex_unlock(&caching_ctl->mutex);
 416         wake_up(&caching_ctl->wait);
 417
 418         put_caching_control(caching_ctl);
 419         atomic_dec(&block_group->space_info->caching_threads);
 420         btrfs_put_block_group(block_group);
 421
 422         return 0;
 423 }
 424
 425 static int cache_block_group(struct btrfs_block_group_cache *cache,
 426                              struct btrfs_trans_handle *trans,
 427                              struct btrfs_root *root,
 428                              int load_cache_only)
 429 {
 430         struct btrfs_fs_info *fs_info = cache->fs_info;
 431         struct btrfs_caching_control *caching_ctl;
 432         struct task_struct *tsk;
 433         int ret = 0;
 434
 435         smp_mb();
 436         if (cache->cached != BTRFS_CACHE_NO)
 437                 return 0;
 438
 439         /*
 440          * We can't do the read from on-disk cache during a commit since we need
 441          * to have the normal tree locking.  Also if we are currently trying to
 442          * allocate blocks for the tree root we can't do the fast caching since
 443          * we likely hold important locks.
 444          */
 445         if (!trans->transaction->in_commit &&
 446             (root && root != root->fs_info->tree_root)) {
 447                 spin_lock(&cache->lock);
 448                 if (cache->cached != BTRFS_CACHE_NO) {
 449                         spin_unlock(&cache->lock);
 450                         return 0;
 451                 }
 452                 cache->cached = BTRFS_CACHE_STARTED;
 453                 spin_unlock(&cache->lock);
 454
 455                 ret = load_free_space_cache(fs_info, cache);
 456
 457                 spin_lock(&cache->lock);
 458                 if (ret == 1) {
 459                         cache->cached = BTRFS_CACHE_FINISHED;
 460                         cache->last_byte_to_unpin = (u64)-1;
 461                 } else {
 462                         cache->cached = BTRFS_CACHE_NO;
 463                 }
 464                 spin_unlock(&cache->lock);
 465                 if (ret == 1) {
 466                         free_excluded_extents(fs_info->extent_root, cache);
 467                         return 0;
 468                 }
 469         }
 470
 471         if (load_cache_only)
 472                 return 0;
 473
 474         caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
 475         BUG_ON(!caching_ctl);
 476
 477         INIT_LIST_HEAD(&caching_ctl->list);
 478         mutex_init(&caching_ctl->mutex);
 479         init_waitqueue_head(&caching_ctl->wait);
 480         caching_ctl->block_group = cache;
 481         caching_ctl->progress = cache->key.objectid;
 482         /* one for caching kthread, one for caching block group list */
 483         atomic_set(&caching_ctl->count, 2);
 484
 485         spin_lock(&cache->lock);
 486         if (cache->cached != BTRFS_CACHE_NO) {
 487                 spin_unlock(&cache->lock);
 488                 kfree(caching_ctl);
 489                 return 0;
 490         }
 491         cache->caching_ctl = caching_ctl;
 492         cache->cached = BTRFS_CACHE_STARTED;
 493         spin_unlock(&cache->lock);
 494
 495         down_write(&fs_info->extent_commit_sem);
 496         list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
 497         up_write(&fs_info->extent_commit_sem);
 498
 499         atomic_inc(&cache->space_info->caching_threads);
 500         btrfs_get_block_group(cache);
 501
 502         tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
 503                           cache->key.objectid);
 504         if (IS_ERR(tsk)) {
 505                 ret = PTR_ERR(tsk);
 506                 printk(KERN_ERR "error running thread %d\n", ret);
 507                 BUG();
 508         }
 509
 510         return ret;
 511 }
 512
 513 /*
 514  * return the block group that starts at or after bytenr
 515  */
 516 static struct btrfs_block_group_cache *
 517 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 518 {
 519         struct btrfs_block_group_cache *cache;
 520
 521         cache = block_group_cache_tree_search(info, bytenr, 0);
 522
 523         return cache;
 524 }
 525
 526 /*
 527  * return the block group that contains the given bytenr
 528  */
 529 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 530                                                  struct btrfs_fs_info *info,
 531                                                  u64 bytenr)
 532 {
 533         struct btrfs_block_group_cache *cache;
 534
 535         cache = block_group_cache_tree_search(info, bytenr, 1);
 536
 537         return cache;
 538 }
 539
 540 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 541                                                   u64 flags)
 542 {
 543         struct list_head *head = &info->space_info;
 544         struct btrfs_space_info *found;
 545
 546         flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM |
 547                  BTRFS_BLOCK_GROUP_METADATA;
 548
 549         rcu_read_lock();
 550         list_for_each_entry_rcu(found, head, list) {
 551                 if (found->flags & flags) {
 552                         rcu_read_unlock();
 553                         return found;
 554                 }
 555         }
 556         rcu_read_unlock();
 557         return NULL;
 558 }
 559
 560 /*
 561  * after adding space to the filesystem, we need to clear the full flags
 562  * on all the space infos.
 563  */
 564 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 565 {
 566         struct list_head *head = &info->space_info;
 567         struct btrfs_space_info *found;
 568
 569         rcu_read_lock();
 570         list_for_each_entry_rcu(found, head, list)
 571                 found->full = 0;
 572         rcu_read_unlock();
 573 }
 574
 575 static u64 div_factor(u64 num, int factor)
 576 {
 577         if (factor == 10)
 578                 return num;
 579         num *= factor;
 580         do_div(num, 10);
 581         return num;
 582 }
 583
 584 static u64 div_factor_fine(u64 num, int factor)
 585 {
 586         if (factor == 100)
 587                 return num;
 588         num *= factor;
 589         do_div(num, 100);
 590         return num;
 591 }
 592
 593 u64 btrfs_find_block_group(struct btrfs_root *root,
 594                            u64 search_start, u64 search_hint, int owner)
 595 {
 596         struct btrfs_block_group_cache *cache;
 597         u64 used;
 598         u64 last = max(search_hint, search_start);
 599         u64 group_start = 0;
 600         int full_search = 0;
 601         int factor = 9;
 602         int wrapped = 0;
 603 again:
 604         while (1) {
 605                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 606                 if (!cache)
 607                         break;
 608
 609                 spin_lock(&cache->lock);
 610                 last = cache->key.objectid + cache->key.offset;
 611                 used = btrfs_block_group_used(&cache->item);
 612
 613                 if ((full_search || !cache->ro) &&
 614                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 615                         if (used + cache->pinned + cache->reserved <
 616                             div_factor(cache->key.offset, factor)) {
 617                                 group_start = cache->key.objectid;
 618                                 spin_unlock(&cache->lock);
 619                                 btrfs_put_block_group(cache);
 620                                 goto found;
 621                         }
 622                 }
 623                 spin_unlock(&cache->lock);
 624                 btrfs_put_block_group(cache);
 625                 cond_resched();
 626         }
 627         if (!wrapped) {
 628                 last = search_start;
 629                 wrapped = 1;
 630                 goto again;
 631         }
 632         if (!full_search && factor < 10) {
 633                 last = search_start;
 634                 full_search = 1;
 635                 factor = 10;
 636                 goto again;
 637         }
 638 found:
 639         return group_start;
 640 }
 641
 642 /* simple helper to search for an existing extent at a given offset */
 643 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 644 {
 645         int ret;
 646         struct btrfs_key key;
 647         struct btrfs_path *path;
 648
 649         path = btrfs_alloc_path();
 650         BUG_ON(!path);
 651         key.objectid = start;
 652         key.offset = len;
 653         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 654         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 655                                 0, 0);
 656         btrfs_free_path(path);
 657         return ret;
 658 }
 659
 660 /*
 661  * helper function to lookup reference count and flags of extent.
 662  *
 663  * the head node for delayed ref is used to store the sum of all the
 664  * reference count modifications queued up in the rbtree. the head
 665  * node may also store the extent flags to set. This way you can check
 666  * to see what the reference count and extent flags would be if all of
 667  * the delayed refs are not processed.
 668  */
 669 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
 670                              struct btrfs_root *root, u64 bytenr,
 671                              u64 num_bytes, u64 *refs, u64 *flags)
 672 {
 673         struct btrfs_delayed_ref_head *head;
 674         struct btrfs_delayed_ref_root *delayed_refs;
 675         struct btrfs_path *path;
 676         struct btrfs_extent_item *ei;
 677         struct extent_buffer *leaf;
 678         struct btrfs_key key;
 679         u32 item_size;
 680         u64 num_refs;
 681         u64 extent_flags;
 682         int ret;
 683
 684         path = btrfs_alloc_path();
 685         if (!path)
 686                 return -ENOMEM;
 687
 688         key.objectid = bytenr;
 689         key.type = BTRFS_EXTENT_ITEM_KEY;
 690         key.offset = num_bytes;
 691         if (!trans) {
 692                 path->skip_locking = 1;
 693                 path->search_commit_root = 1;
 694         }
 695 again:
 696         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
 697                                 &key, path, 0, 0);
 698         if (ret < 0)
 699                 goto out_free;
 700
 701         if (ret == 0) {
 702                 leaf = path->nodes[0];
 703                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 704                 if (item_size >= sizeof(*ei)) {
 705                         ei = btrfs_item_ptr(leaf, path->slots[0],
 706                                             struct btrfs_extent_item);
 707                         num_refs = btrfs_extent_refs(leaf, ei);
 708                         extent_flags = btrfs_extent_flags(leaf, ei);
 709                 } else {
 710 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 711                         struct btrfs_extent_item_v0 *ei0;
 712                         BUG_ON(item_size != sizeof(*ei0));
 713                         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 714                                              struct btrfs_extent_item_v0);
 715                         num_refs = btrfs_extent_refs_v0(leaf, ei0);
 716                         /* FIXME: this isn't correct for data */
 717                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
 718 #else
 719                         BUG();
 720 #endif
 721                 }
 722                 BUG_ON(num_refs == 0);
 723         } else {
 724                 num_refs = 0;
 725                 extent_flags = 0;
 726                 ret = 0;
 727         }
 728
 729         if (!trans)
 730                 goto out;
 731
 732         delayed_refs = &trans->transaction->delayed_refs;
 733         spin_lock(&delayed_refs->lock);
 734         head = btrfs_find_delayed_ref_head(trans, bytenr);
 735         if (head) {
 736                 if (!mutex_trylock(&head->mutex)) {
 737                         atomic_inc(&head->node.refs);
 738                         spin_unlock(&delayed_refs->lock);
 739
 740                         btrfs_release_path(root->fs_info->extent_root, path);
 741
 742                         mutex_lock(&head->mutex);
 743                         mutex_unlock(&head->mutex);
 744                         btrfs_put_delayed_ref(&head->node);
 745                         goto again;
 746                 }
 747                 if (head->extent_op && head->extent_op->update_flags)
 748                         extent_flags |= head->extent_op->flags_to_set;
 749                 else
 750                         BUG_ON(num_refs == 0);
 751
 752                 num_refs += head->node.ref_mod;
 753                 mutex_unlock(&head->mutex);
 754         }
 755         spin_unlock(&delayed_refs->lock);
 756 out:
 757         WARN_ON(num_refs == 0);
 758         if (refs)
 759                 *refs = num_refs;
 760         if (flags)
 761                 *flags = extent_flags;
 762 out_free:
 763         btrfs_free_path(path);
 764         return ret;
 765 }
 766
 767 /*
 768  * Back reference rules.  Back refs have three main goals:
 769  *
 770  * 1) differentiate between all holders of references to an extent so that
 771  *    when a reference is dropped we can make sure it was a valid reference
 772  *    before freeing the extent.
 773  *
 774  * 2) Provide enough information to quickly find the holders of an extent
 775  *    if we notice a given block is corrupted or bad.
 776  *
 777  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 778  *    maintenance.  This is actually the same as #2, but with a slightly
 779  *    different use case.
 780  *
 781  * There are two kinds of back refs. The implicit back refs is optimized
 782  * for pointers in non-shared tree blocks. For a given pointer in a block,
 783  * back refs of this kind provide information about the block's owner tree
 784  * and the pointer's key. These information allow us to find the block by
 785  * b-tree searching. The full back refs is for pointers in tree blocks not
 786  * referenced by their owner trees. The location of tree block is recorded
 787  * in the back refs. Actually the full back refs is generic, and can be
 788  * used in all cases the implicit back refs is used. The major shortcoming
 789  * of the full back refs is its overhead. Every time a tree block gets
 790  * COWed, we have to update back refs entry for all pointers in it.
 791  *
 792  * For a newly allocated tree block, we use implicit back refs for
 793  * pointers in it. This means most tree related operations only involve
 794  * implicit back refs. For a tree block created in old transaction, the
 795  * only way to drop a reference to it is COW it. So we can detect the
 796  * event that tree block loses its owner tree's reference and do the
 797  * back refs conversion.
 798  *
 799  * When a tree block is COW'd through a tree, there are four cases:
 800  *
 801  * The reference count of the block is one and the tree is the block's
 802  * owner tree. Nothing to do in this case.
 803  *
 804  * The reference count of the block is one and the tree is not the
 805  * block's owner tree. In this case, full back refs is used for pointers
 806  * in the block. Remove these full back refs, add implicit back refs for
 807  * every pointers in the new block.
 808  *
 809  * The reference count of the block is greater than one and the tree is
 810  * the block's owner tree. In this case, implicit back refs is used for
 811  * pointers in the block. Add full back refs for every pointers in the
 812  * block, increase lower level extents' reference counts. The original
 813  * implicit back refs are entailed to the new block.
 814  *
 815  * The reference count of the block is greater than one and the tree is
 816  * not the block's owner tree. Add implicit back refs for every pointer in
 817  * the new block, increase lower level extents' reference count.
 818  *
 819  * Back Reference Key composing:
 820  *
 821  * The key objectid corresponds to the first byte in the extent,
 822  * The key type is used to differentiate between types of back refs.
 823  * There are different meanings of the key offset for different types
 824  * of back refs.
 825  *
 826  * File extents can be referenced by:
 827  *
 828  * - multiple snapshots, subvolumes, or different generations in one subvol
 829  * - different files inside a single subvolume
 830  * - different offsets inside a file (bookend extents in file.c)
 831  *
 832  * The extent ref structure for the implicit back refs has fields for:
 833  *
 834  * - Objectid of the subvolume root
 835  * - objectid of the file holding the reference
 836  * - original offset in the file
 837  * - how many bookend extents
 838  *
 839  * The key offset for the implicit back refs is hash of the first
 840  * three fields.
 841  *
 842  * The extent ref structure for the full back refs has field for:
 843  *
 844  * - number of pointers in the tree leaf
 845  *
 846  * The key offset for the implicit back refs is the first byte of
 847  * the tree leaf
 848  *
 849  * When a file extent is allocated, The implicit back refs is used.
 850  * the fields are filled in:
 851  *
 852  *     (root_key.objectid, inode objectid, offset in file, 1)
 853  *
 854  * When a file extent is removed file truncation, we find the
 855  * corresponding implicit back refs and check the following fields:
 856  *
 857  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
 858  *
 859  * Btree extents can be referenced by:
 860  *
 861  * - Different subvolumes
 862  *
 863  * Both the implicit back refs and the full back refs for tree blocks
 864  * only consist of key. The key offset for the implicit back refs is
 865  * objectid of block's owner tree. The key offset for the full back refs
 866  * is the first byte of parent block.
 867  *
 868  * When implicit back refs is used, information about the lowest key and
 869  * level of the tree block are required. These information are stored in
 870  * tree block info structure.
 871  */
 872
 873 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 874 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
 875                                   struct btrfs_root *root,
 876                                   struct btrfs_path *path,
 877                                   u64 owner, u32 extra_size)
 878 {
 879         struct btrfs_extent_item *item;
 880         struct btrfs_extent_item_v0 *ei0;
 881         struct btrfs_extent_ref_v0 *ref0;
 882         struct btrfs_tree_block_info *bi;
 883         struct extent_buffer *leaf;
 884         struct btrfs_key key;
 885         struct btrfs_key found_key;
 886         u32 new_size = sizeof(*item);
 887         u64 refs;
 888         int ret;
 889
 890         leaf = path->nodes[0];
 891         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
 892
 893         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 894         ei0 = btrfs_item_ptr(leaf, path->slots[0],
 895                              struct btrfs_extent_item_v0);
 896         refs = btrfs_extent_refs_v0(leaf, ei0);
 897
 898         if (owner == (u64)-1) {
 899                 while (1) {
 900                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 901                                 ret = btrfs_next_leaf(root, path);
 902                                 if (ret < 0)
 903                                         return ret;
 904                                 BUG_ON(ret > 0);
 905                                 leaf = path->nodes[0];
 906                         }
 907                         btrfs_item_key_to_cpu(leaf, &found_key,
 908                                               path->slots[0]);
 909                         BUG_ON(key.objectid != found_key.objectid);
 910                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
 911                                 path->slots[0]++;
 912                                 continue;
 913                         }
 914                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
 915                                               struct btrfs_extent_ref_v0);
 916                         owner = btrfs_ref_objectid_v0(leaf, ref0);
 917                         break;
 918                 }
 919         }
 920         btrfs_release_path(root, path);
 921
 922         if (owner < BTRFS_FIRST_FREE_OBJECTID)
 923                 new_size += sizeof(*bi);
 924
 925         new_size -= sizeof(*ei0);
 926         ret = btrfs_search_slot(trans, root, &key, path,
 927                                 new_size + extra_size, 1);
 928         if (ret < 0)
 929                 return ret;
 930         BUG_ON(ret);
 931
 932         ret = btrfs_extend_item(trans, root, path, new_size);
 933         BUG_ON(ret);
 934
 935         leaf = path->nodes[0];
 936         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
 937         btrfs_set_extent_refs(leaf, item, refs);
 938         /* FIXME: get real generation */
 939         btrfs_set_extent_generation(leaf, item, 0);
 940         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
 941                 btrfs_set_extent_flags(leaf, item,
 942                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
 943                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
 944                 bi = (struct btrfs_tree_block_info *)(item + 1);
 945                 /* FIXME: get first key of the block */
 946                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
 947                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
 948         } else {
 949                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
 950         }
 951         btrfs_mark_buffer_dirty(leaf);
 952         return 0;
 953 }
 954 #endif
 955
 956 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
 957 {
 958         u32 high_crc = ~(u32)0;
 959         u32 low_crc = ~(u32)0;
 960         __le64 lenum;
 961
 962         lenum = cpu_to_le64(root_objectid);
 963         high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
 964         lenum = cpu_to_le64(owner);
 965         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 966         lenum = cpu_to_le64(offset);
 967         low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
 968
 969         return ((u64)high_crc << 31) ^ (u64)low_crc;
 970 }
 971
 972 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
 973                                      struct btrfs_extent_data_ref *ref)
 974 {
 975         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
 976                                     btrfs_extent_data_ref_objectid(leaf, ref),
 977                                     btrfs_extent_data_ref_offset(leaf, ref));
 978 }
 979
 980 static int match_extent_data_ref(struct extent_buffer *leaf,
 981                                  struct btrfs_extent_data_ref *ref,
 982                                  u64 root_objectid, u64 owner, u64 offset)
 983 {
 984         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
 985             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
 986             btrfs_extent_data_ref_offset(leaf, ref) != offset)
 987                 return 0;
 988         return 1;
 989 }
 990
 991 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
 992                                            struct btrfs_root *root,
 993                                            struct btrfs_path *path,
 994                                            u64 bytenr, u64 parent,
 995                                            u64 root_objectid,
 996                                            u64 owner, u64 offset)
 997 {
 998         struct btrfs_key key;
 999         struct btrfs_extent_data_ref *ref;
1000         struct extent_buffer *leaf;
1001         u32 nritems;
1002         int ret;
1003         int recow;
1004         int err = -ENOENT;
1005
1006         key.objectid = bytenr;
1007         if (parent) {
1008                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1009                 key.offset = parent;
1010         } else {
1011                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1012                 key.offset = hash_extent_data_ref(root_objectid,
1013                                                   owner, offset);
1014         }
1015 again:
1016         recow = 0;
1017         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1018         if (ret < 0) {
1019                 err = ret;
1020                 goto fail;
1021         }
1022
1023         if (parent) {
1024                 if (!ret)
1025                         return 0;
1026 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1027                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1028                 btrfs_release_path(root, path);
1029                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1030                 if (ret < 0) {
1031                         err = ret;
1032                         goto fail;
1033                 }
1034                 if (!ret)
1035                         return 0;
1036 #endif
1037                 goto fail;
1038         }
1039
1040         leaf = path->nodes[0];
1041         nritems = btrfs_header_nritems(leaf);
1042         while (1) {
1043                 if (path->slots[0] >= nritems) {
1044                         ret = btrfs_next_leaf(root, path);
1045                         if (ret < 0)
1046                                 err = ret;
1047                         if (ret)
1048                                 goto fail;
1049
1050                         leaf = path->nodes[0];
1051                         nritems = btrfs_header_nritems(leaf);
1052                         recow = 1;
1053                 }
1054
1055                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1056                 if (key.objectid != bytenr ||
1057                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
1058                         goto fail;
1059
1060                 ref = btrfs_item_ptr(leaf, path->slots[0],
1061                                      struct btrfs_extent_data_ref);
1062
1063                 if (match_extent_data_ref(leaf, ref, root_objectid,
1064                                           owner, offset)) {
1065                         if (recow) {
1066                                 btrfs_release_path(root, path);
1067                                 goto again;
1068                         }
1069                         err = 0;
1070                         break;
1071                 }
1072                 path->slots[0]++;
1073         }
1074 fail:
1075         return err;
1076 }
1077
1078 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1079                                            struct btrfs_root *root,
1080                                            struct btrfs_path *path,
1081                                            u64 bytenr, u64 parent,
1082                                            u64 root_objectid, u64 owner,
1083                                            u64 offset, int refs_to_add)
1084 {
1085         struct btrfs_key key;
1086         struct extent_buffer *leaf;
1087         u32 size;
1088         u32 num_refs;
1089         int ret;
1090
1091         key.objectid = bytenr;
1092         if (parent) {
1093                 key.type = BTRFS_SHARED_DATA_REF_KEY;
1094                 key.offset = parent;
1095                 size = sizeof(struct btrfs_shared_data_ref);
1096         } else {
1097                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1098                 key.offset = hash_extent_data_ref(root_objectid,
1099                                                   owner, offset);
1100                 size = sizeof(struct btrfs_extent_data_ref);
1101         }
1102
1103         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1104         if (ret && ret != -EEXIST)
1105                 goto fail;
1106
1107         leaf = path->nodes[0];
1108         if (parent) {
1109                 struct btrfs_shared_data_ref *ref;
1110                 ref = btrfs_item_ptr(leaf, path->slots[0],
1111                                      struct btrfs_shared_data_ref);
1112                 if (ret == 0) {
1113                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1114                 } else {
1115                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
1116                         num_refs += refs_to_add;
1117                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1118                 }
1119         } else {
1120                 struct btrfs_extent_data_ref *ref;
1121                 while (ret == -EEXIST) {
1122                         ref = btrfs_item_ptr(leaf, path->slots[0],
1123                                              struct btrfs_extent_data_ref);
1124                         if (match_extent_data_ref(leaf, ref, root_objectid,
1125                                                   owner, offset))
1126                                 break;
1127                         btrfs_release_path(root, path);
1128                         key.offset++;
1129                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1130                                                       size);
1131                         if (ret && ret != -EEXIST)
1132                                 goto fail;
1133
1134                         leaf = path->nodes[0];
1135                 }
1136                 ref = btrfs_item_ptr(leaf, path->slots[0],
1137                                      struct btrfs_extent_data_ref);
1138                 if (ret == 0) {
1139                         btrfs_set_extent_data_ref_root(leaf, ref,
1140                                                        root_objectid);
1141                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1142                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1143                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1144                 } else {
1145                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
1146                         num_refs += refs_to_add;
1147                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1148                 }
1149         }
1150         btrfs_mark_buffer_dirty(leaf);
1151         ret = 0;
1152 fail:
1153         btrfs_release_path(root, path);
1154         return ret;
1155 }
1156
1157 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1158                                            struct btrfs_root *root,
1159                                            struct btrfs_path *path,
1160                                            int refs_to_drop)
1161 {
1162         struct btrfs_key key;
1163         struct btrfs_extent_data_ref *ref1 = NULL;
1164         struct btrfs_shared_data_ref *ref2 = NULL;
1165         struct extent_buffer *leaf;
1166         u32 num_refs = 0;
1167         int ret = 0;
1168
1169         leaf = path->nodes[0];
1170         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1171
1172         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1173                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1174                                       struct btrfs_extent_data_ref);
1175                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1176         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1177                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1178                                       struct btrfs_shared_data_ref);
1179                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1180 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1181         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1182                 struct btrfs_extent_ref_v0 *ref0;
1183                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1184                                       struct btrfs_extent_ref_v0);
1185                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1186 #endif
1187         } else {
1188                 BUG();
1189         }
1190
1191         BUG_ON(num_refs < refs_to_drop);
1192         num_refs -= refs_to_drop;
1193
1194         if (num_refs == 0) {
1195                 ret = btrfs_del_item(trans, root, path);
1196         } else {
1197                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1198                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1199                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1200                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1201 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1202                 else {
1203                         struct btrfs_extent_ref_v0 *ref0;
1204                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
1205                                         struct btrfs_extent_ref_v0);
1206                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1207                 }
1208 #endif
1209                 btrfs_mark_buffer_dirty(leaf);
1210         }
1211         return ret;
1212 }
1213
1214 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1215                                           struct btrfs_path *path,
1216                                           struct btrfs_extent_inline_ref *iref)
1217 {
1218         struct btrfs_key key;
1219         struct extent_buffer *leaf;
1220         struct btrfs_extent_data_ref *ref1;
1221         struct btrfs_shared_data_ref *ref2;
1222         u32 num_refs = 0;
1223
1224         leaf = path->nodes[0];
1225         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1226         if (iref) {
1227                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1228                     BTRFS_EXTENT_DATA_REF_KEY) {
1229                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1230                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1231                 } else {
1232                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1233                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1234                 }
1235         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1236                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1237                                       struct btrfs_extent_data_ref);
1238                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1239         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1240                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1241                                       struct btrfs_shared_data_ref);
1242                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1243 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1244         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1245                 struct btrfs_extent_ref_v0 *ref0;
1246                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1247                                       struct btrfs_extent_ref_v0);
1248                 num_refs = btrfs_ref_count_v0(leaf, ref0);
1249 #endif
1250         } else {
1251                 WARN_ON(1);
1252         }
1253         return num_refs;
1254 }
1255
1256 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1257                                           struct btrfs_root *root,
1258                                           struct btrfs_path *path,
1259                                           u64 bytenr, u64 parent,
1260                                           u64 root_objectid)
1261 {
1262         struct btrfs_key key;
1263         int ret;
1264
1265         key.objectid = bytenr;
1266         if (parent) {
1267                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1268                 key.offset = parent;
1269         } else {
1270                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1271                 key.offset = root_objectid;
1272         }
1273
1274         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1275         if (ret > 0)
1276                 ret = -ENOENT;
1277 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1278         if (ret == -ENOENT && parent) {
1279                 btrfs_release_path(root, path);
1280                 key.type = BTRFS_EXTENT_REF_V0_KEY;
1281                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1282                 if (ret > 0)
1283                         ret = -ENOENT;
1284         }
1285 #endif
1286         return ret;
1287 }
1288
1289 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1290                                           struct btrfs_root *root,
1291                                           struct btrfs_path *path,
1292                                           u64 bytenr, u64 parent,
1293                                           u64 root_objectid)
1294 {
1295         struct btrfs_key key;
1296         int ret;
1297
1298         key.objectid = bytenr;
1299         if (parent) {
1300                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1301                 key.offset = parent;
1302         } else {
1303                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1304                 key.offset = root_objectid;
1305         }
1306
1307         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1308         btrfs_release_path(root, path);
1309         return ret;
1310 }
1311
1312 static inline int extent_ref_type(u64 parent, u64 owner)
1313 {
1314         int type;
1315         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1316                 if (parent > 0)
1317                         type = BTRFS_SHARED_BLOCK_REF_KEY;
1318                 else
1319                         type = BTRFS_TREE_BLOCK_REF_KEY;
1320         } else {
1321                 if (parent > 0)
1322                         type = BTRFS_SHARED_DATA_REF_KEY;
1323                 else
1324                         type = BTRFS_EXTENT_DATA_REF_KEY;
1325         }
1326         return type;
1327 }
1328
1329 static int find_next_key(struct btrfs_path *path, int level,
1330                          struct btrfs_key *key)
1331
1332 {
1333         for (; level < BTRFS_MAX_LEVEL; level++) {
1334                 if (!path->nodes[level])
1335                         break;
1336                 if (path->slots[level] + 1 >=
1337                     btrfs_header_nritems(path->nodes[level]))
1338                         continue;
1339                 if (level == 0)
1340                         btrfs_item_key_to_cpu(path->nodes[level], key,
1341                                               path->slots[level] + 1);
1342                 else
1343                         btrfs_node_key_to_cpu(path->nodes[level], key,
1344                                               path->slots[level] + 1);
1345                 return 0;
1346         }
1347         return 1;
1348 }
1349
1350 /*
1351  * look for inline back ref. if back ref is found, *ref_ret is set
1352  * to the address of inline back ref, and 0 is returned.
1353  *
1354  * if back ref isn't found, *ref_ret is set to the address where it
1355  * should be inserted, and -ENOENT is returned.
1356  *
1357  * if insert is true and there are too many inline back refs, the path
1358  * points to the extent item, and -EAGAIN is returned.
1359  *
1360  * NOTE: inline back refs are ordered in the same way that back ref
1361  *       items in the tree are ordered.
1362  */
1363 static noinline_for_stack
1364 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1365                                  struct btrfs_root *root,
1366                                  struct btrfs_path *path,
1367                                  struct btrfs_extent_inline_ref **ref_ret,
1368                                  u64 bytenr, u64 num_bytes,
1369                                  u64 parent, u64 root_objectid,
1370                                  u64 owner, u64 offset, int insert)
1371 {
1372         struct btrfs_key key;
1373         struct extent_buffer *leaf;
1374         struct btrfs_extent_item *ei;
1375         struct btrfs_extent_inline_ref *iref;
1376         u64 flags;
1377         u64 item_size;
1378         unsigned long ptr;
1379         unsigned long end;
1380         int extra_size;
1381         int type;
1382         int want;
1383         int ret;
1384         int err = 0;
1385
1386         key.objectid = bytenr;
1387         key.type = BTRFS_EXTENT_ITEM_KEY;
1388         key.offset = num_bytes;
1389
1390         want = extent_ref_type(parent, owner);
1391         if (insert) {
1392                 extra_size = btrfs_extent_inline_ref_size(want);
1393                 path->keep_locks = 1;
1394         } else
1395                 extra_size = -1;
1396         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1397         if (ret < 0) {
1398                 err = ret;
1399                 goto out;
1400         }
1401         BUG_ON(ret);
1402
1403         leaf = path->nodes[0];
1404         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1405 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1406         if (item_size < sizeof(*ei)) {
1407                 if (!insert) {
1408                         err = -ENOENT;
1409                         goto out;
1410                 }
1411                 ret = convert_extent_item_v0(trans, root, path, owner,
1412                                              extra_size);
1413                 if (ret < 0) {
1414                         err = ret;
1415                         goto out;
1416                 }
1417                 leaf = path->nodes[0];
1418                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1419         }
1420 #endif
1421         BUG_ON(item_size < sizeof(*ei));
1422
1423         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1424         flags = btrfs_extent_flags(leaf, ei);
1425
1426         ptr = (unsigned long)(ei + 1);
1427         end = (unsigned long)ei + item_size;
1428
1429         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1430                 ptr += sizeof(struct btrfs_tree_block_info);
1431                 BUG_ON(ptr > end);
1432         } else {
1433                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1434         }
1435
1436         err = -ENOENT;
1437         while (1) {
1438                 if (ptr >= end) {
1439                         WARN_ON(ptr > end);
1440                         break;
1441                 }
1442                 iref = (struct btrfs_extent_inline_ref *)ptr;
1443                 type = btrfs_extent_inline_ref_type(leaf, iref);
1444                 if (want < type)
1445                         break;
1446                 if (want > type) {
1447                         ptr += btrfs_extent_inline_ref_size(type);
1448                         continue;
1449                 }
1450
1451                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1452                         struct btrfs_extent_data_ref *dref;
1453                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1454                         if (match_extent_data_ref(leaf, dref, root_objectid,
1455                                                   owner, offset)) {
1456                                 err = 0;
1457                                 break;
1458                         }
1459                         if (hash_extent_data_ref_item(leaf, dref) <
1460                             hash_extent_data_ref(root_objectid, owner, offset))
1461                                 break;
1462                 } else {
1463                         u64 ref_offset;
1464                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1465                         if (parent > 0) {
1466                                 if (parent == ref_offset) {
1467                                         err = 0;
1468                                         break;
1469                                 }
1470                                 if (ref_offset < parent)
1471                                         break;
1472                         } else {
1473                                 if (root_objectid == ref_offset) {
1474                                         err = 0;
1475                                         break;
1476                                 }
1477                                 if (ref_offset < root_objectid)
1478                                         break;
1479                         }
1480                 }
1481                 ptr += btrfs_extent_inline_ref_size(type);
1482         }
1483         if (err == -ENOENT && insert) {
1484                 if (item_size + extra_size >=
1485                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1486                         err = -EAGAIN;
1487                         goto out;
1488                 }
1489                 /*
1490                  * To add new inline back ref, we have to make sure
1491                  * there is no corresponding back ref item.
1492                  * For simplicity, we just do not add new inline back
1493                  * ref if there is any kind of item for this block
1494                  */
1495                 if (find_next_key(path, 0, &key) == 0 &&
1496                     key.objectid == bytenr &&
1497                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1498                         err = -EAGAIN;
1499                         goto out;
1500                 }
1501         }
1502         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1503 out:
1504         if (insert) {
1505                 path->keep_locks = 0;
1506                 btrfs_unlock_up_safe(path, 1);
1507         }
1508         return err;
1509 }
1510
1511 /*
1512  * helper to add new inline back ref
1513  */
1514 static noinline_for_stack
1515 int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1516                                 struct btrfs_root *root,
1517                                 struct btrfs_path *path,
1518                                 struct btrfs_extent_inline_ref *iref,
1519                                 u64 parent, u64 root_objectid,
1520                                 u64 owner, u64 offset, int refs_to_add,
1521                                 struct btrfs_delayed_extent_op *extent_op)
1522 {
1523         struct extent_buffer *leaf;
1524         struct btrfs_extent_item *ei;
1525         unsigned long ptr;
1526         unsigned long end;
1527         unsigned long item_offset;
1528         u64 refs;
1529         int size;
1530         int type;
1531         int ret;
1532
1533         leaf = path->nodes[0];
1534         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1535         item_offset = (unsigned long)iref - (unsigned long)ei;
1536
1537         type = extent_ref_type(parent, owner);
1538         size = btrfs_extent_inline_ref_size(type);
1539
1540         ret = btrfs_extend_item(trans, root, path, size);
1541         BUG_ON(ret);
1542
1543         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1544         refs = btrfs_extent_refs(leaf, ei);
1545         refs += refs_to_add;
1546         btrfs_set_extent_refs(leaf, ei, refs);
1547         if (extent_op)
1548                 __run_delayed_extent_op(extent_op, leaf, ei);
1549
1550         ptr = (unsigned long)ei + item_offset;
1551         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1552         if (ptr < end - size)
1553                 memmove_extent_buffer(leaf, ptr + size, ptr,
1554                                       end - size - ptr);
1555
1556         iref = (struct btrfs_extent_inline_ref *)ptr;
1557         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1558         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1559                 struct btrfs_extent_data_ref *dref;
1560                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1561                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1562                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1563                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1564                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1565         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1566                 struct btrfs_shared_data_ref *sref;
1567                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1568                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1569                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1570         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1571                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1572         } else {
1573                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1574         }
1575         btrfs_mark_buffer_dirty(leaf);
1576         return 0;
1577 }
1578
1579 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1580                                  struct btrfs_root *root,
1581                                  struct btrfs_path *path,
1582                                  struct btrfs_extent_inline_ref **ref_ret,
1583                                  u64 bytenr, u64 num_bytes, u64 parent,
1584                                  u64 root_objectid, u64 owner, u64 offset)
1585 {
1586         int ret;
1587
1588         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1589                                            bytenr, num_bytes, parent,
1590                                            root_objectid, owner, offset, 0);
1591         if (ret != -ENOENT)
1592                 return ret;
1593
1594         btrfs_release_path(root, path);
1595         *ref_ret = NULL;
1596
1597         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1598                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1599                                             root_objectid);
1600         } else {
1601                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1602                                              root_objectid, owner, offset);
1603         }
1604         return ret;
1605 }
1606
1607 /*
1608  * helper to update/remove inline back ref
1609  */
1610 static noinline_for_stack
1611 int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1612                                  struct btrfs_root *root,
1613                                  struct btrfs_path *path,
1614                                  struct btrfs_extent_inline_ref *iref,
1615                                  int refs_to_mod,
1616                                  struct btrfs_delayed_extent_op *extent_op)
1617 {
1618         struct extent_buffer *leaf;
1619         struct btrfs_extent_item *ei;
1620         struct btrfs_extent_data_ref *dref = NULL;
1621         struct btrfs_shared_data_ref *sref = NULL;
1622         unsigned long ptr;
1623         unsigned long end;
1624         u32 item_size;
1625         int size;
1626         int type;
1627         int ret;
1628         u64 refs;
1629
1630         leaf = path->nodes[0];
1631         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1632         refs = btrfs_extent_refs(leaf, ei);
1633         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1634         refs += refs_to_mod;
1635         btrfs_set_extent_refs(leaf, ei, refs);
1636         if (extent_op)
1637                 __run_delayed_extent_op(extent_op, leaf, ei);
1638
1639         type = btrfs_extent_inline_ref_type(leaf, iref);
1640
1641         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1642                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1643                 refs = btrfs_extent_data_ref_count(leaf, dref);
1644         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1645                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1646                 refs = btrfs_shared_data_ref_count(leaf, sref);
1647         } else {
1648                 refs = 1;
1649                 BUG_ON(refs_to_mod != -1);
1650         }
1651
1652         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1653         refs += refs_to_mod;
1654
1655         if (refs > 0) {
1656                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1657                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1658                 else
1659                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1660         } else {
1661                 size =  btrfs_extent_inline_ref_size(type);
1662                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1663                 ptr = (unsigned long)iref;
1664                 end = (unsigned long)ei + item_size;
1665                 if (ptr + size < end)
1666                         memmove_extent_buffer(leaf, ptr, ptr + size,
1667                                               end - ptr - size);
1668                 item_size -= size;
1669                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1670                 BUG_ON(ret);
1671         }
1672         btrfs_mark_buffer_dirty(leaf);
1673         return 0;
1674 }
1675
1676 static noinline_for_stack
1677 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1678                                  struct btrfs_root *root,
1679                                  struct btrfs_path *path,
1680                                  u64 bytenr, u64 num_bytes, u64 parent,
1681                                  u64 root_objectid, u64 owner,
1682                                  u64 offset, int refs_to_add,
1683                                  struct btrfs_delayed_extent_op *extent_op)
1684 {
1685         struct btrfs_extent_inline_ref *iref;
1686         int ret;
1687
1688         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1689                                            bytenr, num_bytes, parent,
1690                                            root_objectid, owner, offset, 1);
1691         if (ret == 0) {
1692                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1693                 ret = update_inline_extent_backref(trans, root, path, iref,
1694                                                    refs_to_add, extent_op);
1695         } else if (ret == -ENOENT) {
1696                 ret = setup_inline_extent_backref(trans, root, path, iref,
1697                                                   parent, root_objectid,
1698                                                   owner, offset, refs_to_add,
1699                                                   extent_op);
1700         }
1701         return ret;
1702 }
1703
1704 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1705                                  struct btrfs_root *root,
1706                                  struct btrfs_path *path,
1707                                  u64 bytenr, u64 parent, u64 root_objectid,
1708                                  u64 owner, u64 offset, int refs_to_add)
1709 {
1710         int ret;
1711         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1712                 BUG_ON(refs_to_add != 1);
1713                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1714                                             parent, root_objectid);
1715         } else {
1716                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1717                                              parent, root_objectid,
1718                                              owner, offset, refs_to_add);
1719         }
1720         return ret;
1721 }
1722
1723 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1724                                  struct btrfs_root *root,
1725                                  struct btrfs_path *path,
1726                                  struct btrfs_extent_inline_ref *iref,
1727                                  int refs_to_drop, int is_data)
1728 {
1729         int ret;
1730
1731         BUG_ON(!is_data && refs_to_drop != 1);
1732         if (iref) {
1733                 ret = update_inline_extent_backref(trans, root, path, iref,
1734                                                    -refs_to_drop, NULL);
1735         } else if (is_data) {
1736                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1737         } else {
1738                 ret = btrfs_del_item(trans, root, path);
1739         }
1740         return ret;
1741 }
1742
1743 static void btrfs_issue_discard(struct block_device *bdev,
1744                                 u64 start, u64 len)
1745 {
1746         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL, 0);
1747 }
1748
1749 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1750                                 u64 num_bytes)
1751 {
1752         int ret;
1753         u64 map_length = num_bytes;
1754         struct btrfs_multi_bio *multi = NULL;
1755
1756         if (!btrfs_test_opt(root, DISCARD))
1757                 return 0;
1758
1759         /* Tell the block device(s) that the sectors can be discarded */
1760         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1761                               bytenr, &map_length, &multi, 0);
1762         if (!ret) {
1763                 struct btrfs_bio_stripe *stripe = multi->stripes;
1764                 int i;
1765
1766                 if (map_length > num_bytes)
1767                         map_length = num_bytes;
1768
1769                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
1770                         btrfs_issue_discard(stripe->dev->bdev,
1771                                             stripe->physical,
1772                                             map_length);
1773                 }
1774                 kfree(multi);
1775         }
1776
1777         return ret;
1778 }
1779
1780 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1781                          struct btrfs_root *root,
1782                          u64 bytenr, u64 num_bytes, u64 parent,
1783                          u64 root_objectid, u64 owner, u64 offset)
1784 {
1785         int ret;
1786         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1787                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1788
1789         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1790                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1791                                         parent, root_objectid, (int)owner,
1792                                         BTRFS_ADD_DELAYED_REF, NULL);
1793         } else {
1794                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1795                                         parent, root_objectid, owner, offset,
1796                                         BTRFS_ADD_DELAYED_REF, NULL);
1797         }
1798         return ret;
1799 }
1800
1801 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1802                                   struct btrfs_root *root,
1803                                   u64 bytenr, u64 num_bytes,
1804                                   u64 parent, u64 root_objectid,
1805                                   u64 owner, u64 offset, int refs_to_add,
1806                                   struct btrfs_delayed_extent_op *extent_op)
1807 {
1808         struct btrfs_path *path;
1809         struct extent_buffer *leaf;
1810         struct btrfs_extent_item *item;
1811         u64 refs;
1812         int ret;
1813         int err = 0;
1814
1815         path = btrfs_alloc_path();
1816         if (!path)
1817                 return -ENOMEM;
1818
1819         path->reada = 1;
1820         path->leave_spinning = 1;
1821         /* this will setup the path even if it fails to insert the back ref */
1822         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1823                                            path, bytenr, num_bytes, parent,
1824                                            root_objectid, owner, offset,
1825                                            refs_to_add, extent_op);
1826         if (ret == 0)
1827                 goto out;
1828
1829         if (ret != -EAGAIN) {
1830                 err = ret;
1831                 goto out;
1832         }
1833
1834         leaf = path->nodes[0];
1835         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1836         refs = btrfs_extent_refs(leaf, item);
1837         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1838         if (extent_op)
1839                 __run_delayed_extent_op(extent_op, leaf, item);
1840
1841         btrfs_mark_buffer_dirty(leaf);
1842         btrfs_release_path(root->fs_info->extent_root, path);
1843
1844         path->reada = 1;
1845         path->leave_spinning = 1;
1846
1847         /* now insert the actual backref */
1848         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1849                                     path, bytenr, parent, root_objectid,
1850                                     owner, offset, refs_to_add);
1851         BUG_ON(ret);
1852 out:
1853         btrfs_free_path(path);
1854         return err;
1855 }
1856
1857 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1858                                 struct btrfs_root *root,
1859                                 struct btrfs_delayed_ref_node *node,
1860                                 struct btrfs_delayed_extent_op *extent_op,
1861                                 int insert_reserved)
1862 {
1863         int ret = 0;
1864         struct btrfs_delayed_data_ref *ref;
1865         struct btrfs_key ins;
1866         u64 parent = 0;
1867         u64 ref_root = 0;
1868         u64 flags = 0;
1869
1870         ins.objectid = node->bytenr;
1871         ins.offset = node->num_bytes;
1872         ins.type = BTRFS_EXTENT_ITEM_KEY;
1873
1874         ref = btrfs_delayed_node_to_data_ref(node);
1875         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1876                 parent = ref->parent;
1877         else
1878                 ref_root = ref->root;
1879
1880         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1881                 if (extent_op) {
1882                         BUG_ON(extent_op->update_key);
1883                         flags |= extent_op->flags_to_set;
1884                 }
1885                 ret = alloc_reserved_file_extent(trans, root,
1886                                                  parent, ref_root, flags,
1887                                                  ref->objectid, ref->offset,
1888                                                  &ins, node->ref_mod);
1889         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1890                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1891                                              node->num_bytes, parent,
1892                                              ref_root, ref->objectid,
1893                                              ref->offset, node->ref_mod,
1894                                              extent_op);
1895         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1896                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1897                                           node->num_bytes, parent,
1898                                           ref_root, ref->objectid,
1899                                           ref->offset, node->ref_mod,
1900                                           extent_op);
1901         } else {
1902                 BUG();
1903         }
1904         return ret;
1905 }
1906
1907 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1908                                     struct extent_buffer *leaf,
1909                                     struct btrfs_extent_item *ei)
1910 {
1911         u64 flags = btrfs_extent_flags(leaf, ei);
1912         if (extent_op->update_flags) {
1913                 flags |= extent_op->flags_to_set;
1914                 btrfs_set_extent_flags(leaf, ei, flags);
1915         }
1916
1917         if (extent_op->update_key) {
1918                 struct btrfs_tree_block_info *bi;
1919                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1920                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1921                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1922         }
1923 }
1924
1925 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1926                                  struct btrfs_root *root,
1927                                  struct btrfs_delayed_ref_node *node,
1928                                  struct btrfs_delayed_extent_op *extent_op)
1929 {
1930         struct btrfs_key key;
1931         struct btrfs_path *path;
1932         struct btrfs_extent_item *ei;
1933         struct extent_buffer *leaf;
1934         u32 item_size;
1935         int ret;
1936         int err = 0;
1937
1938         path = btrfs_alloc_path();
1939         if (!path)
1940                 return -ENOMEM;
1941
1942         key.objectid = node->bytenr;
1943         key.type = BTRFS_EXTENT_ITEM_KEY;
1944         key.offset = node->num_bytes;
1945
1946         path->reada = 1;
1947         path->leave_spinning = 1;
1948         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1949                                 path, 0, 1);
1950         if (ret < 0) {
1951                 err = ret;
1952                 goto out;
1953         }
1954         if (ret > 0) {
1955                 err = -EIO;
1956                 goto out;
1957         }
1958
1959         leaf = path->nodes[0];
1960         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1961 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1962         if (item_size < sizeof(*ei)) {
1963                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1964                                              path, (u64)-1, 0);
1965                 if (ret < 0) {
1966                         err = ret;
1967                         goto out;
1968                 }
1969                 leaf = path->nodes[0];
1970                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1971         }
1972 #endif
1973         BUG_ON(item_size < sizeof(*ei));
1974         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1975         __run_delayed_extent_op(extent_op, leaf, ei);
1976
1977         btrfs_mark_buffer_dirty(leaf);
1978 out:
1979         btrfs_free_path(path);
1980         return err;
1981 }
1982
1983 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1984                                 struct btrfs_root *root,
1985                                 struct btrfs_delayed_ref_node *node,
1986                                 struct btrfs_delayed_extent_op *extent_op,
1987                                 int insert_reserved)
1988 {
1989         int ret = 0;
1990         struct btrfs_delayed_tree_ref *ref;
1991         struct btrfs_key ins;
1992         u64 parent = 0;
1993         u64 ref_root = 0;
1994
1995         ins.objectid = node->bytenr;
1996         ins.offset = node->num_bytes;
1997         ins.type = BTRFS_EXTENT_ITEM_KEY;
1998
1999         ref = btrfs_delayed_node_to_tree_ref(node);
2000         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2001                 parent = ref->parent;
2002         else
2003                 ref_root = ref->root;
2004
2005         BUG_ON(node->ref_mod != 1);
2006         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2007                 BUG_ON(!extent_op || !extent_op->update_flags ||
2008                        !extent_op->update_key);
2009                 ret = alloc_reserved_tree_block(trans, root,
2010                                                 parent, ref_root,
2011                                                 extent_op->flags_to_set,
2012                                                 &extent_op->key,
2013                                                 ref->level, &ins);
2014         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2015                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2016                                              node->num_bytes, parent, ref_root,
2017                                              ref->level, 0, 1, extent_op);
2018         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2019                 ret = __btrfs_free_extent(trans, root, node->bytenr,
2020                                           node->num_bytes, parent, ref_root,
2021                                           ref->level, 0, 1, extent_op);
2022         } else {
2023                 BUG();
2024         }
2025         return ret;
2026 }
2027
2028 /* helper function to actually process a single delayed ref entry */
2029 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2030                                struct btrfs_root *root,
2031                                struct btrfs_delayed_ref_node *node,
2032                                struct btrfs_delayed_extent_op *extent_op,
2033                                int insert_reserved)
2034 {
2035         int ret;
2036         if (btrfs_delayed_ref_is_head(node)) {
2037                 struct btrfs_delayed_ref_head *head;
2038                 /*
2039                  * we've hit the end of the chain and we were supposed
2040                  * to insert this extent into the tree.  But, it got
2041                  * deleted before we ever needed to insert it, so all
2042                  * we have to do is clean up the accounting
2043                  */
2044                 BUG_ON(extent_op);
2045                 head = btrfs_delayed_node_to_head(node);
2046                 if (insert_reserved) {
2047                         btrfs_pin_extent(root, node->bytenr,
2048                                          node->num_bytes, 1);
2049                         if (head->is_data) {
2050                                 ret = btrfs_del_csums(trans, root,
2051                                                       node->bytenr,
2052                                                       node->num_bytes);
2053                                 BUG_ON(ret);
2054                         }
2055                 }
2056                 mutex_unlock(&head->mutex);
2057                 return 0;
2058         }
2059
2060         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2061             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2062                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2063                                            insert_reserved);
2064         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2065                  node->type == BTRFS_SHARED_DATA_REF_KEY)
2066                 ret = run_delayed_data_ref(trans, root, node, extent_op,
2067                                            insert_reserved);
2068         else
2069                 BUG();
2070         return ret;
2071 }
2072
2073 static noinline struct btrfs_delayed_ref_node *
2074 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2075 {
2076         struct rb_node *node;
2077         struct btrfs_delayed_ref_node *ref;
2078         int action = BTRFS_ADD_DELAYED_REF;
2079 again:
2080         /*
2081          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2082          * this prevents ref count from going down to zero when
2083          * there still are pending delayed ref.
2084          */
2085         node = rb_prev(&head->node.rb_node);
2086         while (1) {
2087                 if (!node)
2088                         break;
2089                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2090                                 rb_node);
2091                 if (ref->bytenr != head->node.bytenr)
2092                         break;
2093                 if (ref->action == action)
2094                         return ref;
2095                 node = rb_prev(node);
2096         }
2097         if (action == BTRFS_ADD_DELAYED_REF) {
2098                 action = BTRFS_DROP_DELAYED_REF;
2099                 goto again;
2100         }
2101         return NULL;
2102 }
2103
2104 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
2105                                        struct btrfs_root *root,
2106                                        struct list_head *cluster)
2107 {
2108         struct btrfs_delayed_ref_root *delayed_refs;
2109         struct btrfs_delayed_ref_node *ref;
2110         struct btrfs_delayed_ref_head *locked_ref = NULL;
2111         struct btrfs_delayed_extent_op *extent_op;
2112         int ret;
2113         int count = 0;
2114         int must_insert_reserved = 0;
2115
2116         delayed_refs = &trans->transaction->delayed_refs;
2117         while (1) {
2118                 if (!locked_ref) {
2119                         /* pick a new head ref from the cluster list */
2120                         if (list_empty(cluster))
2121                                 break;
2122
2123                         locked_ref = list_entry(cluster->next,
2124                                      struct btrfs_delayed_ref_head, cluster);
2125
2126                         /* grab the lock that says we are going to process
2127                          * all the refs for this head */
2128                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
2129
2130                         /*
2131                          * we may have dropped the spin lock to get the head
2132                          * mutex lock, and that might have given someone else
2133                          * time to free the head.  If that's true, it has been
2134                          * removed from our list and we can move on.
2135                          */
2136                         if (ret == -EAGAIN) {
2137                                 locked_ref = NULL;
2138                                 count++;
2139                                 continue;
2140                         }
2141                 }
2142
2143                 /*
2144                  * record the must insert reserved flag before we
2145                  * drop the spin lock.
2146                  */
2147                 must_insert_reserved = locked_ref->must_insert_reserved;
2148                 locked_ref->must_insert_reserved = 0;
2149
2150                 extent_op = locked_ref->extent_op;
2151                 locked_ref->extent_op = NULL;
2152
2153                 /*
2154                  * locked_ref is the head node, so we have to go one
2155                  * node back for any delayed ref updates
2156                  */
2157                 ref = select_delayed_ref(locked_ref);
2158                 if (!ref) {
2159                         /* All delayed refs have been processed, Go ahead
2160                          * and send the head node to run_one_delayed_ref,
2161                          * so that any accounting fixes can happen
2162                          */
2163                         ref = &locked_ref->node;
2164
2165                         if (extent_op && must_insert_reserved) {
2166                                 kfree(extent_op);
2167                                 extent_op = NULL;
2168                         }
2169
2170                         if (extent_op) {
2171                                 spin_unlock(&delayed_refs->lock);
2172
2173                                 ret = run_delayed_extent_op(trans, root,
2174                                                             ref, extent_op);
2175                                 BUG_ON(ret);
2176                                 kfree(extent_op);
2177
2178                                 cond_resched();
2179                                 spin_lock(&delayed_refs->lock);
2180                                 continue;
2181                         }
2182
2183                         list_del_init(&locked_ref->cluster);
2184                         locked_ref = NULL;
2185                 }
2186
2187                 ref->in_tree = 0;
2188                 rb_erase(&ref->rb_node, &delayed_refs->root);
2189                 delayed_refs->num_entries--;
2190
2191                 spin_unlock(&delayed_refs->lock);
2192
2193                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2194                                           must_insert_reserved);
2195                 BUG_ON(ret);
2196
2197                 btrfs_put_delayed_ref(ref);
2198                 kfree(extent_op);
2199                 count++;
2200
2201                 cond_resched();
2202                 spin_lock(&delayed_refs->lock);
2203         }
2204         return count;
2205 }
2206
2207 /*
2208  * this starts processing the delayed reference count updates and
2209  * extent insertions we have queued up so far.  count can be
2210  * 0, which means to process everything in the tree at the start
2211  * of the run (but not newly added entries), or it can be some target
2212  * number you'd like to process.
2213  */
2214 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2215                            struct btrfs_root *root, unsigned long count)
2216 {
2217         struct rb_node *node;
2218         struct btrfs_delayed_ref_root *delayed_refs;
2219         struct btrfs_delayed_ref_node *ref;
2220         struct list_head cluster;
2221         int ret;
2222         int run_all = count == (unsigned long)-1;
2223         int run_most = 0;
2224
2225         if (root == root->fs_info->extent_root)
2226                 root = root->fs_info->tree_root;
2227
2228         delayed_refs = &trans->transaction->delayed_refs;
2229         INIT_LIST_HEAD(&cluster);
2230 again:
2231         spin_lock(&delayed_refs->lock);
2232         if (count == 0) {
2233                 count = delayed_refs->num_entries * 2;
2234                 run_most = 1;
2235         }
2236         while (1) {
2237                 if (!(run_all || run_most) &&
2238                     delayed_refs->num_heads_ready < 64)
2239                         break;
2240
2241                 /*
2242                  * go find something we can process in the rbtree.  We start at
2243                  * the beginning of the tree, and then build a cluster
2244                  * of refs to process starting at the first one we are able to
2245                  * lock
2246                  */
2247                 ret = btrfs_find_ref_cluster(trans, &cluster,
2248                                              delayed_refs->run_delayed_start);
2249                 if (ret)
2250                         break;
2251
2252                 ret = run_clustered_refs(trans, root, &cluster);
2253                 BUG_ON(ret < 0);
2254
2255                 count -= min_t(unsigned long, ret, count);
2256
2257                 if (count == 0)
2258                         break;
2259         }
2260
2261         if (run_all) {
2262                 node = rb_first(&delayed_refs->root);
2263                 if (!node)
2264                         goto out;
2265                 count = (unsigned long)-1;
2266
2267                 while (node) {
2268                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2269                                        rb_node);
2270                         if (btrfs_delayed_ref_is_head(ref)) {
2271                                 struct btrfs_delayed_ref_head *head;
2272
2273                                 head = btrfs_delayed_node_to_head(ref);
2274                                 atomic_inc(&ref->refs);
2275
2276                                 spin_unlock(&delayed_refs->lock);
2277                                 mutex_lock(&head->mutex);
2278                                 mutex_unlock(&head->mutex);
2279
2280                                 btrfs_put_delayed_ref(ref);
2281                                 cond_resched();
2282                                 goto again;
2283                         }
2284                         node = rb_next(node);
2285                 }
2286                 spin_unlock(&delayed_refs->lock);
2287                 schedule_timeout(1);
2288                 goto again;
2289         }
2290 out:
2291         spin_unlock(&delayed_refs->lock);
2292         return 0;
2293 }
2294
2295 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2296                                 struct btrfs_root *root,
2297                                 u64 bytenr, u64 num_bytes, u64 flags,
2298                                 int is_data)
2299 {
2300         struct btrfs_delayed_extent_op *extent_op;
2301         int ret;
2302
2303         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2304         if (!extent_op)
2305                 return -ENOMEM;
2306
2307         extent_op->flags_to_set = flags;
2308         extent_op->update_flags = 1;
2309         extent_op->update_key = 0;
2310         extent_op->is_data = is_data ? 1 : 0;
2311
2312         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2313         if (ret)
2314                 kfree(extent_op);
2315         return ret;
2316 }
2317
2318 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2319                                       struct btrfs_root *root,
2320                                       struct btrfs_path *path,
2321                                       u64 objectid, u64 offset, u64 bytenr)
2322 {
2323         struct btrfs_delayed_ref_head *head;
2324         struct btrfs_delayed_ref_node *ref;
2325         struct btrfs_delayed_data_ref *data_ref;
2326         struct btrfs_delayed_ref_root *delayed_refs;
2327         struct rb_node *node;
2328         int ret = 0;
2329
2330         ret = -ENOENT;
2331         delayed_refs = &trans->transaction->delayed_refs;
2332         spin_lock(&delayed_refs->lock);
2333         head = btrfs_find_delayed_ref_head(trans, bytenr);
2334         if (!head)
2335                 goto out;
2336
2337         if (!mutex_trylock(&head->mutex)) {
2338                 atomic_inc(&head->node.refs);
2339                 spin_unlock(&delayed_refs->lock);
2340
2341                 btrfs_release_path(root->fs_info->extent_root, path);
2342
2343                 mutex_lock(&head->mutex);
2344                 mutex_unlock(&head->mutex);
2345                 btrfs_put_delayed_ref(&head->node);
2346                 return -EAGAIN;
2347         }
2348
2349         node = rb_prev(&head->node.rb_node);
2350         if (!node)
2351                 goto out_unlock;
2352
2353         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2354
2355         if (ref->bytenr != bytenr)
2356                 goto out_unlock;
2357
2358         ret = 1;
2359         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2360                 goto out_unlock;
2361
2362         data_ref = btrfs_delayed_node_to_data_ref(ref);
2363
2364         node = rb_prev(node);
2365         if (node) {
2366                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2367                 if (ref->bytenr == bytenr)
2368                         goto out_unlock;
2369         }
2370
2371         if (data_ref->root != root->root_key.objectid ||
2372             data_ref->objectid != objectid || data_ref->offset != offset)
2373                 goto out_unlock;
2374
2375         ret = 0;
2376 out_unlock:
2377         mutex_unlock(&head->mutex);
2378 out:
2379         spin_unlock(&delayed_refs->lock);
2380         return ret;
2381 }
2382
2383 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2384                                         struct btrfs_root *root,
2385                                         struct btrfs_path *path,
2386                                         u64 objectid, u64 offset, u64 bytenr)
2387 {
2388         struct btrfs_root *extent_root = root->fs_info->extent_root;
2389         struct extent_buffer *leaf;
2390         struct btrfs_extent_data_ref *ref;
2391         struct btrfs_extent_inline_ref *iref;
2392         struct btrfs_extent_item *ei;
2393         struct btrfs_key key;
2394         u32 item_size;
2395         int ret;
2396
2397         key.objectid = bytenr;
2398         key.offset = (u64)-1;
2399         key.type = BTRFS_EXTENT_ITEM_KEY;
2400
2401         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2402         if (ret < 0)
2403                 goto out;
2404         BUG_ON(ret == 0);
2405
2406         ret = -ENOENT;
2407         if (path->slots[0] == 0)
2408                 goto out;
2409
2410         path->slots[0]--;
2411         leaf = path->nodes[0];
2412         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2413
2414         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2415                 goto out;
2416
2417         ret = 1;
2418         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2419 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2420         if (item_size < sizeof(*ei)) {
2421                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2422                 goto out;
2423         }
2424 #endif
2425         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2426
2427         if (item_size != sizeof(*ei) +
2428             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2429                 goto out;
2430
2431         if (btrfs_extent_generation(leaf, ei) <=
2432             btrfs_root_last_snapshot(&root->root_item))
2433                 goto out;
2434
2435         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2436         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2437             BTRFS_EXTENT_DATA_REF_KEY)
2438                 goto out;
2439
2440         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2441         if (btrfs_extent_refs(leaf, ei) !=
2442             btrfs_extent_data_ref_count(leaf, ref) ||
2443             btrfs_extent_data_ref_root(leaf, ref) !=
2444             root->root_key.objectid ||
2445             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2446             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2447                 goto out;
2448
2449         ret = 0;
2450 out:
2451         return ret;
2452 }
2453
2454 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2455                           struct btrfs_root *root,
2456                           u64 objectid, u64 offset, u64 bytenr)
2457 {
2458         struct btrfs_path *path;
2459         int ret;
2460         int ret2;
2461
2462         path = btrfs_alloc_path();
2463         if (!path)
2464                 return -ENOENT;
2465
2466         do {
2467                 ret = check_committed_ref(trans, root, path, objectid,
2468                                           offset, bytenr);
2469                 if (ret && ret != -ENOENT)
2470                         goto out;
2471
2472                 ret2 = check_delayed_ref(trans, root, path, objectid,
2473                                          offset, bytenr);
2474         } while (ret2 == -EAGAIN);
2475
2476         if (ret2 && ret2 != -ENOENT) {
2477                 ret = ret2;
2478                 goto out;
2479         }
2480
2481         if (ret != -ENOENT || ret2 != -ENOENT)
2482                 ret = 0;
2483 out:
2484         btrfs_free_path(path);
2485         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2486                 WARN_ON(ret > 0);
2487         return ret;
2488 }
2489
2490 #if 0
2491 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2492                     struct extent_buffer *buf, u32 nr_extents)
2493 {
2494         struct btrfs_key key;
2495         struct btrfs_file_extent_item *fi;
2496         u64 root_gen;
2497         u32 nritems;
2498         int i;
2499         int level;
2500         int ret = 0;
2501         int shared = 0;
2502
2503         if (!root->ref_cows)
2504                 return 0;
2505
2506         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2507                 shared = 0;
2508                 root_gen = root->root_key.offset;
2509         } else {
2510                 shared = 1;
2511                 root_gen = trans->transid - 1;
2512         }
2513
2514         level = btrfs_header_level(buf);
2515         nritems = btrfs_header_nritems(buf);
2516
2517         if (level == 0) {
2518                 struct btrfs_leaf_ref *ref;
2519                 struct btrfs_extent_info *info;
2520
2521                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2522                 if (!ref) {
2523                         ret = -ENOMEM;
2524                         goto out;
2525                 }
2526
2527                 ref->root_gen = root_gen;
2528                 ref->bytenr = buf->start;
2529                 ref->owner = btrfs_header_owner(buf);
2530                 ref->generation = btrfs_header_generation(buf);
2531                 ref->nritems = nr_extents;
2532                 info = ref->extents;
2533
2534                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2535                         u64 disk_bytenr;
2536                         btrfs_item_key_to_cpu(buf, &key, i);
2537                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2538                                 continue;
2539                         fi = btrfs_item_ptr(buf, i,
2540                                             struct btrfs_file_extent_item);
2541                         if (btrfs_file_extent_type(buf, fi) ==
2542                             BTRFS_FILE_EXTENT_INLINE)
2543                                 continue;
2544                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2545                         if (disk_bytenr == 0)
2546                                 continue;
2547
2548                         info->bytenr = disk_bytenr;
2549                         info->num_bytes =
2550                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2551                         info->objectid = key.objectid;
2552                         info->offset = key.offset;
2553                         info++;
2554                 }
2555
2556                 ret = btrfs_add_leaf_ref(root, ref, shared);
2557                 if (ret == -EEXIST && shared) {
2558                         struct btrfs_leaf_ref *old;
2559                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2560                         BUG_ON(!old);
2561                         btrfs_remove_leaf_ref(root, old);
2562                         btrfs_free_leaf_ref(root, old);
2563                         ret = btrfs_add_leaf_ref(root, ref, shared);
2564                 }
2565                 WARN_ON(ret);
2566                 btrfs_free_leaf_ref(root, ref);
2567         }
2568 out:
2569         return ret;
2570 }
2571
2572 /* when a block goes through cow, we update the reference counts of
2573  * everything that block points to.  The internal pointers of the block
2574  * can be in just about any order, and it is likely to have clusters of
2575  * things that are close together and clusters of things that are not.
2576  *
2577  * To help reduce the seeks that come with updating all of these reference
2578  * counts, sort them by byte number before actual updates are done.
2579  *
2580  * struct refsort is used to match byte number to slot in the btree block.
2581  * we sort based on the byte number and then use the slot to actually
2582  * find the item.
2583  *
2584  * struct refsort is smaller than strcut btrfs_item and smaller than
2585  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2586  * for a btree block, there's no way for a kmalloc of refsorts for a
2587  * single node to be bigger than a page.
2588  */
2589 struct refsort {
2590         u64 bytenr;
2591         u32 slot;
2592 };
2593
2594 /*
2595  * for passing into sort()
2596  */
2597 static int refsort_cmp(const void *a_void, const void *b_void)
2598 {
2599         const struct refsort *a = a_void;
2600         const struct refsort *b = b_void;
2601
2602         if (a->bytenr < b->bytenr)
2603                 return -1;
2604         if (a->bytenr > b->bytenr)
2605                 return 1;
2606         return 0;
2607 }
2608 #endif
2609
2610 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2611                            struct btrfs_root *root,
2612                            struct extent_buffer *buf,
2613                            int full_backref, int inc)
2614 {
2615         u64 bytenr;
2616         u64 num_bytes;
2617         u64 parent;
2618         u64 ref_root;
2619         u32 nritems;
2620         struct btrfs_key key;
2621         struct btrfs_file_extent_item *fi;
2622         int i;
2623         int level;
2624         int ret = 0;
2625         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2626                             u64, u64, u64, u64, u64, u64);
2627
2628         ref_root = btrfs_header_owner(buf);
2629         nritems = btrfs_header_nritems(buf);
2630         level = btrfs_header_level(buf);
2631
2632         if (!root->ref_cows && level == 0)
2633                 return 0;
2634
2635         if (inc)
2636                 process_func = btrfs_inc_extent_ref;
2637         else
2638                 process_func = btrfs_free_extent;
2639
2640         if (full_backref)
2641                 parent = buf->start;
2642         else
2643                 parent = 0;
2644
2645         for (i = 0; i < nritems; i++) {
2646                 if (level == 0) {
2647                         btrfs_item_key_to_cpu(buf, &key, i);
2648                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2649                                 continue;
2650                         fi = btrfs_item_ptr(buf, i,
2651                                             struct btrfs_file_extent_item);
2652                         if (btrfs_file_extent_type(buf, fi) ==
2653                             BTRFS_FILE_EXTENT_INLINE)
2654                                 continue;
2655                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2656                         if (bytenr == 0)
2657                                 continue;
2658
2659                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2660                         key.offset -= btrfs_file_extent_offset(buf, fi);
2661                         ret = process_func(trans, root, bytenr, num_bytes,
2662                                            parent, ref_root, key.objectid,
2663                                            key.offset);
2664                         if (ret)
2665                                 goto fail;
2666                 } else {
2667                         bytenr = btrfs_node_blockptr(buf, i);
2668                         num_bytes = btrfs_level_size(root, level - 1);
2669                         ret = process_func(trans, root, bytenr, num_bytes,
2670                                            parent, ref_root, level - 1, 0);
2671                         if (ret)
2672                                 goto fail;
2673                 }
2674         }
2675         return 0;
2676 fail:
2677         BUG();
2678         return ret;
2679 }
2680
2681 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2682                   struct extent_buffer *buf, int full_backref)
2683 {
2684         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2685 }
2686
2687 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2688                   struct extent_buffer *buf, int full_backref)
2689 {
2690         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2691 }
2692
2693 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2694                                  struct btrfs_root *root,
2695                                  struct btrfs_path *path,
2696                                  struct btrfs_block_group_cache *cache)
2697 {
2698         int ret;
2699         struct btrfs_root *extent_root = root->fs_info->extent_root;
2700         unsigned long bi;
2701         struct extent_buffer *leaf;
2702
2703         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2704         if (ret < 0)
2705                 goto fail;
2706         BUG_ON(ret);
2707
2708         leaf = path->nodes[0];
2709         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2710         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2711         btrfs_mark_buffer_dirty(leaf);
2712         btrfs_release_path(extent_root, path);
2713 fail:
2714         if (ret)
2715                 return ret;
2716         return 0;
2717
2718 }
2719
2720 static struct btrfs_block_group_cache *
2721 next_block_group(struct btrfs_root *root,
2722                  struct btrfs_block_group_cache *cache)
2723 {
2724         struct rb_node *node;
2725         spin_lock(&root->fs_info->block_group_cache_lock);
2726         node = rb_next(&cache->cache_node);
2727         btrfs_put_block_group(cache);
2728         if (node) {
2729                 cache = rb_entry(node, struct btrfs_block_group_cache,
2730                                  cache_node);
2731                 btrfs_get_block_group(cache);
2732         } else
2733                 cache = NULL;
2734         spin_unlock(&root->fs_info->block_group_cache_lock);
2735         return cache;
2736 }
2737
2738 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
2739                             struct btrfs_trans_handle *trans,
2740                             struct btrfs_path *path)
2741 {
2742         struct btrfs_root *root = block_group->fs_info->tree_root;
2743         struct inode *inode = NULL;
2744         u64 alloc_hint = 0;
2745         int dcs = BTRFS_DC_ERROR;
2746         int num_pages = 0;
2747         int retries = 0;
2748         int ret = 0;
2749
2750         /*
2751          * If this block group is smaller than 100 megs don't bother caching the
2752          * block group.
2753          */
2754         if (block_group->key.offset < (100 * 1024 * 1024)) {
2755                 spin_lock(&block_group->lock);
2756                 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
2757                 spin_unlock(&block_group->lock);
2758                 return 0;
2759         }
2760
2761 again:
2762         inode = lookup_free_space_inode(root, block_group, path);
2763         if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
2764                 ret = PTR_ERR(inode);
2765                 btrfs_release_path(root, path);
2766                 goto out;
2767         }
2768
2769         if (IS_ERR(inode)) {
2770                 BUG_ON(retries);
2771                 retries++;
2772
2773                 if (block_group->ro)
2774                         goto out_free;
2775
2776                 ret = create_free_space_inode(root, trans, block_group, path);
2777                 if (ret)
2778                         goto out_free;
2779                 goto again;
2780         }
2781
2782         /*
2783          * We want to set the generation to 0, that way if anything goes wrong
2784          * from here on out we know not to trust this cache when we load up next
2785          * time.
2786          */
2787         BTRFS_I(inode)->generation = 0;
2788         ret = btrfs_update_inode(trans, root, inode);
2789         WARN_ON(ret);
2790
2791         if (i_size_read(inode) > 0) {
2792                 ret = btrfs_truncate_free_space_cache(root, trans, path,
2793                                                       inode);
2794                 if (ret)
2795                         goto out_put;
2796         }
2797
2798         spin_lock(&block_group->lock);
2799         if (block_group->cached != BTRFS_CACHE_FINISHED) {
2800                 /* We're not cached, don't bother trying to write stuff out */
2801                 dcs = BTRFS_DC_WRITTEN;
2802                 spin_unlock(&block_group->lock);
2803                 goto out_put;
2804         }
2805         spin_unlock(&block_group->lock);
2806
2807         num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
2808         if (!num_pages)
2809                 num_pages = 1;
2810
2811         /*
2812          * Just to make absolutely sure we have enough space, we're going to
2813          * preallocate 12 pages worth of space for each block group.  In
2814          * practice we ought to use at most 8, but we need extra space so we can
2815          * add our header and have a terminator between the extents and the
2816          * bitmaps.
2817          */
2818         num_pages *= 16;
2819         num_pages *= PAGE_CACHE_SIZE;
2820
2821         ret = btrfs_check_data_free_space(inode, num_pages);
2822         if (ret)
2823                 goto out_put;
2824
2825         ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
2826                                               num_pages, num_pages,
2827                                               &alloc_hint);
2828         if (!ret)
2829                 dcs = BTRFS_DC_SETUP;
2830         btrfs_free_reserved_data_space(inode, num_pages);
2831 out_put:
2832         iput(inode);
2833 out_free:
2834         btrfs_release_path(root, path);
2835 out:
2836         spin_lock(&block_group->lock);
2837         block_group->disk_cache_state = dcs;
2838         spin_unlock(&block_group->lock);
2839
2840         return ret;
2841 }
2842
2843 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2844                                    struct btrfs_root *root)
2845 {
2846         struct btrfs_block_group_cache *cache;
2847         int err = 0;
2848         struct btrfs_path *path;
2849         u64 last = 0;
2850
2851         path = btrfs_alloc_path();
2852         if (!path)
2853                 return -ENOMEM;
2854
2855 again:
2856         while (1) {
2857                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2858                 while (cache) {
2859                         if (cache->disk_cache_state == BTRFS_DC_CLEAR)
2860                                 break;
2861                         cache = next_block_group(root, cache);
2862                 }
2863                 if (!cache) {
2864                         if (last == 0)
2865                                 break;
2866                         last = 0;
2867                         continue;
2868                 }
2869                 err = cache_save_setup(cache, trans, path);
2870                 last = cache->key.objectid + cache->key.offset;
2871                 btrfs_put_block_group(cache);
2872         }
2873
2874         while (1) {
2875                 if (last == 0) {
2876                         err = btrfs_run_delayed_refs(trans, root,
2877                                                      (unsigned long)-1);
2878                         BUG_ON(err);
2879                 }
2880
2881                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2882                 while (cache) {
2883                         if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
2884                                 btrfs_put_block_group(cache);
2885                                 goto again;
2886                         }
2887
2888                         if (cache->dirty)
2889                                 break;
2890                         cache = next_block_group(root, cache);
2891                 }
2892                 if (!cache) {
2893                         if (last == 0)
2894                                 break;
2895                         last = 0;
2896                         continue;
2897                 }
2898
2899                 if (cache->disk_cache_state == BTRFS_DC_SETUP)
2900                         cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
2901                 cache->dirty = 0;
2902                 last = cache->key.objectid + cache->key.offset;
2903
2904                 err = write_one_cache_group(trans, root, path, cache);
2905                 BUG_ON(err);
2906                 btrfs_put_block_group(cache);
2907         }
2908
2909         while (1) {
2910                 /*
2911                  * I don't think this is needed since we're just marking our
2912                  * preallocated extent as written, but just in case it can't
2913                  * hurt.
2914                  */
2915                 if (last == 0) {
2916                         err = btrfs_run_delayed_refs(trans, root,
2917                                                      (unsigned long)-1);
2918                         BUG_ON(err);
2919                 }
2920
2921                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2922                 while (cache) {
2923                         /*
2924                          * Really this shouldn't happen, but it could if we
2925                          * couldn't write the entire preallocated extent and
2926                          * splitting the extent resulted in a new block.
2927                          */
2928                         if (cache->dirty) {
2929                                 btrfs_put_block_group(cache);
2930                                 goto again;
2931                         }
2932                         if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2933                                 break;
2934                         cache = next_block_group(root, cache);
2935                 }
2936                 if (!cache) {
2937                         if (last == 0)
2938                                 break;
2939                         last = 0;
2940                         continue;
2941                 }
2942
2943                 btrfs_write_out_cache(root, trans, cache, path);
2944
2945                 /*
2946                  * If we didn't have an error then the cache state is still
2947                  * NEED_WRITE, so we can set it to WRITTEN.
2948                  */
2949                 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
2950                         cache->disk_cache_state = BTRFS_DC_WRITTEN;
2951                 last = cache->key.objectid + cache->key.offset;
2952                 btrfs_put_block_group(cache);
2953         }
2954
2955         btrfs_free_path(path);
2956         return 0;
2957 }
2958
2959 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2960 {
2961         struct btrfs_block_group_cache *block_group;
2962         int readonly = 0;
2963
2964         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2965         if (!block_group || block_group->ro)
2966                 readonly = 1;
2967         if (block_group)
2968                 btrfs_put_block_group(block_group);
2969         return readonly;
2970 }
2971
2972 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2973                              u64 total_bytes, u64 bytes_used,
2974                              struct btrfs_space_info **space_info)
2975 {
2976         struct btrfs_space_info *found;
2977         int i;
2978         int factor;
2979
2980         if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2981                      BTRFS_BLOCK_GROUP_RAID10))
2982                 factor = 2;
2983         else
2984                 factor = 1;
2985
2986         found = __find_space_info(info, flags);
2987         if (found) {
2988                 spin_lock(&found->lock);
2989                 found->total_bytes += total_bytes;
2990                 found->disk_total += total_bytes * factor;
2991                 found->bytes_used += bytes_used;
2992                 found->disk_used += bytes_used * factor;
2993                 found->full = 0;
2994                 spin_unlock(&found->lock);
2995                 *space_info = found;
2996                 return 0;
2997         }
2998         found = kzalloc(sizeof(*found), GFP_NOFS);
2999         if (!found)
3000                 return -ENOMEM;
3001
3002         for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3003                 INIT_LIST_HEAD(&found->block_groups[i]);
3004         init_rwsem(&found->groups_sem);
3005         spin_lock_init(&found->lock);
3006         found->flags = flags & (BTRFS_BLOCK_GROUP_DATA |
3007                                 BTRFS_BLOCK_GROUP_SYSTEM |
3008                                 BTRFS_BLOCK_GROUP_METADATA);
3009         found->total_bytes = total_bytes;
3010         found->disk_total = total_bytes * factor;
3011         found->bytes_used = bytes_used;
3012         found->disk_used = bytes_used * factor;
3013         found->bytes_pinned = 0;
3014         found->bytes_reserved = 0;
3015         found->bytes_readonly = 0;
3016         found->bytes_may_use = 0;
3017         found->full = 0;
3018         found->force_alloc = 0;
3019         *space_info = found;
3020         list_add_rcu(&found->list, &info->space_info);
3021         atomic_set(&found->caching_threads, 0);
3022         return 0;
3023 }
3024
3025 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3026 {
3027         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
3028                                    BTRFS_BLOCK_GROUP_RAID1 |
3029                                    BTRFS_BLOCK_GROUP_RAID10 |
3030                                    BTRFS_BLOCK_GROUP_DUP);
3031         if (extra_flags) {
3032                 if (flags & BTRFS_BLOCK_GROUP_DATA)
3033                         fs_info->avail_data_alloc_bits |= extra_flags;
3034                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3035                         fs_info->avail_metadata_alloc_bits |= extra_flags;
3036                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3037                         fs_info->avail_system_alloc_bits |= extra_flags;
3038         }
3039 }
3040
3041 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3042 {
3043         /*
3044          * we add in the count of missing devices because we want
3045          * to make sure that any RAID levels on a degraded FS
3046          * continue to be honored.
3047          */
3048         u64 num_devices = root->fs_info->fs_devices->rw_devices +
3049                 root->fs_info->fs_devices->missing_devices;
3050
3051         if (num_devices == 1)
3052                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
3053         if (num_devices < 4)
3054                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3055
3056         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
3057             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3058                       BTRFS_BLOCK_GROUP_RAID10))) {
3059                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
3060         }
3061
3062         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
3063             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
3064                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
3065         }
3066
3067         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
3068             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
3069              (flags & BTRFS_BLOCK_GROUP_RAID10) |
3070              (flags & BTRFS_BLOCK_GROUP_DUP)))
3071                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
3072         return flags;
3073 }
3074
3075 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
3076 {
3077         if (flags & BTRFS_BLOCK_GROUP_DATA)
3078                 flags |= root->fs_info->avail_data_alloc_bits &
3079                          root->fs_info->data_alloc_profile;
3080         else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3081                 flags |= root->fs_info->avail_system_alloc_bits &
3082                          root->fs_info->system_alloc_profile;
3083         else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3084                 flags |= root->fs_info->avail_metadata_alloc_bits &
3085                          root->fs_info->metadata_alloc_profile;
3086         return btrfs_reduce_alloc_profile(root, flags);
3087 }
3088
3089 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3090 {
3091         u64 flags;
3092
3093         if (data)
3094                 flags = BTRFS_BLOCK_GROUP_DATA;
3095         else if (root == root->fs_info->chunk_root)
3096                 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3097         else
3098                 flags = BTRFS_BLOCK_GROUP_METADATA;
3099
3100         return get_alloc_profile(root, flags);
3101 }
3102
3103 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
3104 {
3105         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
3106                                                        BTRFS_BLOCK_GROUP_DATA);
3107 }
3108
3109 /*
3110  * This will check the space that the inode allocates from to make sure we have
3111  * enough space for bytes.
3112  */
3113 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3114 {
3115         struct btrfs_space_info *data_sinfo;
3116         struct btrfs_root *root = BTRFS_I(inode)->root;
3117         u64 used;
3118         int ret = 0, committed = 0, alloc_chunk = 1;
3119
3120         /* make sure bytes are sectorsize aligned */
3121         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3122
3123         if (root == root->fs_info->tree_root) {
3124                 alloc_chunk = 0;
3125                 committed = 1;
3126         }
3127
3128         data_sinfo = BTRFS_I(inode)->space_info;
3129         if (!data_sinfo)
3130                 goto alloc;
3131
3132 again:
3133         /* make sure we have enough space to handle the data first */
3134         spin_lock(&data_sinfo->lock);
3135         used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3136                 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3137                 data_sinfo->bytes_may_use;
3138
3139         if (used + bytes > data_sinfo->total_bytes) {
3140                 struct btrfs_trans_handle *trans;
3141
3142                 /*
3143                  * if we don't have enough free bytes in this space then we need
3144                  * to alloc a new chunk.
3145                  */
3146                 if (!data_sinfo->full && alloc_chunk) {
3147                         u64 alloc_target;
3148
3149                         data_sinfo->force_alloc = 1;
3150                         spin_unlock(&data_sinfo->lock);
3151 alloc:
3152                         alloc_target = btrfs_get_alloc_profile(root, 1);
3153                         trans = btrfs_join_transaction(root, 1);
3154                         if (IS_ERR(trans))
3155                                 return PTR_ERR(trans);
3156
3157                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3158                                              bytes + 2 * 1024 * 1024,
3159                                              alloc_target, 0);
3160                         btrfs_end_transaction(trans, root);
3161                         if (ret < 0) {
3162                                 if (ret != -ENOSPC)
3163                                         return ret;
3164                                 else
3165                                         goto commit_trans;
3166                         }
3167
3168                         if (!data_sinfo) {
3169                                 btrfs_set_inode_space_info(root, inode);
3170                                 data_sinfo = BTRFS_I(inode)->space_info;
3171                         }
3172                         goto again;
3173                 }
3174                 spin_unlock(&data_sinfo->lock);
3175
3176                 /* commit the current transaction and try again */
3177 commit_trans:
3178                 if (!committed && !root->fs_info->open_ioctl_trans) {
3179                         committed = 1;
3180                         trans = btrfs_join_transaction(root, 1);
3181                         if (IS_ERR(trans))
3182                                 return PTR_ERR(trans);
3183                         ret = btrfs_commit_transaction(trans, root);
3184                         if (ret)
3185                                 return ret;
3186                         goto again;
3187                 }
3188
3189 #if 0 /* I hope we never need this code again, just in case */
3190                 printk(KERN_ERR "no space left, need %llu, %llu bytes_used, "
3191                        "%llu bytes_reserved, " "%llu bytes_pinned, "
3192                        "%llu bytes_readonly, %llu may use %llu total\n",
3193                        (unsigned long long)bytes,
3194                        (unsigned long long)data_sinfo->bytes_used,
3195                        (unsigned long long)data_sinfo->bytes_reserved,
3196                        (unsigned long long)data_sinfo->bytes_pinned,
3197                        (unsigned long long)data_sinfo->bytes_readonly,
3198                        (unsigned long long)data_sinfo->bytes_may_use,
3199                        (unsigned long long)data_sinfo->total_bytes);
3200 #endif
3201                 return -ENOSPC;
3202         }
3203         data_sinfo->bytes_may_use += bytes;
3204         BTRFS_I(inode)->reserved_bytes += bytes;
3205         spin_unlock(&data_sinfo->lock);
3206
3207         return 0;
3208 }
3209
3210 /*
3211  * called when we are clearing an delalloc extent from the
3212  * inode's io_tree or there was an error for whatever reason
3213  * after calling btrfs_check_data_free_space
3214  */
3215 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3216 {
3217         struct btrfs_root *root = BTRFS_I(inode)->root;
3218         struct btrfs_space_info *data_sinfo;
3219
3220         /* make sure bytes are sectorsize aligned */
3221         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3222
3223         data_sinfo = BTRFS_I(inode)->space_info;
3224         spin_lock(&data_sinfo->lock);
3225         data_sinfo->bytes_may_use -= bytes;
3226         BTRFS_I(inode)->reserved_bytes -= bytes;
3227         spin_unlock(&data_sinfo->lock);
3228 }
3229
3230 static void force_metadata_allocation(struct btrfs_fs_info *info)
3231 {
3232         struct list_head *head = &info->space_info;
3233         struct btrfs_space_info *found;
3234
3235         rcu_read_lock();
3236         list_for_each_entry_rcu(found, head, list) {
3237                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3238                         found->force_alloc = 1;
3239         }
3240         rcu_read_unlock();
3241 }
3242
3243 static int should_alloc_chunk(struct btrfs_root *root,
3244                               struct btrfs_space_info *sinfo, u64 alloc_bytes)
3245 {
3246         u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3247         u64 thresh;
3248
3249         if (sinfo->bytes_used + sinfo->bytes_reserved +
3250             alloc_bytes + 256 * 1024 * 1024 < num_bytes)
3251                 return 0;
3252
3253         if (sinfo->bytes_used + sinfo->bytes_reserved +
3254             alloc_bytes < div_factor(num_bytes, 8))
3255                 return 0;
3256
3257         thresh = btrfs_super_total_bytes(&root->fs_info->super_copy);
3258         thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5));
3259
3260         if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3))
3261                 return 0;
3262
3263         return 1;
3264 }
3265
3266 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3267                           struct btrfs_root *extent_root, u64 alloc_bytes,
3268                           u64 flags, int force)
3269 {
3270         struct btrfs_space_info *space_info;
3271         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3272         int ret = 0;
3273
3274         mutex_lock(&fs_info->chunk_mutex);
3275
3276         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3277
3278         space_info = __find_space_info(extent_root->fs_info, flags);
3279         if (!space_info) {
3280                 ret = update_space_info(extent_root->fs_info, flags,
3281                                         0, 0, &space_info);
3282                 BUG_ON(ret);
3283         }
3284         BUG_ON(!space_info);
3285
3286         spin_lock(&space_info->lock);
3287         if (space_info->force_alloc)
3288                 force = 1;
3289         if (space_info->full) {
3290                 spin_unlock(&space_info->lock);
3291                 goto out;
3292         }
3293
3294         if (!force && !should_alloc_chunk(extent_root, space_info,
3295                                           alloc_bytes)) {
3296                 spin_unlock(&space_info->lock);
3297                 goto out;
3298         }
3299         spin_unlock(&space_info->lock);
3300
3301         /*
3302          * If we have mixed data/metadata chunks we want to make sure we keep
3303          * allocating mixed chunks instead of individual chunks.
3304          */
3305         if (btrfs_mixed_space_info(space_info))
3306                 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3307
3308         /*
3309          * if we're doing a data chunk, go ahead and make sure that
3310          * we keep a reasonable number of metadata chunks allocated in the
3311          * FS as well.
3312          */
3313         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3314                 fs_info->data_chunk_allocations++;
3315                 if (!(fs_info->data_chunk_allocations %
3316                       fs_info->metadata_ratio))
3317                         force_metadata_allocation(fs_info);
3318         }
3319
3320         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3321         spin_lock(&space_info->lock);
3322         if (ret)
3323                 space_info->full = 1;
3324         else
3325                 ret = 1;
3326         space_info->force_alloc = 0;
3327         spin_unlock(&space_info->lock);
3328 out:
3329         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3330         return ret;
3331 }
3332
3333 /*
3334  * shrink metadata reservation for delalloc
3335  */
3336 static int shrink_delalloc(struct btrfs_trans_handle *trans,
3337                            struct btrfs_root *root, u64 to_reclaim, int sync)
3338 {
3339         struct btrfs_block_rsv *block_rsv;
3340         struct btrfs_space_info *space_info;
3341         u64 reserved;
3342         u64 max_reclaim;
3343         u64 reclaimed = 0;
3344         long time_left;
3345         int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT;
3346         int loops = 0;
3347         unsigned long progress;
3348
3349         block_rsv = &root->fs_info->delalloc_block_rsv;
3350         space_info = block_rsv->space_info;
3351
3352         smp_mb();
3353         reserved = space_info->bytes_reserved;
3354         progress = space_info->reservation_progress;
3355
3356         if (reserved == 0)
3357                 return 0;
3358
3359         max_reclaim = min(reserved, to_reclaim);
3360
3361         while (loops < 1024) {
3362                 /* have the flusher threads jump in and do some IO */
3363                 smp_mb();
3364                 nr_pages = min_t(unsigned long, nr_pages,
3365                        root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT);
3366                 writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages);
3367
3368                 spin_lock(&space_info->lock);
3369                 if (reserved > space_info->bytes_reserved)
3370                         reclaimed += reserved - space_info->bytes_reserved;
3371                 reserved = space_info->bytes_reserved;
3372                 spin_unlock(&space_info->lock);
3373
3374                 loops++;
3375
3376                 if (reserved == 0 || reclaimed >= max_reclaim)
3377                         break;
3378
3379                 if (trans && trans->transaction->blocked)
3380                         return -EAGAIN;
3381
3382                 time_left = schedule_timeout_interruptible(1);
3383
3384                 /* We were interrupted, exit */
3385                 if (time_left)
3386                         break;
3387
3388                 /* we've kicked the IO a few times, if anything has been freed,
3389                  * exit.  There is no sense in looping here for a long time
3390                  * when we really need to commit the transaction, or there are
3391                  * just too many writers without enough free space
3392                  */
3393
3394                 if (loops > 3) {
3395                         smp_mb();
3396                         if (progress != space_info->reservation_progress)
3397                                 break;
3398                 }
3399
3400         }
3401         return reclaimed >= to_reclaim;
3402 }
3403
3404 /*
3405  * Retries tells us how many times we've called reserve_metadata_bytes.  The
3406  * idea is if this is the first call (retries == 0) then we will add to our
3407  * reserved count if we can't make the allocation in order to hold our place
3408  * while we go and try and free up space.  That way for retries > 1 we don't try
3409  * and add space, we just check to see if the amount of unused space is >= the
3410  * total space, meaning that our reservation is valid.
3411  *
3412  * However if we don't intend to retry this reservation, pass -1 as retries so
3413  * that it short circuits this logic.
3414  */
3415 static int reserve_metadata_bytes(struct btrfs_trans_handle *trans,
3416                                   struct btrfs_root *root,
3417                                   struct btrfs_block_rsv *block_rsv,
3418                                   u64 orig_bytes, int flush)
3419 {
3420         struct btrfs_space_info *space_info = block_rsv->space_info;
3421         u64 unused;
3422         u64 num_bytes = orig_bytes;
3423         int retries = 0;
3424         int ret = 0;
3425         bool reserved = false;
3426         bool committed = false;
3427
3428 again:
3429         ret = -ENOSPC;
3430         if (reserved)
3431                 num_bytes = 0;
3432
3433         spin_lock(&space_info->lock);
3434         unused = space_info->bytes_used + space_info->bytes_reserved +
3435                  space_info->bytes_pinned + space_info->bytes_readonly +
3436                  space_info->bytes_may_use;
3437
3438         /*
3439          * The idea here is that we've not already over-reserved the block group
3440          * then we can go ahead and save our reservation first and then start
3441          * flushing if we need to.  Otherwise if we've already overcommitted
3442          * lets start flushing stuff first and then come back and try to make
3443          * our reservation.
3444          */
3445         if (unused <= space_info->total_bytes) {
3446                 unused = space_info->total_bytes - unused;
3447                 if (unused >= num_bytes) {
3448                         if (!reserved)
3449                                 space_info->bytes_reserved += orig_bytes;
3450                         ret = 0;
3451                 } else {
3452                         /*
3453                          * Ok set num_bytes to orig_bytes since we aren't
3454                          * overocmmitted, this way we only try and reclaim what
3455                          * we need.
3456                          */
3457                         num_bytes = orig_bytes;
3458                 }
3459         } else {
3460                 /*
3461                  * Ok we're over committed, set num_bytes to the overcommitted
3462                  * amount plus the amount of bytes that we need for this
3463                  * reservation.
3464                  */
3465                 num_bytes = unused - space_info->total_bytes +
3466                         (orig_bytes * (retries + 1));
3467         }
3468
3469         /*
3470          * Couldn't make our reservation, save our place so while we're trying
3471          * to reclaim space we can actually use it instead of somebody else
3472          * stealing it from us.
3473          */
3474         if (ret && !reserved) {
3475                 space_info->bytes_reserved += orig_bytes;
3476                 reserved = true;
3477         }
3478
3479         spin_unlock(&space_info->lock);
3480
3481         if (!ret)
3482                 return 0;
3483
3484         if (!flush)
3485                 goto out;
3486
3487         /*
3488          * We do synchronous shrinking since we don't actually unreserve
3489          * metadata until after the IO is completed.
3490          */
3491         ret = shrink_delalloc(trans, root, num_bytes, 1);
3492         if (ret > 0)
3493                 return 0;
3494         else if (ret < 0)
3495                 goto out;
3496
3497         /*
3498          * So if we were overcommitted it's possible that somebody else flushed
3499          * out enough space and we simply didn't have enough space to reclaim,
3500          * so go back around and try again.
3501          */
3502         if (retries < 2) {
3503                 retries++;
3504                 goto again;
3505         }
3506
3507         spin_lock(&space_info->lock);
3508         /*
3509          * Not enough space to be reclaimed, don't bother committing the
3510          * transaction.
3511          */
3512         if (space_info->bytes_pinned < orig_bytes)
3513                 ret = -ENOSPC;
3514         spin_unlock(&space_info->lock);
3515         if (ret)
3516                 goto out;
3517
3518         ret = -EAGAIN;
3519         if (trans || committed)
3520                 goto out;
3521
3522         ret = -ENOSPC;
3523         trans = btrfs_join_transaction(root, 1);
3524         if (IS_ERR(trans))
3525                 goto out;
3526         ret = btrfs_commit_transaction(trans, root);
3527         if (!ret) {
3528                 trans = NULL;
3529                 committed = true;
3530                 goto again;
3531         }
3532
3533 out:
3534         if (reserved) {
3535                 spin_lock(&space_info->lock);
3536                 space_info->bytes_reserved -= orig_bytes;
3537                 spin_unlock(&space_info->lock);
3538         }
3539
3540         return ret;
3541 }
3542
3543 static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
3544                                              struct btrfs_root *root)
3545 {
3546         struct btrfs_block_rsv *block_rsv;
3547         if (root->ref_cows)
3548                 block_rsv = trans->block_rsv;
3549         else
3550                 block_rsv = root->block_rsv;
3551
3552         if (!block_rsv)
3553                 block_rsv = &root->fs_info->empty_block_rsv;
3554
3555         return block_rsv;
3556 }
3557
3558 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
3559                                u64 num_bytes)
3560 {
3561         int ret = -ENOSPC;
3562         spin_lock(&block_rsv->lock);
3563         if (block_rsv->reserved >= num_bytes) {
3564                 block_rsv->reserved -= num_bytes;
3565                 if (block_rsv->reserved < block_rsv->size)
3566                         block_rsv->full = 0;
3567                 ret = 0;
3568         }
3569         spin_unlock(&block_rsv->lock);
3570         return ret;
3571 }
3572
3573 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
3574                                 u64 num_bytes, int update_size)
3575 {
3576         spin_lock(&block_rsv->lock);
3577         block_rsv->reserved += num_bytes;
3578         if (update_size)
3579                 block_rsv->size += num_bytes;
3580         else if (block_rsv->reserved >= block_rsv->size)
3581                 block_rsv->full = 1;
3582         spin_unlock(&block_rsv->lock);
3583 }
3584
3585 void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv,
3586                              struct btrfs_block_rsv *dest, u64 num_bytes)
3587 {
3588         struct btrfs_space_info *space_info = block_rsv->space_info;
3589
3590         spin_lock(&block_rsv->lock);
3591         if (num_bytes == (u64)-1)
3592                 num_bytes = block_rsv->size;
3593         block_rsv->size -= num_bytes;
3594         if (block_rsv->reserved >= block_rsv->size) {
3595                 num_bytes = block_rsv->reserved - block_rsv->size;
3596                 block_rsv->reserved = block_rsv->size;
3597                 block_rsv->full = 1;
3598         } else {
3599                 num_bytes = 0;
3600         }
3601         spin_unlock(&block_rsv->lock);
3602
3603         if (num_bytes > 0) {
3604                 if (dest) {
3605                         spin_lock(&dest->lock);
3606                         if (!dest->full) {
3607                                 u64 bytes_to_add;
3608
3609                                 bytes_to_add = dest->size - dest->reserved;
3610                                 bytes_to_add = min(num_bytes, bytes_to_add);
3611                                 dest->reserved += bytes_to_add;
3612                                 if (dest->reserved >= dest->size)
3613                                         dest->full = 1;
3614                                 num_bytes -= bytes_to_add;
3615                         }
3616                         spin_unlock(&dest->lock);
3617                 }
3618                 if (num_bytes) {
3619                         spin_lock(&space_info->lock);
3620                         space_info->bytes_reserved -= num_bytes;
3621                         space_info->reservation_progress++;
3622                         spin_unlock(&space_info->lock);
3623                 }
3624         }
3625 }
3626
3627 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
3628                                    struct btrfs_block_rsv *dst, u64 num_bytes)
3629 {
3630         int ret;
3631
3632         ret = block_rsv_use_bytes(src, num_bytes);
3633         if (ret)
3634                 return ret;
3635
3636         block_rsv_add_bytes(dst, num_bytes, 1);
3637         return 0;
3638 }
3639
3640 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv)
3641 {
3642         memset(rsv, 0, sizeof(*rsv));
3643         spin_lock_init(&rsv->lock);
3644         atomic_set(&rsv->usage, 1);
3645         rsv->priority = 6;
3646         INIT_LIST_HEAD(&rsv->list);
3647 }
3648
3649 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root)
3650 {
3651         struct btrfs_block_rsv *block_rsv;
3652         struct btrfs_fs_info *fs_info = root->fs_info;
3653
3654         block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
3655         if (!block_rsv)
3656                 return NULL;
3657
3658         btrfs_init_block_rsv(block_rsv);
3659         block_rsv->space_info = __find_space_info(fs_info,
3660                                                   BTRFS_BLOCK_GROUP_METADATA);
3661         return block_rsv;
3662 }
3663
3664 void btrfs_free_block_rsv(struct btrfs_root *root,
3665                           struct btrfs_block_rsv *rsv)
3666 {
3667         if (rsv && atomic_dec_and_test(&rsv->usage)) {
3668                 btrfs_block_rsv_release(root, rsv, (u64)-1);
3669                 if (!rsv->durable)
3670                         kfree(rsv);
3671         }
3672 }
3673
3674 /*
3675  * make the block_rsv struct be able to capture freed space.
3676  * the captured space will re-add to the the block_rsv struct
3677  * after transaction commit
3678  */
3679 void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info,
3680                                  struct btrfs_block_rsv *block_rsv)
3681 {
3682         block_rsv->durable = 1;
3683         mutex_lock(&fs_info->durable_block_rsv_mutex);
3684         list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list);
3685         mutex_unlock(&fs_info->durable_block_rsv_mutex);
3686 }
3687
3688 int btrfs_block_rsv_add(struct btrfs_trans_handle *trans,
3689                         struct btrfs_root *root,
3690                         struct btrfs_block_rsv *block_rsv,
3691                         u64 num_bytes)
3692 {
3693         int ret;
3694
3695         if (num_bytes == 0)
3696                 return 0;
3697
3698         ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1);
3699         if (!ret) {
3700                 block_rsv_add_bytes(block_rsv, num_bytes, 1);
3701                 return 0;
3702         }
3703
3704         return ret;
3705 }
3706
3707 int btrfs_block_rsv_check(struct btrfs_trans_handle *trans,
3708                           struct btrfs_root *root,
3709                           struct btrfs_block_rsv *block_rsv,
3710                           u64 min_reserved, int min_factor)
3711 {
3712         u64 num_bytes = 0;
3713         int commit_trans = 0;
3714         int ret = -ENOSPC;
3715
3716         if (!block_rsv)
3717                 return 0;
3718
3719         spin_lock(&block_rsv->lock);
3720         if (min_factor > 0)
3721                 num_bytes = div_factor(block_rsv->size, min_factor);
3722         if (min_reserved > num_bytes)
3723                 num_bytes = min_reserved;
3724
3725         if (block_rsv->reserved >= num_bytes) {
3726                 ret = 0;
3727         } else {
3728                 num_bytes -= block_rsv->reserved;
3729                 if (block_rsv->durable &&
3730                     block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes)
3731                         commit_trans = 1;
3732         }
3733         spin_unlock(&block_rsv->lock);
3734         if (!ret)
3735                 return 0;
3736
3737         if (block_rsv->refill_used) {
3738                 ret = reserve_metadata_bytes(trans, root, block_rsv,
3739                                              num_bytes, 0);
3740                 if (!ret) {
3741                         block_rsv_add_bytes(block_rsv, num_bytes, 0);
3742                         return 0;
3743                 }
3744         }
3745
3746         if (commit_trans) {
3747                 if (trans)
3748                         return -EAGAIN;
3749
3750                 trans = btrfs_join_transaction(root, 1);
3751                 BUG_ON(IS_ERR(trans));
3752                 ret = btrfs_commit_transaction(trans, root);
3753                 return 0;
3754         }
3755
3756         return -ENOSPC;
3757 }
3758
3759 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
3760                             struct btrfs_block_rsv *dst_rsv,
3761                             u64 num_bytes)
3762 {
3763         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3764 }
3765
3766 void btrfs_block_rsv_release(struct btrfs_root *root,
3767                              struct btrfs_block_rsv *block_rsv,
3768                              u64 num_bytes)
3769 {
3770         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3771         if (global_rsv->full || global_rsv == block_rsv ||
3772             block_rsv->space_info != global_rsv->space_info)
3773                 global_rsv = NULL;
3774         block_rsv_release_bytes(block_rsv, global_rsv, num_bytes);
3775 }
3776
3777 /*
3778  * helper to calculate size of global block reservation.
3779  * the desired value is sum of space used by extent tree,
3780  * checksum tree and root tree
3781  */
3782 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
3783 {
3784         struct btrfs_space_info *sinfo;
3785         u64 num_bytes;
3786         u64 meta_used;
3787         u64 data_used;
3788         int csum_size = btrfs_super_csum_size(&fs_info->super_copy);
3789 #if 0
3790         /*
3791          * per tree used space accounting can be inaccuracy, so we
3792          * can't rely on it.
3793          */
3794         spin_lock(&fs_info->extent_root->accounting_lock);
3795         num_bytes = btrfs_root_used(&fs_info->extent_root->root_item);
3796         spin_unlock(&fs_info->extent_root->accounting_lock);
3797
3798         spin_lock(&fs_info->csum_root->accounting_lock);
3799         num_bytes += btrfs_root_used(&fs_info->csum_root->root_item);
3800         spin_unlock(&fs_info->csum_root->accounting_lock);
3801
3802         spin_lock(&fs_info->tree_root->accounting_lock);
3803         num_bytes += btrfs_root_used(&fs_info->tree_root->root_item);
3804         spin_unlock(&fs_info->tree_root->accounting_lock);
3805 #endif
3806         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
3807         spin_lock(&sinfo->lock);
3808         data_used = sinfo->bytes_used;
3809         spin_unlock(&sinfo->lock);
3810
3811         sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3812         spin_lock(&sinfo->lock);
3813         if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
3814                 data_used = 0;
3815         meta_used = sinfo->bytes_used;
3816         spin_unlock(&sinfo->lock);
3817
3818         num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
3819                     csum_size * 2;
3820         num_bytes += div64_u64(data_used + meta_used, 50);
3821
3822         if (num_bytes * 3 > meta_used)
3823                 num_bytes = div64_u64(meta_used, 3);
3824
3825         return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
3826 }
3827
3828 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
3829 {
3830         struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
3831         struct btrfs_space_info *sinfo = block_rsv->space_info;
3832         u64 num_bytes;
3833
3834         num_bytes = calc_global_metadata_size(fs_info);
3835
3836         spin_lock(&block_rsv->lock);
3837         spin_lock(&sinfo->lock);
3838
3839         block_rsv->size = num_bytes;
3840
3841         num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
3842                     sinfo->bytes_reserved + sinfo->bytes_readonly +
3843                     sinfo->bytes_may_use;
3844
3845         if (sinfo->total_bytes > num_bytes) {
3846                 num_bytes = sinfo->total_bytes - num_bytes;
3847                 block_rsv->reserved += num_bytes;
3848                 sinfo->bytes_reserved += num_bytes;
3849         }
3850
3851         if (block_rsv->reserved >= block_rsv->size) {
3852                 num_bytes = block_rsv->reserved - block_rsv->size;
3853                 sinfo->bytes_reserved -= num_bytes;
3854                 sinfo->reservation_progress++;
3855                 block_rsv->reserved = block_rsv->size;
3856                 block_rsv->full = 1;
3857         }
3858 #if 0
3859         printk(KERN_INFO"global block rsv size %llu reserved %llu\n",
3860                 block_rsv->size, block_rsv->reserved);
3861 #endif
3862         spin_unlock(&sinfo->lock);
3863         spin_unlock(&block_rsv->lock);
3864 }
3865
3866 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
3867 {
3868         struct btrfs_space_info *space_info;
3869
3870         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3871         fs_info->chunk_block_rsv.space_info = space_info;
3872         fs_info->chunk_block_rsv.priority = 10;
3873
3874         space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
3875         fs_info->global_block_rsv.space_info = space_info;
3876         fs_info->global_block_rsv.priority = 10;
3877         fs_info->global_block_rsv.refill_used = 1;
3878         fs_info->delalloc_block_rsv.space_info = space_info;
3879         fs_info->trans_block_rsv.space_info = space_info;
3880         fs_info->empty_block_rsv.space_info = space_info;
3881         fs_info->empty_block_rsv.priority = 10;
3882
3883         fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
3884         fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
3885         fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
3886         fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
3887         fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
3888
3889         btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv);
3890
3891         btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv);
3892
3893         update_global_block_rsv(fs_info);
3894 }
3895
3896 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
3897 {
3898         block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1);
3899         WARN_ON(fs_info->delalloc_block_rsv.size > 0);
3900         WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
3901         WARN_ON(fs_info->trans_block_rsv.size > 0);
3902         WARN_ON(fs_info->trans_block_rsv.reserved > 0);
3903         WARN_ON(fs_info->chunk_block_rsv.size > 0);
3904         WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
3905 }
3906
3907 static u64 calc_trans_metadata_size(struct btrfs_root *root, int num_items)
3908 {
3909         return (root->leafsize + root->nodesize * (BTRFS_MAX_LEVEL - 1)) *
3910                 3 * num_items;
3911 }
3912
3913 int btrfs_trans_reserve_metadata(struct btrfs_trans_handle *trans,
3914                                  struct btrfs_root *root,
3915                                  int num_items)
3916 {
3917         u64 num_bytes;
3918         int ret;
3919
3920         if (num_items == 0 || root->fs_info->chunk_root == root)
3921                 return 0;
3922
3923         num_bytes = calc_trans_metadata_size(root, num_items);
3924         ret = btrfs_block_rsv_add(trans, root, &root->fs_info->trans_block_rsv,
3925                                   num_bytes);
3926         if (!ret) {
3927                 trans->bytes_reserved += num_bytes;
3928                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3929         }
3930         return ret;
3931 }
3932
3933 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
3934                                   struct btrfs_root *root)
3935 {
3936         if (!trans->bytes_reserved)
3937                 return;
3938
3939         BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv);
3940         btrfs_block_rsv_release(root, trans->block_rsv,
3941                                 trans->bytes_reserved);
3942         trans->bytes_reserved = 0;
3943 }
3944
3945 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
3946                                   struct inode *inode)
3947 {
3948         struct btrfs_root *root = BTRFS_I(inode)->root;
3949         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3950         struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
3951
3952         /*
3953          * one for deleting orphan item, one for updating inode and
3954          * two for calling btrfs_truncate_inode_items.
3955          *
3956          * btrfs_truncate_inode_items is a delete operation, it frees
3957          * more space than it uses in most cases. So two units of
3958          * metadata space should be enough for calling it many times.
3959          * If all of the metadata space is used, we can commit
3960          * transaction and use space it freed.
3961          */
3962         u64 num_bytes = calc_trans_metadata_size(root, 4);
3963         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3964 }
3965
3966 void btrfs_orphan_release_metadata(struct inode *inode)
3967 {
3968         struct btrfs_root *root = BTRFS_I(inode)->root;
3969         u64 num_bytes = calc_trans_metadata_size(root, 4);
3970         btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
3971 }
3972
3973 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans,
3974                                 struct btrfs_pending_snapshot *pending)
3975 {
3976         struct btrfs_root *root = pending->root;
3977         struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
3978         struct btrfs_block_rsv *dst_rsv = &pending->block_rsv;
3979         /*
3980          * two for root back/forward refs, two for directory entries
3981          * and one for root of the snapshot.
3982          */
3983         u64 num_bytes = calc_trans_metadata_size(root, 5);
3984         dst_rsv->space_info = src_rsv->space_info;
3985         return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
3986 }
3987
3988 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes)
3989 {
3990         return num_bytes >>= 3;
3991 }
3992
3993 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
3994 {
3995         struct btrfs_root *root = BTRFS_I(inode)->root;
3996         struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
3997         u64 to_reserve;
3998         int nr_extents;
3999         int reserved_extents;
4000         int ret;
4001
4002         if (btrfs_transaction_in_commit(root->fs_info))
4003                 schedule_timeout(1);
4004
4005         num_bytes = ALIGN(num_bytes, root->sectorsize);
4006
4007         nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents) + 1;
4008         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4009
4010         if (nr_extents > reserved_extents) {
4011                 nr_extents -= reserved_extents;
4012                 to_reserve = calc_trans_metadata_size(root, nr_extents);
4013         } else {
4014                 nr_extents = 0;
4015                 to_reserve = 0;
4016         }
4017
4018         to_reserve += calc_csum_metadata_size(inode, num_bytes);
4019         ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1);
4020         if (ret)
4021                 return ret;
4022
4023         atomic_add(nr_extents, &BTRFS_I(inode)->reserved_extents);
4024         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
4025
4026         block_rsv_add_bytes(block_rsv, to_reserve, 1);
4027
4028         if (block_rsv->size > 512 * 1024 * 1024)
4029                 shrink_delalloc(NULL, root, to_reserve, 0);
4030
4031         return 0;
4032 }
4033
4034 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
4035 {
4036         struct btrfs_root *root = BTRFS_I(inode)->root;
4037         u64 to_free;
4038         int nr_extents;
4039         int reserved_extents;
4040
4041         num_bytes = ALIGN(num_bytes, root->sectorsize);
4042         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
4043         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents) < 0);
4044
4045         reserved_extents = atomic_read(&BTRFS_I(inode)->reserved_extents);
4046         do {
4047                 int old, new;
4048
4049                 nr_extents = atomic_read(&BTRFS_I(inode)->outstanding_extents);
4050                 if (nr_extents >= reserved_extents) {
4051                         nr_extents = 0;
4052                         break;
4053                 }
4054                 old = reserved_extents;
4055                 nr_extents = reserved_extents - nr_extents;
4056                 new = reserved_extents - nr_extents;
4057                 old = atomic_cmpxchg(&BTRFS_I(inode)->reserved_extents,
4058                                      reserved_extents, new);
4059                 if (likely(old == reserved_extents))
4060                         break;
4061                 reserved_extents = old;
4062         } while (1);
4063
4064         to_free = calc_csum_metadata_size(inode, num_bytes);
4065         if (nr_extents > 0)
4066                 to_free += calc_trans_metadata_size(root, nr_extents);
4067
4068         btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
4069                                 to_free);
4070 }
4071
4072 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
4073 {
4074         int ret;
4075
4076         ret = btrfs_check_data_free_space(inode, num_bytes);
4077         if (ret)
4078                 return ret;
4079
4080         ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
4081         if (ret) {
4082                 btrfs_free_reserved_data_space(inode, num_bytes);
4083                 return ret;
4084         }
4085
4086         return 0;
4087 }
4088
4089 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
4090 {
4091         btrfs_delalloc_release_metadata(inode, num_bytes);
4092         btrfs_free_reserved_data_space(inode, num_bytes);
4093 }
4094
4095 static int update_block_group(struct btrfs_trans_handle *trans,
4096                               struct btrfs_root *root,
4097                               u64 bytenr, u64 num_bytes, int alloc)
4098 {
4099         struct btrfs_block_group_cache *cache = NULL;
4100         struct btrfs_fs_info *info = root->fs_info;
4101         u64 total = num_bytes;
4102         u64 old_val;
4103         u64 byte_in_group;
4104         int factor;
4105
4106         /* block accounting for super block */
4107         spin_lock(&info->delalloc_lock);
4108         old_val = btrfs_super_bytes_used(&info->super_copy);
4109         if (alloc)
4110                 old_val += num_bytes;
4111         else
4112                 old_val -= num_bytes;
4113         btrfs_set_super_bytes_used(&info->super_copy, old_val);
4114         spin_unlock(&info->delalloc_lock);
4115
4116         while (total) {
4117                 cache = btrfs_lookup_block_group(info, bytenr);
4118                 if (!cache)
4119                         return -1;
4120                 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
4121                                     BTRFS_BLOCK_GROUP_RAID1 |
4122                                     BTRFS_BLOCK_GROUP_RAID10))
4123                         factor = 2;
4124                 else
4125                         factor = 1;
4126                 /*
4127                  * If this block group has free space cache written out, we
4128                  * need to make sure to load it if we are removing space.  This
4129                  * is because we need the unpinning stage to actually add the
4130                  * space back to the block group, otherwise we will leak space.
4131                  */
4132                 if (!alloc && cache->cached == BTRFS_CACHE_NO)
4133                         cache_block_group(cache, trans, NULL, 1);
4134
4135                 byte_in_group = bytenr - cache->key.objectid;
4136                 WARN_ON(byte_in_group > cache->key.offset);
4137
4138                 spin_lock(&cache->space_info->lock);
4139                 spin_lock(&cache->lock);
4140
4141                 if (btrfs_super_cache_generation(&info->super_copy) != 0 &&
4142                     cache->disk_cache_state < BTRFS_DC_CLEAR)
4143                         cache->disk_cache_state = BTRFS_DC_CLEAR;
4144
4145                 cache->dirty = 1;
4146                 old_val = btrfs_block_group_used(&cache->item);
4147                 num_bytes = min(total, cache->key.offset - byte_in_group);
4148                 if (alloc) {
4149                         old_val += num_bytes;
4150                         btrfs_set_block_group_used(&cache->item, old_val);
4151                         cache->reserved -= num_bytes;
4152                         cache->space_info->bytes_reserved -= num_bytes;
4153                         cache->space_info->reservation_progress++;
4154                         cache->space_info->bytes_used += num_bytes;
4155                         cache->space_info->disk_used += num_bytes * factor;
4156                         spin_unlock(&cache->lock);
4157                         spin_unlock(&cache->space_info->lock);
4158                 } else {
4159                         old_val -= num_bytes;
4160                         btrfs_set_block_group_used(&cache->item, old_val);
4161                         cache->pinned += num_bytes;
4162                         cache->space_info->bytes_pinned += num_bytes;
4163                         cache->space_info->bytes_used -= num_bytes;
4164                         cache->space_info->disk_used -= num_bytes * factor;
4165                         spin_unlock(&cache->lock);
4166                         spin_unlock(&cache->space_info->lock);
4167
4168                         set_extent_dirty(info->pinned_extents,
4169                                          bytenr, bytenr + num_bytes - 1,
4170                                          GFP_NOFS | __GFP_NOFAIL);
4171                 }
4172                 btrfs_put_block_group(cache);
4173                 total -= num_bytes;
4174                 bytenr += num_bytes;
4175         }
4176         return 0;
4177 }
4178
4179 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
4180 {
4181         struct btrfs_block_group_cache *cache;
4182         u64 bytenr;
4183
4184         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
4185         if (!cache)
4186                 return 0;
4187
4188         bytenr = cache->key.objectid;
4189         btrfs_put_block_group(cache);
4190
4191         return bytenr;
4192 }
4193
4194 static int pin_down_extent(struct btrfs_root *root,
4195                            struct btrfs_block_group_cache *cache,
4196                            u64 bytenr, u64 num_bytes, int reserved)
4197 {
4198         spin_lock(&cache->space_info->lock);
4199         spin_lock(&cache->lock);
4200         cache->pinned += num_bytes;
4201         cache->space_info->bytes_pinned += num_bytes;
4202         if (reserved) {
4203                 cache->reserved -= num_bytes;
4204                 cache->space_info->bytes_reserved -= num_bytes;
4205                 cache->space_info->reservation_progress++;
4206         }
4207         spin_unlock(&cache->lock);
4208         spin_unlock(&cache->space_info->lock);
4209
4210         set_extent_dirty(root->fs_info->pinned_extents, bytenr,
4211                          bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
4212         return 0;
4213 }
4214
4215 /*
4216  * this function must be called within transaction
4217  */
4218 int btrfs_pin_extent(struct btrfs_root *root,
4219                      u64 bytenr, u64 num_bytes, int reserved)
4220 {
4221         struct btrfs_block_group_cache *cache;
4222
4223         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
4224         BUG_ON(!cache);
4225
4226         pin_down_extent(root, cache, bytenr, num_bytes, reserved);
4227
4228         btrfs_put_block_group(cache);
4229         return 0;
4230 }
4231
4232 /*
4233  * update size of reserved extents. this function may return -EAGAIN
4234  * if 'reserve' is true or 'sinfo' is false.
4235  */
4236 static int update_reserved_bytes(struct btrfs_block_group_cache *cache,
4237                                  u64 num_bytes, int reserve, int sinfo)
4238 {
4239         int ret = 0;
4240         if (sinfo) {
4241                 struct btrfs_space_info *space_info = cache->space_info;
4242                 spin_lock(&space_info->lock);
4243                 spin_lock(&cache->lock);
4244                 if (reserve) {
4245                         if (cache->ro) {
4246                                 ret = -EAGAIN;
4247                         } else {
4248                                 cache->reserved += num_bytes;
4249                                 space_info->bytes_reserved += num_bytes;
4250                         }
4251                 } else {
4252                         if (cache->ro)
4253                                 space_info->bytes_readonly += num_bytes;
4254                         cache->reserved -= num_bytes;
4255                         space_info->bytes_reserved -= num_bytes;
4256                         space_info->reservation_progress++;
4257                 }
4258                 spin_unlock(&cache->lock);
4259                 spin_unlock(&space_info->lock);
4260         } else {
4261                 spin_lock(&cache->lock);
4262                 if (cache->ro) {
4263                         ret = -EAGAIN;
4264                 } else {
4265                         if (reserve)
4266                                 cache->reserved += num_bytes;
4267                         else
4268                                 cache->reserved -= num_bytes;
4269                 }
4270                 spin_unlock(&cache->lock);
4271         }
4272         return ret;
4273 }
4274
4275 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
4276                                 struct btrfs_root *root)
4277 {
4278         struct btrfs_fs_info *fs_info = root->fs_info;
4279         struct btrfs_caching_control *next;
4280         struct btrfs_caching_control *caching_ctl;
4281         struct btrfs_block_group_cache *cache;
4282
4283         down_write(&fs_info->extent_commit_sem);
4284
4285         list_for_each_entry_safe(caching_ctl, next,
4286                                  &fs_info->caching_block_groups, list) {
4287                 cache = caching_ctl->block_group;
4288                 if (block_group_cache_done(cache)) {
4289                         cache->last_byte_to_unpin = (u64)-1;
4290                         list_del_init(&caching_ctl->list);
4291                         put_caching_control(caching_ctl);
4292                 } else {
4293                         cache->last_byte_to_unpin = caching_ctl->progress;
4294                 }
4295         }
4296
4297         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4298                 fs_info->pinned_extents = &fs_info->freed_extents[1];
4299         else
4300                 fs_info->pinned_extents = &fs_info->freed_extents[0];
4301
4302         up_write(&fs_info->extent_commit_sem);
4303
4304         update_global_block_rsv(fs_info);
4305         return 0;
4306 }
4307
4308 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
4309 {
4310         struct btrfs_fs_info *fs_info = root->fs_info;
4311         struct btrfs_block_group_cache *cache = NULL;
4312         u64 len;
4313
4314         while (start <= end) {
4315                 if (!cache ||
4316                     start >= cache->key.objectid + cache->key.offset) {
4317                         if (cache)
4318                                 btrfs_put_block_group(cache);
4319                         cache = btrfs_lookup_block_group(fs_info, start);
4320                         BUG_ON(!cache);
4321                 }
4322
4323                 len = cache->key.objectid + cache->key.offset - start;
4324                 len = min(len, end + 1 - start);
4325
4326                 if (start < cache->last_byte_to_unpin) {
4327                         len = min(len, cache->last_byte_to_unpin - start);
4328                         btrfs_add_free_space(cache, start, len);
4329                 }
4330
4331                 start += len;
4332
4333                 spin_lock(&cache->space_info->lock);
4334                 spin_lock(&cache->lock);
4335                 cache->pinned -= len;
4336                 cache->space_info->bytes_pinned -= len;
4337                 if (cache->ro) {
4338                         cache->space_info->bytes_readonly += len;
4339                 } else if (cache->reserved_pinned > 0) {
4340                         len = min(len, cache->reserved_pinned);
4341                         cache->reserved_pinned -= len;
4342                         cache->space_info->bytes_reserved += len;
4343                 }
4344                 spin_unlock(&cache->lock);
4345                 spin_unlock(&cache->space_info->lock);
4346         }
4347
4348         if (cache)
4349                 btrfs_put_block_group(cache);
4350         return 0;
4351 }
4352
4353 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
4354                                struct btrfs_root *root)
4355 {
4356         struct btrfs_fs_info *fs_info = root->fs_info;
4357         struct extent_io_tree *unpin;
4358         struct btrfs_block_rsv *block_rsv;
4359         struct btrfs_block_rsv *next_rsv;
4360         u64 start;
4361         u64 end;
4362         int idx;
4363         int ret;
4364
4365         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
4366                 unpin = &fs_info->freed_extents[1];
4367         else
4368                 unpin = &fs_info->freed_extents[0];
4369
4370         while (1) {
4371                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4372                                             EXTENT_DIRTY);
4373                 if (ret)
4374                         break;
4375
4376                 ret = btrfs_discard_extent(root, start, end + 1 - start);
4377
4378                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4379                 unpin_extent_range(root, start, end);
4380                 cond_resched();
4381         }
4382
4383         mutex_lock(&fs_info->durable_block_rsv_mutex);
4384         list_for_each_entry_safe(block_rsv, next_rsv,
4385                                  &fs_info->durable_block_rsv_list, list) {
4386
4387                 idx = trans->transid & 0x1;
4388                 if (block_rsv->freed[idx] > 0) {
4389                         block_rsv_add_bytes(block_rsv,
4390                                             block_rsv->freed[idx], 0);
4391                         block_rsv->freed[idx] = 0;
4392                 }
4393                 if (atomic_read(&block_rsv->usage) == 0) {
4394                         btrfs_block_rsv_release(root, block_rsv, (u64)-1);
4395
4396                         if (block_rsv->freed[0] == 0 &&
4397                             block_rsv->freed[1] == 0) {
4398                                 list_del_init(&block_rsv->list);
4399                                 kfree(block_rsv);
4400                         }
4401                 } else {
4402                         btrfs_block_rsv_release(root, block_rsv, 0);
4403                 }
4404         }
4405         mutex_unlock(&fs_info->durable_block_rsv_mutex);
4406
4407         return 0;
4408 }
4409
4410 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
4411                                 struct btrfs_root *root,
4412                                 u64 bytenr, u64 num_bytes, u64 parent,
4413                                 u64 root_objectid, u64 owner_objectid,
4414                                 u64 owner_offset, int refs_to_drop,
4415                                 struct btrfs_delayed_extent_op *extent_op)
4416 {
4417         struct btrfs_key key;
4418         struct btrfs_path *path;
4419         struct btrfs_fs_info *info = root->fs_info;
4420         struct btrfs_root *extent_root = info->extent_root;
4421         struct extent_buffer *leaf;
4422         struct btrfs_extent_item *ei;
4423         struct btrfs_extent_inline_ref *iref;
4424         int ret;
4425         int is_data;
4426         int extent_slot = 0;
4427         int found_extent = 0;
4428         int num_to_del = 1;
4429         u32 item_size;
4430         u64 refs;
4431
4432         path = btrfs_alloc_path();
4433         if (!path)
4434                 return -ENOMEM;
4435
4436         path->reada = 1;
4437         path->leave_spinning = 1;
4438
4439         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
4440         BUG_ON(!is_data && refs_to_drop != 1);
4441
4442         ret = lookup_extent_backref(trans, extent_root, path, &iref,
4443                                     bytenr, num_bytes, parent,
4444                                     root_objectid, owner_objectid,
4445                                     owner_offset);
4446         if (ret == 0) {
4447                 extent_slot = path->slots[0];
4448                 while (extent_slot >= 0) {
4449                         btrfs_item_key_to_cpu(path->nodes[0], &key,
4450                                               extent_slot);
4451                         if (key.objectid != bytenr)
4452                                 break;
4453                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
4454                             key.offset == num_bytes) {
4455                                 found_extent = 1;
4456                                 break;
4457                         }
4458                         if (path->slots[0] - extent_slot > 5)
4459                                 break;
4460                         extent_slot--;
4461                 }
4462 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4463                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
4464                 if (found_extent && item_size < sizeof(*ei))
4465                         found_extent = 0;
4466 #endif
4467                 if (!found_extent) {
4468                         BUG_ON(iref);
4469                         ret = remove_extent_backref(trans, extent_root, path,
4470                                                     NULL, refs_to_drop,
4471                                                     is_data);
4472                         BUG_ON(ret);
4473                         btrfs_release_path(extent_root, path);
4474                         path->leave_spinning = 1;
4475
4476                         key.objectid = bytenr;
4477                         key.type = BTRFS_EXTENT_ITEM_KEY;
4478                         key.offset = num_bytes;
4479
4480                         ret = btrfs_search_slot(trans, extent_root,
4481                                                 &key, path, -1, 1);
4482                         if (ret) {
4483                                 printk(KERN_ERR "umm, got %d back from search"
4484                                        ", was looking for %llu\n", ret,
4485                                        (unsigned long long)bytenr);
4486                                 btrfs_print_leaf(extent_root, path->nodes[0]);
4487                         }
4488                         BUG_ON(ret);
4489                         extent_slot = path->slots[0];
4490                 }
4491         } else {
4492                 btrfs_print_leaf(extent_root, path->nodes[0]);
4493                 WARN_ON(1);
4494                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
4495                        "parent %llu root %llu  owner %llu offset %llu\n",
4496                        (unsigned long long)bytenr,
4497                        (unsigned long long)parent,
4498                        (unsigned long long)root_objectid,
4499                        (unsigned long long)owner_objectid,
4500                        (unsigned long long)owner_offset);
4501         }
4502
4503         leaf = path->nodes[0];
4504         item_size = btrfs_item_size_nr(leaf, extent_slot);
4505 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4506         if (item_size < sizeof(*ei)) {
4507                 BUG_ON(found_extent || extent_slot != path->slots[0]);
4508                 ret = convert_extent_item_v0(trans, extent_root, path,
4509                                              owner_objectid, 0);
4510                 BUG_ON(ret < 0);
4511
4512                 btrfs_release_path(extent_root, path);
4513                 path->leave_spinning = 1;
4514
4515                 key.objectid = bytenr;
4516                 key.type = BTRFS_EXTENT_ITEM_KEY;
4517                 key.offset = num_bytes;
4518
4519                 ret = btrfs_search_slot(trans, extent_root, &key, path,
4520                                         -1, 1);
4521                 if (ret) {
4522                         printk(KERN_ERR "umm, got %d back from search"
4523                                ", was looking for %llu\n", ret,
4524                                (unsigned long long)bytenr);
4525                         btrfs_print_leaf(extent_root, path->nodes[0]);
4526                 }
4527                 BUG_ON(ret);
4528                 extent_slot = path->slots[0];
4529                 leaf = path->nodes[0];
4530                 item_size = btrfs_item_size_nr(leaf, extent_slot);
4531         }
4532 #endif
4533         BUG_ON(item_size < sizeof(*ei));
4534         ei = btrfs_item_ptr(leaf, extent_slot,
4535                             struct btrfs_extent_item);
4536         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4537                 struct btrfs_tree_block_info *bi;
4538                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
4539                 bi = (struct btrfs_tree_block_info *)(ei + 1);
4540                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
4541         }
4542
4543         refs = btrfs_extent_refs(leaf, ei);
4544         BUG_ON(refs < refs_to_drop);
4545         refs -= refs_to_drop;
4546
4547         if (refs > 0) {
4548                 if (extent_op)
4549                         __run_delayed_extent_op(extent_op, leaf, ei);
4550                 /*
4551                  * In the case of inline back ref, reference count will
4552                  * be updated by remove_extent_backref
4553                  */
4554                 if (iref) {
4555                         BUG_ON(!found_extent);
4556                 } else {
4557                         btrfs_set_extent_refs(leaf, ei, refs);
4558                         btrfs_mark_buffer_dirty(leaf);
4559                 }
4560                 if (found_extent) {
4561                         ret = remove_extent_backref(trans, extent_root, path,
4562                                                     iref, refs_to_drop,
4563                                                     is_data);
4564                         BUG_ON(ret);
4565                 }
4566         } else {
4567                 if (found_extent) {
4568                         BUG_ON(is_data && refs_to_drop !=
4569                                extent_data_ref_count(root, path, iref));
4570                         if (iref) {
4571                                 BUG_ON(path->slots[0] != extent_slot);
4572                         } else {
4573                                 BUG_ON(path->slots[0] != extent_slot + 1);
4574                                 path->slots[0] = extent_slot;
4575                                 num_to_del = 2;
4576                         }
4577                 }
4578
4579                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
4580                                       num_to_del);
4581                 BUG_ON(ret);
4582                 btrfs_release_path(extent_root, path);
4583
4584                 if (is_data) {
4585                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
4586                         BUG_ON(ret);
4587                 } else {
4588                         invalidate_mapping_pages(info->btree_inode->i_mapping,
4589                              bytenr >> PAGE_CACHE_SHIFT,
4590                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
4591                 }
4592
4593                 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
4594                 BUG_ON(ret);
4595         }
4596         btrfs_free_path(path);
4597         return ret;
4598 }
4599
4600 /*
4601  * when we free an block, it is possible (and likely) that we free the last
4602  * delayed ref for that extent as well.  This searches the delayed ref tree for
4603  * a given extent, and if there are no other delayed refs to be processed, it
4604  * removes it from the tree.
4605  */
4606 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
4607                                       struct btrfs_root *root, u64 bytenr)
4608 {
4609         struct btrfs_delayed_ref_head *head;
4610         struct btrfs_delayed_ref_root *delayed_refs;
4611         struct btrfs_delayed_ref_node *ref;
4612         struct rb_node *node;
4613         int ret = 0;
4614
4615         delayed_refs = &trans->transaction->delayed_refs;
4616         spin_lock(&delayed_refs->lock);
4617         head = btrfs_find_delayed_ref_head(trans, bytenr);
4618         if (!head)
4619                 goto out;
4620
4621         node = rb_prev(&head->node.rb_node);
4622         if (!node)
4623                 goto out;
4624
4625         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
4626
4627         /* there are still entries for this ref, we can't drop it */
4628         if (ref->bytenr == bytenr)
4629                 goto out;
4630
4631         if (head->extent_op) {
4632                 if (!head->must_insert_reserved)
4633                         goto out;
4634                 kfree(head->extent_op);
4635                 head->extent_op = NULL;
4636         }
4637
4638         /*
4639          * waiting for the lock here would deadlock.  If someone else has it
4640          * locked they are already in the process of dropping it anyway
4641          */
4642         if (!mutex_trylock(&head->mutex))
4643                 goto out;
4644
4645         /*
4646          * at this point we have a head with no other entries.  Go
4647          * ahead and process it.
4648          */
4649         head->node.in_tree = 0;
4650         rb_erase(&head->node.rb_node, &delayed_refs->root);
4651
4652         delayed_refs->num_entries--;
4653
4654         /*
4655          * we don't take a ref on the node because we're removing it from the
4656          * tree, so we just steal the ref the tree was holding.
4657          */
4658         delayed_refs->num_heads--;
4659         if (list_empty(&head->cluster))
4660                 delayed_refs->num_heads_ready--;
4661
4662         list_del_init(&head->cluster);
4663         spin_unlock(&delayed_refs->lock);
4664
4665         BUG_ON(head->extent_op);
4666         if (head->must_insert_reserved)
4667                 ret = 1;
4668
4669         mutex_unlock(&head->mutex);
4670         btrfs_put_delayed_ref(&head->node);
4671         return ret;
4672 out:
4673         spin_unlock(&delayed_refs->lock);
4674         return 0;
4675 }
4676
4677 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
4678                            struct btrfs_root *root,
4679                            struct extent_buffer *buf,
4680                            u64 parent, int last_ref)
4681 {
4682         struct btrfs_block_rsv *block_rsv;
4683         struct btrfs_block_group_cache *cache = NULL;
4684         int ret;
4685
4686         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4687                 ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
4688                                                 parent, root->root_key.objectid,
4689                                                 btrfs_header_level(buf),
4690                                                 BTRFS_DROP_DELAYED_REF, NULL);
4691                 BUG_ON(ret);
4692         }
4693
4694         if (!last_ref)
4695                 return;
4696
4697         block_rsv = get_block_rsv(trans, root);
4698         cache = btrfs_lookup_block_group(root->fs_info, buf->start);
4699         if (block_rsv->space_info != cache->space_info)
4700                 goto out;
4701
4702         if (btrfs_header_generation(buf) == trans->transid) {
4703                 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4704                         ret = check_ref_cleanup(trans, root, buf->start);
4705                         if (!ret)
4706                                 goto pin;
4707                 }
4708
4709                 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
4710                         pin_down_extent(root, cache, buf->start, buf->len, 1);
4711                         goto pin;
4712                 }
4713
4714                 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
4715
4716                 btrfs_add_free_space(cache, buf->start, buf->len);
4717                 ret = update_reserved_bytes(cache, buf->len, 0, 0);
4718                 if (ret == -EAGAIN) {
4719                         /* block group became read-only */
4720                         update_reserved_bytes(cache, buf->len, 0, 1);
4721                         goto out;
4722                 }
4723
4724                 ret = 1;
4725                 spin_lock(&block_rsv->lock);
4726                 if (block_rsv->reserved < block_rsv->size) {
4727                         block_rsv->reserved += buf->len;
4728                         ret = 0;
4729                 }
4730                 spin_unlock(&block_rsv->lock);
4731
4732                 if (ret) {
4733                         spin_lock(&cache->space_info->lock);
4734                         cache->space_info->bytes_reserved -= buf->len;
4735                         cache->space_info->reservation_progress++;
4736                         spin_unlock(&cache->space_info->lock);
4737                 }
4738                 goto out;
4739         }
4740 pin:
4741         if (block_rsv->durable && !cache->ro) {
4742                 ret = 0;
4743                 spin_lock(&cache->lock);
4744                 if (!cache->ro) {
4745                         cache->reserved_pinned += buf->len;
4746                         ret = 1;
4747                 }
4748                 spin_unlock(&cache->lock);
4749
4750                 if (ret) {
4751                         spin_lock(&block_rsv->lock);
4752                         block_rsv->freed[trans->transid & 0x1] += buf->len;
4753                         spin_unlock(&block_rsv->lock);
4754                 }
4755         }
4756 out:
4757         btrfs_put_block_group(cache);
4758 }
4759
4760 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4761                       struct btrfs_root *root,
4762                       u64 bytenr, u64 num_bytes, u64 parent,
4763                       u64 root_objectid, u64 owner, u64 offset)
4764 {
4765         int ret;
4766
4767         /*
4768          * tree log blocks never actually go into the extent allocation
4769          * tree, just update pinning info and exit early.
4770          */
4771         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4772                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4773                 /* unlocks the pinned mutex */
4774                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4775                 ret = 0;
4776         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4777                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4778                                         parent, root_objectid, (int)owner,
4779                                         BTRFS_DROP_DELAYED_REF, NULL);
4780                 BUG_ON(ret);
4781         } else {
4782                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4783                                         parent, root_objectid, owner,
4784                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4785                 BUG_ON(ret);
4786         }
4787         return ret;
4788 }
4789
4790 static u64 stripe_align(struct btrfs_root *root, u64 val)
4791 {
4792         u64 mask = ((u64)root->stripesize - 1);
4793         u64 ret = (val + mask) & ~mask;
4794         return ret;
4795 }
4796
4797 /*
4798  * when we wait for progress in the block group caching, its because
4799  * our allocation attempt failed at least once.  So, we must sleep
4800  * and let some progress happen before we try again.
4801  *
4802  * This function will sleep at least once waiting for new free space to
4803  * show up, and then it will check the block group free space numbers
4804  * for our min num_bytes.  Another option is to have it go ahead
4805  * and look in the rbtree for a free extent of a given size, but this
4806  * is a good start.
4807  */
4808 static noinline int
4809 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4810                                 u64 num_bytes)
4811 {
4812         struct btrfs_caching_control *caching_ctl;
4813         DEFINE_WAIT(wait);
4814
4815         caching_ctl = get_caching_control(cache);
4816         if (!caching_ctl)
4817                 return 0;
4818
4819         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4820                    (cache->free_space >= num_bytes));
4821
4822         put_caching_control(caching_ctl);
4823         return 0;
4824 }
4825
4826 static noinline int
4827 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4828 {
4829         struct btrfs_caching_control *caching_ctl;
4830         DEFINE_WAIT(wait);
4831
4832         caching_ctl = get_caching_control(cache);
4833         if (!caching_ctl)
4834                 return 0;
4835
4836         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4837
4838         put_caching_control(caching_ctl);
4839         return 0;
4840 }
4841
4842 static int get_block_group_index(struct btrfs_block_group_cache *cache)
4843 {
4844         int index;
4845         if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
4846                 index = 0;
4847         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
4848                 index = 1;
4849         else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
4850                 index = 2;
4851         else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
4852                 index = 3;
4853         else
4854                 index = 4;
4855         return index;
4856 }
4857
4858 enum btrfs_loop_type {
4859         LOOP_FIND_IDEAL = 0,
4860         LOOP_CACHING_NOWAIT = 1,
4861         LOOP_CACHING_WAIT = 2,
4862         LOOP_ALLOC_CHUNK = 3,
4863         LOOP_NO_EMPTY_SIZE = 4,
4864 };
4865
4866 /*
4867  * walks the btree of allocated extents and find a hole of a given size.
4868  * The key ins is changed to record the hole:
4869  * ins->objectid == block start
4870  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4871  * ins->offset == number of blocks
4872  * Any available blocks before search_start are skipped.
4873  */
4874 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4875                                      struct btrfs_root *orig_root,
4876                                      u64 num_bytes, u64 empty_size,
4877                                      u64 search_start, u64 search_end,
4878                                      u64 hint_byte, struct btrfs_key *ins,
4879                                      int data)
4880 {
4881         int ret = 0;
4882         struct btrfs_root *root = orig_root->fs_info->extent_root;
4883         struct btrfs_free_cluster *last_ptr = NULL;
4884         struct btrfs_block_group_cache *block_group = NULL;
4885         int empty_cluster = 2 * 1024 * 1024;
4886         int allowed_chunk_alloc = 0;
4887         int done_chunk_alloc = 0;
4888         struct btrfs_space_info *space_info;
4889         int last_ptr_loop = 0;
4890         int loop = 0;
4891         int index = 0;
4892         bool found_uncached_bg = false;
4893         bool failed_cluster_refill = false;
4894         bool failed_alloc = false;
4895         bool use_cluster = true;
4896         u64 ideal_cache_percent = 0;
4897         u64 ideal_cache_offset = 0;
4898
4899         WARN_ON(num_bytes < root->sectorsize);
4900         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4901         ins->objectid = 0;
4902         ins->offset = 0;
4903
4904         space_info = __find_space_info(root->fs_info, data);
4905         if (!space_info) {
4906                 printk(KERN_ERR "No space info for %d\n", data);
4907                 return -ENOSPC;
4908         }
4909
4910         /*
4911          * If the space info is for both data and metadata it means we have a
4912          * small filesystem and we can't use the clustering stuff.
4913          */
4914         if (btrfs_mixed_space_info(space_info))
4915                 use_cluster = false;
4916
4917         if (orig_root->ref_cows || empty_size)
4918                 allowed_chunk_alloc = 1;
4919
4920         if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
4921                 last_ptr = &root->fs_info->meta_alloc_cluster;
4922                 if (!btrfs_test_opt(root, SSD))
4923                         empty_cluster = 64 * 1024;
4924         }
4925
4926         if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
4927             btrfs_test_opt(root, SSD)) {
4928                 last_ptr = &root->fs_info->data_alloc_cluster;
4929         }
4930
4931         if (last_ptr) {
4932                 spin_lock(&last_ptr->lock);
4933                 if (last_ptr->block_group)
4934                         hint_byte = last_ptr->window_start;
4935                 spin_unlock(&last_ptr->lock);
4936         }
4937
4938         search_start = max(search_start, first_logical_byte(root, 0));
4939         search_start = max(search_start, hint_byte);
4940
4941         if (!last_ptr)
4942                 empty_cluster = 0;
4943
4944         if (search_start == hint_byte) {
4945 ideal_cache:
4946                 block_group = btrfs_lookup_block_group(root->fs_info,
4947                                                        search_start);
4948                 /*
4949                  * we don't want to use the block group if it doesn't match our
4950                  * allocation bits, or if its not cached.
4951                  *
4952                  * However if we are re-searching with an ideal block group
4953                  * picked out then we don't care that the block group is cached.
4954                  */
4955                 if (block_group && block_group_bits(block_group, data) &&
4956                     (block_group->cached != BTRFS_CACHE_NO ||
4957                      search_start == ideal_cache_offset)) {
4958                         down_read(&space_info->groups_sem);
4959                         if (list_empty(&block_group->list) ||
4960                             block_group->ro) {
4961                                 /*
4962                                  * someone is removing this block group,
4963                                  * we can't jump into the have_block_group
4964                                  * target because our list pointers are not
4965                                  * valid
4966                                  */
4967                                 btrfs_put_block_group(block_group);
4968                                 up_read(&space_info->groups_sem);
4969                         } else {
4970                                 index = get_block_group_index(block_group);
4971                                 goto have_block_group;
4972                         }
4973                 } else if (block_group) {
4974                         btrfs_put_block_group(block_group);
4975                 }
4976         }
4977 search:
4978         down_read(&space_info->groups_sem);
4979         list_for_each_entry(block_group, &space_info->block_groups[index],
4980                             list) {
4981                 u64 offset;
4982                 int cached;
4983
4984                 btrfs_get_block_group(block_group);
4985                 search_start = block_group->key.objectid;
4986
4987                 /*
4988                  * this can happen if we end up cycling through all the
4989                  * raid types, but we want to make sure we only allocate
4990                  * for the proper type.
4991                  */
4992                 if (!block_group_bits(block_group, data)) {
4993                     u64 extra = BTRFS_BLOCK_GROUP_DUP |
4994                                 BTRFS_BLOCK_GROUP_RAID1 |
4995                                 BTRFS_BLOCK_GROUP_RAID10;
4996
4997                         /*
4998                          * if they asked for extra copies and this block group
4999                          * doesn't provide them, bail.  This does allow us to
5000                          * fill raid0 from raid1.
5001                          */
5002                         if ((data & extra) && !(block_group->flags & extra))
5003                                 goto loop;
5004                 }
5005
5006 have_block_group:
5007                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
5008                         u64 free_percent;
5009
5010                         ret = cache_block_group(block_group, trans,
5011                                                 orig_root, 1);
5012                         if (block_group->cached == BTRFS_CACHE_FINISHED)
5013                                 goto have_block_group;
5014
5015                         free_percent = btrfs_block_group_used(&block_group->item);
5016                         free_percent *= 100;
5017                         free_percent = div64_u64(free_percent,
5018                                                  block_group->key.offset);
5019                         free_percent = 100 - free_percent;
5020                         if (free_percent > ideal_cache_percent &&
5021                             likely(!block_group->ro)) {
5022                                 ideal_cache_offset = block_group->key.objectid;
5023                                 ideal_cache_percent = free_percent;
5024                         }
5025
5026                         /*
5027                          * We only want to start kthread caching if we are at
5028                          * the point where we will wait for caching to make
5029                          * progress, or if our ideal search is over and we've
5030                          * found somebody to start caching.
5031                          */
5032                         if (loop > LOOP_CACHING_NOWAIT ||
5033                             (loop > LOOP_FIND_IDEAL &&
5034                              atomic_read(&space_info->caching_threads) < 2)) {
5035                                 ret = cache_block_group(block_group, trans,
5036                                                         orig_root, 0);
5037                                 BUG_ON(ret);
5038                         }
5039                         found_uncached_bg = true;
5040
5041                         /*
5042                          * If loop is set for cached only, try the next block
5043                          * group.
5044                          */
5045                         if (loop == LOOP_FIND_IDEAL)
5046                                 goto loop;
5047                 }
5048
5049                 cached = block_group_cache_done(block_group);
5050                 if (unlikely(!cached))
5051                         found_uncached_bg = true;
5052
5053                 if (unlikely(block_group->ro))
5054                         goto loop;
5055
5056                 /*
5057                  * Ok we want to try and use the cluster allocator, so lets look
5058                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5059                  * have tried the cluster allocator plenty of times at this
5060                  * point and not have found anything, so we are likely way too
5061                  * fragmented for the clustering stuff to find anything, so lets
5062                  * just skip it and let the allocator find whatever block it can
5063                  * find
5064                  */
5065                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
5066                         /*
5067                          * the refill lock keeps out other
5068                          * people trying to start a new cluster
5069                          */
5070                         spin_lock(&last_ptr->refill_lock);
5071                         if (last_ptr->block_group &&
5072                             (last_ptr->block_group->ro ||
5073                             !block_group_bits(last_ptr->block_group, data))) {
5074                                 offset = 0;
5075                                 goto refill_cluster;
5076                         }
5077
5078                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
5079                                                  num_bytes, search_start);
5080                         if (offset) {
5081                                 /* we have a block, we're done */
5082                                 spin_unlock(&last_ptr->refill_lock);
5083                                 goto checks;
5084                         }
5085
5086                         spin_lock(&last_ptr->lock);
5087                         /*
5088                          * whoops, this cluster doesn't actually point to
5089                          * this block group.  Get a ref on the block
5090                          * group is does point to and try again
5091                          */
5092                         if (!last_ptr_loop && last_ptr->block_group &&
5093                             last_ptr->block_group != block_group) {
5094
5095                                 btrfs_put_block_group(block_group);
5096                                 block_group = last_ptr->block_group;
5097                                 btrfs_get_block_group(block_group);
5098                                 spin_unlock(&last_ptr->lock);
5099                                 spin_unlock(&last_ptr->refill_lock);
5100
5101                                 last_ptr_loop = 1;
5102                                 search_start = block_group->key.objectid;
5103                                 /*
5104                                  * we know this block group is properly
5105                                  * in the list because
5106                                  * btrfs_remove_block_group, drops the
5107                                  * cluster before it removes the block
5108                                  * group from the list
5109                                  */
5110                                 goto have_block_group;
5111                         }
5112                         spin_unlock(&last_ptr->lock);
5113 refill_cluster:
5114                         /*
5115                          * this cluster didn't work out, free it and
5116                          * start over
5117                          */
5118                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5119
5120                         last_ptr_loop = 0;
5121
5122                         /* allocate a cluster in this block group */
5123                         ret = btrfs_find_space_cluster(trans, root,
5124                                                block_group, last_ptr,
5125                                                offset, num_bytes,
5126                                                empty_cluster + empty_size);
5127                         if (ret == 0) {
5128                                 /*
5129                                  * now pull our allocation out of this
5130                                  * cluster
5131                                  */
5132                                 offset = btrfs_alloc_from_cluster(block_group,
5133                                                   last_ptr, num_bytes,
5134                                                   search_start);
5135                                 if (offset) {
5136                                         /* we found one, proceed */
5137                                         spin_unlock(&last_ptr->refill_lock);
5138                                         goto checks;
5139                                 }
5140                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
5141                                    && !failed_cluster_refill) {
5142                                 spin_unlock(&last_ptr->refill_lock);
5143
5144                                 failed_cluster_refill = true;
5145                                 wait_block_group_cache_progress(block_group,
5146                                        num_bytes + empty_cluster + empty_size);
5147                                 goto have_block_group;
5148                         }
5149
5150                         /*
5151                          * at this point we either didn't find a cluster
5152                          * or we weren't able to allocate a block from our
5153                          * cluster.  Free the cluster we've been trying
5154                          * to use, and go to the next block group
5155                          */
5156                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
5157                         spin_unlock(&last_ptr->refill_lock);
5158                         goto loop;
5159                 }
5160
5161                 offset = btrfs_find_space_for_alloc(block_group, search_start,
5162                                                     num_bytes, empty_size);
5163                 /*
5164                  * If we didn't find a chunk, and we haven't failed on this
5165                  * block group before, and this block group is in the middle of
5166                  * caching and we are ok with waiting, then go ahead and wait
5167                  * for progress to be made, and set failed_alloc to true.
5168                  *
5169                  * If failed_alloc is true then we've already waited on this
5170                  * block group once and should move on to the next block group.
5171                  */
5172                 if (!offset && !failed_alloc && !cached &&
5173                     loop > LOOP_CACHING_NOWAIT) {
5174                         wait_block_group_cache_progress(block_group,
5175                                                 num_bytes + empty_size);
5176                         failed_alloc = true;
5177                         goto have_block_group;
5178                 } else if (!offset) {
5179                         goto loop;
5180                 }
5181 checks:
5182                 search_start = stripe_align(root, offset);
5183                 /* move on to the next group */
5184                 if (search_start + num_bytes >= search_end) {
5185                         btrfs_add_free_space(block_group, offset, num_bytes);
5186                         goto loop;
5187                 }
5188
5189                 /* move on to the next group */
5190                 if (search_start + num_bytes >
5191                     block_group->key.objectid + block_group->key.offset) {
5192                         btrfs_add_free_space(block_group, offset, num_bytes);
5193                         goto loop;
5194                 }
5195
5196                 ins->objectid = search_start;
5197                 ins->offset = num_bytes;
5198
5199                 if (offset < search_start)
5200                         btrfs_add_free_space(block_group, offset,
5201                                              search_start - offset);
5202                 BUG_ON(offset > search_start);
5203
5204                 ret = update_reserved_bytes(block_group, num_bytes, 1,
5205                                             (data & BTRFS_BLOCK_GROUP_DATA));
5206                 if (ret == -EAGAIN) {
5207                         btrfs_add_free_space(block_group, offset, num_bytes);
5208                         goto loop;
5209                 }
5210
5211                 /* we are all good, lets return */
5212                 ins->objectid = search_start;
5213                 ins->offset = num_bytes;
5214
5215                 if (offset < search_start)
5216                         btrfs_add_free_space(block_group, offset,
5217                                              search_start - offset);
5218                 BUG_ON(offset > search_start);
5219                 break;
5220 loop:
5221                 failed_cluster_refill = false;
5222                 failed_alloc = false;
5223                 BUG_ON(index != get_block_group_index(block_group));
5224                 btrfs_put_block_group(block_group);
5225         }
5226         up_read(&space_info->groups_sem);
5227
5228         if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
5229                 goto search;
5230
5231         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5232          *                      for them to make caching progress.  Also
5233          *                      determine the best possible bg to cache
5234          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5235          *                      caching kthreads as we move along
5236          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5237          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5238          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5239          *                      again
5240          */
5241         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
5242             (found_uncached_bg || empty_size || empty_cluster ||
5243              allowed_chunk_alloc)) {
5244                 index = 0;
5245                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
5246                         found_uncached_bg = false;
5247                         loop++;
5248                         if (!ideal_cache_percent &&
5249                             atomic_read(&space_info->caching_threads))
5250                                 goto search;
5251
5252                         /*
5253                          * 1 of the following 2 things have happened so far
5254                          *
5255                          * 1) We found an ideal block group for caching that
5256                          * is mostly full and will cache quickly, so we might
5257                          * as well wait for it.
5258                          *
5259                          * 2) We searched for cached only and we didn't find
5260                          * anything, and we didn't start any caching kthreads
5261                          * either, so chances are we will loop through and
5262                          * start a couple caching kthreads, and then come back
5263                          * around and just wait for them.  This will be slower
5264                          * because we will have 2 caching kthreads reading at
5265                          * the same time when we could have just started one
5266                          * and waited for it to get far enough to give us an
5267                          * allocation, so go ahead and go to the wait caching
5268                          * loop.
5269                          */
5270                         loop = LOOP_CACHING_WAIT;
5271                         search_start = ideal_cache_offset;
5272                         ideal_cache_percent = 0;
5273                         goto ideal_cache;
5274                 } else if (loop == LOOP_FIND_IDEAL) {
5275                         /*
5276                          * Didn't find a uncached bg, wait on anything we find
5277                          * next.
5278                          */
5279                         loop = LOOP_CACHING_WAIT;
5280                         goto search;
5281                 }
5282
5283                 if (loop < LOOP_CACHING_WAIT) {
5284                         loop++;
5285                         goto search;
5286                 }
5287
5288                 if (loop == LOOP_ALLOC_CHUNK) {
5289                         empty_size = 0;
5290                         empty_cluster = 0;
5291                 }
5292
5293                 if (allowed_chunk_alloc) {
5294                         ret = do_chunk_alloc(trans, root, num_bytes +
5295                                              2 * 1024 * 1024, data, 1);
5296                         allowed_chunk_alloc = 0;
5297                         done_chunk_alloc = 1;
5298                 } else if (!done_chunk_alloc) {
5299                         space_info->force_alloc = 1;
5300                 }
5301
5302                 if (loop < LOOP_NO_EMPTY_SIZE) {
5303                         loop++;
5304                         goto search;
5305                 }
5306                 ret = -ENOSPC;
5307         } else if (!ins->objectid) {
5308                 ret = -ENOSPC;
5309         }
5310
5311         /* we found what we needed */
5312         if (ins->objectid) {
5313                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
5314                         trans->block_group = block_group->key.objectid;
5315
5316                 btrfs_put_block_group(block_group);
5317                 ret = 0;
5318         }
5319
5320         return ret;
5321 }
5322
5323 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
5324                             int dump_block_groups)
5325 {
5326         struct btrfs_block_group_cache *cache;
5327         int index = 0;
5328
5329         spin_lock(&info->lock);
5330         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
5331                (unsigned long long)(info->total_bytes - info->bytes_used -
5332                                     info->bytes_pinned - info->bytes_reserved -
5333                                     info->bytes_readonly),
5334                (info->full) ? "" : "not ");
5335         printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, "
5336                "reserved=%llu, may_use=%llu, readonly=%llu\n",
5337                (unsigned long long)info->total_bytes,
5338                (unsigned long long)info->bytes_used,
5339                (unsigned long long)info->bytes_pinned,
5340                (unsigned long long)info->bytes_reserved,
5341                (unsigned long long)info->bytes_may_use,
5342                (unsigned long long)info->bytes_readonly);
5343         spin_unlock(&info->lock);
5344
5345         if (!dump_block_groups)
5346                 return;
5347
5348         down_read(&info->groups_sem);
5349 again:
5350         list_for_each_entry(cache, &info->block_groups[index], list) {
5351                 spin_lock(&cache->lock);
5352                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
5353                        "%llu pinned %llu reserved\n",
5354                        (unsigned long long)cache->key.objectid,
5355                        (unsigned long long)cache->key.offset,
5356                        (unsigned long long)btrfs_block_group_used(&cache->item),
5357                        (unsigned long long)cache->pinned,
5358                        (unsigned long long)cache->reserved);
5359                 btrfs_dump_free_space(cache, bytes);
5360                 spin_unlock(&cache->lock);
5361         }
5362         if (++index < BTRFS_NR_RAID_TYPES)
5363                 goto again;
5364         up_read(&info->groups_sem);
5365 }
5366
5367 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
5368                          struct btrfs_root *root,
5369                          u64 num_bytes, u64 min_alloc_size,
5370                          u64 empty_size, u64 hint_byte,
5371                          u64 search_end, struct btrfs_key *ins,
5372                          u64 data)
5373 {
5374         int ret;
5375         u64 search_start = 0;
5376
5377         data = btrfs_get_alloc_profile(root, data);
5378 again:
5379         /*
5380          * the only place that sets empty_size is btrfs_realloc_node, which
5381          * is not called recursively on allocations
5382          */
5383         if (empty_size || root->ref_cows)
5384                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
5385                                      num_bytes + 2 * 1024 * 1024, data, 0);
5386
5387         WARN_ON(num_bytes < root->sectorsize);
5388         ret = find_free_extent(trans, root, num_bytes, empty_size,
5389                                search_start, search_end, hint_byte,
5390                                ins, data);
5391
5392         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
5393                 num_bytes = num_bytes >> 1;
5394                 num_bytes = num_bytes & ~(root->sectorsize - 1);
5395                 num_bytes = max(num_bytes, min_alloc_size);
5396                 do_chunk_alloc(trans, root->fs_info->extent_root,
5397                                num_bytes, data, 1);
5398                 goto again;
5399         }
5400         if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) {
5401                 struct btrfs_space_info *sinfo;
5402
5403                 sinfo = __find_space_info(root->fs_info, data);
5404                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
5405                        "wanted %llu\n", (unsigned long long)data,
5406                        (unsigned long long)num_bytes);
5407                 dump_space_info(sinfo, num_bytes, 1);
5408         }
5409
5410         return ret;
5411 }
5412
5413 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
5414 {
5415         struct btrfs_block_group_cache *cache;
5416         int ret = 0;
5417
5418         cache = btrfs_lookup_block_group(root->fs_info, start);
5419         if (!cache) {
5420                 printk(KERN_ERR "Unable to find block group for %llu\n",
5421                        (unsigned long long)start);
5422                 return -ENOSPC;
5423         }
5424
5425         ret = btrfs_discard_extent(root, start, len);
5426
5427         btrfs_add_free_space(cache, start, len);
5428         update_reserved_bytes(cache, len, 0, 1);
5429         btrfs_put_block_group(cache);
5430
5431         return ret;
5432 }
5433
5434 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5435                                       struct btrfs_root *root,
5436                                       u64 parent, u64 root_objectid,
5437                                       u64 flags, u64 owner, u64 offset,
5438                                       struct btrfs_key *ins, int ref_mod)
5439 {
5440         int ret;
5441         struct btrfs_fs_info *fs_info = root->fs_info;
5442         struct btrfs_extent_item *extent_item;
5443         struct btrfs_extent_inline_ref *iref;
5444         struct btrfs_path *path;
5445         struct extent_buffer *leaf;
5446         int type;
5447         u32 size;
5448
5449         if (parent > 0)
5450                 type = BTRFS_SHARED_DATA_REF_KEY;
5451         else
5452                 type = BTRFS_EXTENT_DATA_REF_KEY;
5453
5454         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
5455
5456         path = btrfs_alloc_path();
5457         BUG_ON(!path);
5458
5459         path->leave_spinning = 1;
5460         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5461                                       ins, size);
5462         BUG_ON(ret);
5463
5464         leaf = path->nodes[0];
5465         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5466                                      struct btrfs_extent_item);
5467         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
5468         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5469         btrfs_set_extent_flags(leaf, extent_item,
5470                                flags | BTRFS_EXTENT_FLAG_DATA);
5471
5472         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
5473         btrfs_set_extent_inline_ref_type(leaf, iref, type);
5474         if (parent > 0) {
5475                 struct btrfs_shared_data_ref *ref;
5476                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
5477                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5478                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
5479         } else {
5480                 struct btrfs_extent_data_ref *ref;
5481                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
5482                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
5483                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
5484                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
5485                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
5486         }
5487
5488         btrfs_mark_buffer_dirty(path->nodes[0]);
5489         btrfs_free_path(path);
5490
5491         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5492         if (ret) {
5493                 printk(KERN_ERR "btrfs update block group failed for %llu "
5494                        "%llu\n", (unsigned long long)ins->objectid,
5495                        (unsigned long long)ins->offset);
5496                 BUG();
5497         }
5498         return ret;
5499 }
5500
5501 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
5502                                      struct btrfs_root *root,
5503                                      u64 parent, u64 root_objectid,
5504                                      u64 flags, struct btrfs_disk_key *key,
5505                                      int level, struct btrfs_key *ins)
5506 {
5507         int ret;
5508         struct btrfs_fs_info *fs_info = root->fs_info;
5509         struct btrfs_extent_item *extent_item;
5510         struct btrfs_tree_block_info *block_info;
5511         struct btrfs_extent_inline_ref *iref;
5512         struct btrfs_path *path;
5513         struct extent_buffer *leaf;
5514         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
5515
5516         path = btrfs_alloc_path();
5517         BUG_ON(!path);
5518
5519         path->leave_spinning = 1;
5520         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
5521                                       ins, size);
5522         BUG_ON(ret);
5523
5524         leaf = path->nodes[0];
5525         extent_item = btrfs_item_ptr(leaf, path->slots[0],
5526                                      struct btrfs_extent_item);
5527         btrfs_set_extent_refs(leaf, extent_item, 1);
5528         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
5529         btrfs_set_extent_flags(leaf, extent_item,
5530                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
5531         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
5532
5533         btrfs_set_tree_block_key(leaf, block_info, key);
5534         btrfs_set_tree_block_level(leaf, block_info, level);
5535
5536         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
5537         if (parent > 0) {
5538                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
5539                 btrfs_set_extent_inline_ref_type(leaf, iref,
5540                                                  BTRFS_SHARED_BLOCK_REF_KEY);
5541                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
5542         } else {
5543                 btrfs_set_extent_inline_ref_type(leaf, iref,
5544                                                  BTRFS_TREE_BLOCK_REF_KEY);
5545                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
5546         }
5547
5548         btrfs_mark_buffer_dirty(leaf);
5549         btrfs_free_path(path);
5550
5551         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
5552         if (ret) {
5553                 printk(KERN_ERR "btrfs update block group failed for %llu "
5554                        "%llu\n", (unsigned long long)ins->objectid,
5555                        (unsigned long long)ins->offset);
5556                 BUG();
5557         }
5558         return ret;
5559 }
5560
5561 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
5562                                      struct btrfs_root *root,
5563                                      u64 root_objectid, u64 owner,
5564                                      u64 offset, struct btrfs_key *ins)
5565 {
5566         int ret;
5567
5568         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
5569
5570         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
5571                                          0, root_objectid, owner, offset,
5572                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
5573         return ret;
5574 }
5575
5576 /*
5577  * this is used by the tree logging recovery code.  It records that
5578  * an extent has been allocated and makes sure to clear the free
5579  * space cache bits as well
5580  */
5581 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
5582                                    struct btrfs_root *root,
5583                                    u64 root_objectid, u64 owner, u64 offset,
5584                                    struct btrfs_key *ins)
5585 {
5586         int ret;
5587         struct btrfs_block_group_cache *block_group;
5588         struct btrfs_caching_control *caching_ctl;
5589         u64 start = ins->objectid;
5590         u64 num_bytes = ins->offset;
5591
5592         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
5593         cache_block_group(block_group, trans, NULL, 0);
5594         caching_ctl = get_caching_control(block_group);
5595
5596         if (!caching_ctl) {
5597                 BUG_ON(!block_group_cache_done(block_group));
5598                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5599                 BUG_ON(ret);
5600         } else {
5601                 mutex_lock(&caching_ctl->mutex);
5602
5603                 if (start >= caching_ctl->progress) {
5604                         ret = add_excluded_extent(root, start, num_bytes);
5605                         BUG_ON(ret);
5606                 } else if (start + num_bytes <= caching_ctl->progress) {
5607                         ret = btrfs_remove_free_space(block_group,
5608                                                       start, num_bytes);
5609                         BUG_ON(ret);
5610                 } else {
5611                         num_bytes = caching_ctl->progress - start;
5612                         ret = btrfs_remove_free_space(block_group,
5613                                                       start, num_bytes);
5614                         BUG_ON(ret);
5615
5616                         start = caching_ctl->progress;
5617                         num_bytes = ins->objectid + ins->offset -
5618                                     caching_ctl->progress;
5619                         ret = add_excluded_extent(root, start, num_bytes);
5620                         BUG_ON(ret);
5621                 }
5622
5623                 mutex_unlock(&caching_ctl->mutex);
5624                 put_caching_control(caching_ctl);
5625         }
5626
5627         ret = update_reserved_bytes(block_group, ins->offset, 1, 1);
5628         BUG_ON(ret);
5629         btrfs_put_block_group(block_group);
5630         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
5631                                          0, owner, offset, ins, 1);
5632         return ret;
5633 }
5634
5635 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
5636                                             struct btrfs_root *root,
5637                                             u64 bytenr, u32 blocksize,
5638                                             int level)
5639 {
5640         struct extent_buffer *buf;
5641
5642         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5643         if (!buf)
5644                 return ERR_PTR(-ENOMEM);
5645         btrfs_set_header_generation(buf, trans->transid);
5646         btrfs_set_buffer_lockdep_class(buf, level);
5647         btrfs_tree_lock(buf);
5648         clean_tree_block(trans, root, buf);
5649
5650         btrfs_set_lock_blocking(buf);
5651         btrfs_set_buffer_uptodate(buf);
5652
5653         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
5654                 /*
5655                  * we allow two log transactions at a time, use different
5656                  * EXENT bit to differentiate dirty pages.
5657                  */
5658                 if (root->log_transid % 2 == 0)
5659                         set_extent_dirty(&root->dirty_log_pages, buf->start,
5660                                         buf->start + buf->len - 1, GFP_NOFS);
5661                 else
5662                         set_extent_new(&root->dirty_log_pages, buf->start,
5663                                         buf->start + buf->len - 1, GFP_NOFS);
5664         } else {
5665                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
5666                          buf->start + buf->len - 1, GFP_NOFS);
5667         }
5668         trans->blocks_used++;
5669         /* this returns a buffer locked for blocking */
5670         return buf;
5671 }
5672
5673 static struct btrfs_block_rsv *
5674 use_block_rsv(struct btrfs_trans_handle *trans,
5675               struct btrfs_root *root, u32 blocksize)
5676 {
5677         struct btrfs_block_rsv *block_rsv;
5678         struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5679         int ret;
5680
5681         block_rsv = get_block_rsv(trans, root);
5682
5683         if (block_rsv->size == 0) {
5684                 ret = reserve_metadata_bytes(trans, root, block_rsv,
5685                                              blocksize, 0);
5686                 /*
5687                  * If we couldn't reserve metadata bytes try and use some from
5688                  * the global reserve.
5689                  */
5690                 if (ret && block_rsv != global_rsv) {
5691                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5692                         if (!ret)
5693                                 return global_rsv;
5694                         return ERR_PTR(ret);
5695                 } else if (ret) {
5696                         return ERR_PTR(ret);
5697                 }
5698                 return block_rsv;
5699         }
5700
5701         ret = block_rsv_use_bytes(block_rsv, blocksize);
5702         if (!ret)
5703                 return block_rsv;
5704         if (ret) {
5705                 WARN_ON(1);
5706                 ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize,
5707                                              0);
5708                 if (!ret) {
5709                         spin_lock(&block_rsv->lock);
5710                         block_rsv->size += blocksize;
5711                         spin_unlock(&block_rsv->lock);
5712                         return block_rsv;
5713                 } else if (ret && block_rsv != global_rsv) {
5714                         ret = block_rsv_use_bytes(global_rsv, blocksize);
5715                         if (!ret)
5716                                 return global_rsv;
5717                 }
5718         }
5719
5720         return ERR_PTR(-ENOSPC);
5721 }
5722
5723 static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize)
5724 {
5725         block_rsv_add_bytes(block_rsv, blocksize, 0);
5726         block_rsv_release_bytes(block_rsv, NULL, 0);
5727 }
5728
5729 /*
5730  * finds a free extent and does all the dirty work required for allocation
5731  * returns the key for the extent through ins, and a tree buffer for
5732  * the first block of the extent through buf.
5733  *
5734  * returns the tree buffer or NULL.
5735  */
5736 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
5737                                         struct btrfs_root *root, u32 blocksize,
5738                                         u64 parent, u64 root_objectid,
5739                                         struct btrfs_disk_key *key, int level,
5740                                         u64 hint, u64 empty_size)
5741 {
5742         struct btrfs_key ins;
5743         struct btrfs_block_rsv *block_rsv;
5744         struct extent_buffer *buf;
5745         u64 flags = 0;
5746         int ret;
5747
5748
5749         block_rsv = use_block_rsv(trans, root, blocksize);
5750         if (IS_ERR(block_rsv))
5751                 return ERR_CAST(block_rsv);
5752
5753         ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
5754                                    empty_size, hint, (u64)-1, &ins, 0);
5755         if (ret) {
5756                 unuse_block_rsv(block_rsv, blocksize);
5757                 return ERR_PTR(ret);
5758         }
5759
5760         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
5761                                     blocksize, level);
5762         BUG_ON(IS_ERR(buf));
5763
5764         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
5765                 if (parent == 0)
5766                         parent = ins.objectid;
5767                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5768         } else
5769                 BUG_ON(parent > 0);
5770
5771         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
5772                 struct btrfs_delayed_extent_op *extent_op;
5773                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
5774                 BUG_ON(!extent_op);
5775                 if (key)
5776                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
5777                 else
5778                         memset(&extent_op->key, 0, sizeof(extent_op->key));
5779                 extent_op->flags_to_set = flags;
5780                 extent_op->update_key = 1;
5781                 extent_op->update_flags = 1;
5782                 extent_op->is_data = 0;
5783
5784                 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
5785                                         ins.offset, parent, root_objectid,
5786                                         level, BTRFS_ADD_DELAYED_EXTENT,
5787                                         extent_op);
5788                 BUG_ON(ret);
5789         }
5790         return buf;
5791 }
5792
5793 struct walk_control {
5794         u64 refs[BTRFS_MAX_LEVEL];
5795         u64 flags[BTRFS_MAX_LEVEL];
5796         struct btrfs_key update_progress;
5797         int stage;
5798         int level;
5799         int shared_level;
5800         int update_ref;
5801         int keep_locks;
5802         int reada_slot;
5803         int reada_count;
5804 };
5805
5806 #define DROP_REFERENCE  1
5807 #define UPDATE_BACKREF  2
5808
5809 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
5810                                      struct btrfs_root *root,
5811                                      struct walk_control *wc,
5812                                      struct btrfs_path *path)
5813 {
5814         u64 bytenr;
5815         u64 generation;
5816         u64 refs;
5817         u64 flags;
5818         u32 nritems;
5819         u32 blocksize;
5820         struct btrfs_key key;
5821         struct extent_buffer *eb;
5822         int ret;
5823         int slot;
5824         int nread = 0;
5825
5826         if (path->slots[wc->level] < wc->reada_slot) {
5827                 wc->reada_count = wc->reada_count * 2 / 3;
5828                 wc->reada_count = max(wc->reada_count, 2);
5829         } else {
5830                 wc->reada_count = wc->reada_count * 3 / 2;
5831                 wc->reada_count = min_t(int, wc->reada_count,
5832                                         BTRFS_NODEPTRS_PER_BLOCK(root));
5833         }
5834
5835         eb = path->nodes[wc->level];
5836         nritems = btrfs_header_nritems(eb);
5837         blocksize = btrfs_level_size(root, wc->level - 1);
5838
5839         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5840                 if (nread >= wc->reada_count)
5841                         break;
5842
5843                 cond_resched();
5844                 bytenr = btrfs_node_blockptr(eb, slot);
5845                 generation = btrfs_node_ptr_generation(eb, slot);
5846
5847                 if (slot == path->slots[wc->level])
5848                         goto reada;
5849
5850                 if (wc->stage == UPDATE_BACKREF &&
5851                     generation <= root->root_key.offset)
5852                         continue;
5853
5854                 /* We don't lock the tree block, it's OK to be racy here */
5855                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5856                                                &refs, &flags);
5857                 BUG_ON(ret);
5858                 BUG_ON(refs == 0);
5859
5860                 if (wc->stage == DROP_REFERENCE) {
5861                         if (refs == 1)
5862                                 goto reada;
5863
5864                         if (wc->level == 1 &&
5865                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5866                                 continue;
5867                         if (!wc->update_ref ||
5868                             generation <= root->root_key.offset)
5869                                 continue;
5870                         btrfs_node_key_to_cpu(eb, &key, slot);
5871                         ret = btrfs_comp_cpu_keys(&key,
5872                                                   &wc->update_progress);
5873                         if (ret < 0)
5874                                 continue;
5875                 } else {
5876                         if (wc->level == 1 &&
5877                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5878                                 continue;
5879                 }
5880 reada:
5881                 ret = readahead_tree_block(root, bytenr, blocksize,
5882                                            generation);
5883                 if (ret)
5884                         break;
5885                 nread++;
5886         }
5887         wc->reada_slot = slot;
5888 }
5889
5890 /*
5891  * hepler to process tree block while walking down the tree.
5892  *
5893  * when wc->stage == UPDATE_BACKREF, this function updates
5894  * back refs for pointers in the block.
5895  *
5896  * NOTE: return value 1 means we should stop walking down.
5897  */
5898 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5899                                    struct btrfs_root *root,
5900                                    struct btrfs_path *path,
5901                                    struct walk_control *wc, int lookup_info)
5902 {
5903         int level = wc->level;
5904         struct extent_buffer *eb = path->nodes[level];
5905         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5906         int ret;
5907
5908         if (wc->stage == UPDATE_BACKREF &&
5909             btrfs_header_owner(eb) != root->root_key.objectid)
5910                 return 1;
5911
5912         /*
5913          * when reference count of tree block is 1, it won't increase
5914          * again. once full backref flag is set, we never clear it.
5915          */
5916         if (lookup_info &&
5917             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5918              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5919                 BUG_ON(!path->locks[level]);
5920                 ret = btrfs_lookup_extent_info(trans, root,
5921                                                eb->start, eb->len,
5922                                                &wc->refs[level],
5923                                                &wc->flags[level]);
5924                 BUG_ON(ret);
5925                 BUG_ON(wc->refs[level] == 0);
5926         }
5927
5928         if (wc->stage == DROP_REFERENCE) {
5929                 if (wc->refs[level] > 1)
5930                         return 1;
5931
5932                 if (path->locks[level] && !wc->keep_locks) {
5933                         btrfs_tree_unlock(eb);
5934                         path->locks[level] = 0;
5935                 }
5936                 return 0;
5937         }
5938
5939         /* wc->stage == UPDATE_BACKREF */
5940         if (!(wc->flags[level] & flag)) {
5941                 BUG_ON(!path->locks[level]);
5942                 ret = btrfs_inc_ref(trans, root, eb, 1);
5943                 BUG_ON(ret);
5944                 ret = btrfs_dec_ref(trans, root, eb, 0);
5945                 BUG_ON(ret);
5946                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5947                                                   eb->len, flag, 0);
5948                 BUG_ON(ret);
5949                 wc->flags[level] |= flag;
5950         }
5951
5952         /*
5953          * the block is shared by multiple trees, so it's not good to
5954          * keep the tree lock
5955          */
5956         if (path->locks[level] && level > 0) {
5957                 btrfs_tree_unlock(eb);
5958                 path->locks[level] = 0;
5959         }
5960         return 0;
5961 }
5962
5963 /*
5964  * hepler to process tree block pointer.
5965  *
5966  * when wc->stage == DROP_REFERENCE, this function checks
5967  * reference count of the block pointed to. if the block
5968  * is shared and we need update back refs for the subtree
5969  * rooted at the block, this function changes wc->stage to
5970  * UPDATE_BACKREF. if the block is shared and there is no
5971  * need to update back, this function drops the reference
5972  * to the block.
5973  *
5974  * NOTE: return value 1 means we should stop walking down.
5975  */
5976 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5977                                  struct btrfs_root *root,
5978                                  struct btrfs_path *path,
5979                                  struct walk_control *wc, int *lookup_info)
5980 {
5981         u64 bytenr;
5982         u64 generation;
5983         u64 parent;
5984         u32 blocksize;
5985         struct btrfs_key key;
5986         struct extent_buffer *next;
5987         int level = wc->level;
5988         int reada = 0;
5989         int ret = 0;
5990
5991         generation = btrfs_node_ptr_generation(path->nodes[level],
5992                                                path->slots[level]);
5993         /*
5994          * if the lower level block was created before the snapshot
5995          * was created, we know there is no need to update back refs
5996          * for the subtree
5997          */
5998         if (wc->stage == UPDATE_BACKREF &&
5999             generation <= root->root_key.offset) {
6000                 *lookup_info = 1;
6001                 return 1;
6002         }
6003
6004         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
6005         blocksize = btrfs_level_size(root, level - 1);
6006
6007         next = btrfs_find_tree_block(root, bytenr, blocksize);
6008         if (!next) {
6009                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
6010                 if (!next)
6011                         return -ENOMEM;
6012                 reada = 1;
6013         }
6014         btrfs_tree_lock(next);
6015         btrfs_set_lock_blocking(next);
6016
6017         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
6018                                        &wc->refs[level - 1],
6019                                        &wc->flags[level - 1]);
6020         BUG_ON(ret);
6021         BUG_ON(wc->refs[level - 1] == 0);
6022         *lookup_info = 0;
6023
6024         if (wc->stage == DROP_REFERENCE) {
6025                 if (wc->refs[level - 1] > 1) {
6026                         if (level == 1 &&
6027                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6028                                 goto skip;
6029
6030                         if (!wc->update_ref ||
6031                             generation <= root->root_key.offset)
6032                                 goto skip;
6033
6034                         btrfs_node_key_to_cpu(path->nodes[level], &key,
6035                                               path->slots[level]);
6036                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
6037                         if (ret < 0)
6038                                 goto skip;
6039
6040                         wc->stage = UPDATE_BACKREF;
6041                         wc->shared_level = level - 1;
6042                 }
6043         } else {
6044                 if (level == 1 &&
6045                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
6046                         goto skip;
6047         }
6048
6049         if (!btrfs_buffer_uptodate(next, generation)) {
6050                 btrfs_tree_unlock(next);
6051                 free_extent_buffer(next);
6052                 next = NULL;
6053                 *lookup_info = 1;
6054         }
6055
6056         if (!next) {
6057                 if (reada && level == 1)
6058                         reada_walk_down(trans, root, wc, path);
6059                 next = read_tree_block(root, bytenr, blocksize, generation);
6060                 btrfs_tree_lock(next);
6061                 btrfs_set_lock_blocking(next);
6062         }
6063
6064         level--;
6065         BUG_ON(level != btrfs_header_level(next));
6066         path->nodes[level] = next;
6067         path->slots[level] = 0;
6068         path->locks[level] = 1;
6069         wc->level = level;
6070         if (wc->level == 1)
6071                 wc->reada_slot = 0;
6072         return 0;
6073 skip:
6074         wc->refs[level - 1] = 0;
6075         wc->flags[level - 1] = 0;
6076         if (wc->stage == DROP_REFERENCE) {
6077                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6078                         parent = path->nodes[level]->start;
6079                 } else {
6080                         BUG_ON(root->root_key.objectid !=
6081                                btrfs_header_owner(path->nodes[level]));
6082                         parent = 0;
6083                 }
6084
6085                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
6086                                         root->root_key.objectid, level - 1, 0);
6087                 BUG_ON(ret);
6088         }
6089         btrfs_tree_unlock(next);
6090         free_extent_buffer(next);
6091         *lookup_info = 1;
6092         return 1;
6093 }
6094
6095 /*
6096  * hepler to process tree block while walking up the tree.
6097  *
6098  * when wc->stage == DROP_REFERENCE, this function drops
6099  * reference count on the block.
6100  *
6101  * when wc->stage == UPDATE_BACKREF, this function changes
6102  * wc->stage back to DROP_REFERENCE if we changed wc->stage
6103  * to UPDATE_BACKREF previously while processing the block.
6104  *
6105  * NOTE: return value 1 means we should stop walking up.
6106  */
6107 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
6108                                  struct btrfs_root *root,
6109                                  struct btrfs_path *path,
6110                                  struct walk_control *wc)
6111 {
6112         int ret;
6113         int level = wc->level;
6114         struct extent_buffer *eb = path->nodes[level];
6115         u64 parent = 0;
6116
6117         if (wc->stage == UPDATE_BACKREF) {
6118                 BUG_ON(wc->shared_level < level);
6119                 if (level < wc->shared_level)
6120                         goto out;
6121
6122                 ret = find_next_key(path, level + 1, &wc->update_progress);
6123                 if (ret > 0)
6124                         wc->update_ref = 0;
6125
6126                 wc->stage = DROP_REFERENCE;
6127                 wc->shared_level = -1;
6128                 path->slots[level] = 0;
6129
6130                 /*
6131                  * check reference count again if the block isn't locked.
6132                  * we should start walking down the tree again if reference
6133                  * count is one.
6134                  */
6135                 if (!path->locks[level]) {
6136                         BUG_ON(level == 0);
6137                         btrfs_tree_lock(eb);
6138                         btrfs_set_lock_blocking(eb);
6139                         path->locks[level] = 1;
6140
6141                         ret = btrfs_lookup_extent_info(trans, root,
6142                                                        eb->start, eb->len,
6143                                                        &wc->refs[level],
6144                                                        &wc->flags[level]);
6145                         BUG_ON(ret);
6146                         BUG_ON(wc->refs[level] == 0);
6147                         if (wc->refs[level] == 1) {
6148                                 btrfs_tree_unlock(eb);
6149                                 path->locks[level] = 0;
6150                                 return 1;
6151                         }
6152                 }
6153         }
6154
6155         /* wc->stage == DROP_REFERENCE */
6156         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
6157
6158         if (wc->refs[level] == 1) {
6159                 if (level == 0) {
6160                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6161                                 ret = btrfs_dec_ref(trans, root, eb, 1);
6162                         else
6163                                 ret = btrfs_dec_ref(trans, root, eb, 0);
6164                         BUG_ON(ret);
6165                 }
6166                 /* make block locked assertion in clean_tree_block happy */
6167                 if (!path->locks[level] &&
6168                     btrfs_header_generation(eb) == trans->transid) {
6169                         btrfs_tree_lock(eb);
6170                         btrfs_set_lock_blocking(eb);
6171                         path->locks[level] = 1;
6172                 }
6173                 clean_tree_block(trans, root, eb);
6174         }
6175
6176         if (eb == root->node) {
6177                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6178                         parent = eb->start;
6179                 else
6180                         BUG_ON(root->root_key.objectid !=
6181                                btrfs_header_owner(eb));
6182         } else {
6183                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
6184                         parent = path->nodes[level + 1]->start;
6185                 else
6186                         BUG_ON(root->root_key.objectid !=
6187                                btrfs_header_owner(path->nodes[level + 1]));
6188         }
6189
6190         btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
6191 out:
6192         wc->refs[level] = 0;
6193         wc->flags[level] = 0;
6194         return 0;
6195 }
6196
6197 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
6198                                    struct btrfs_root *root,
6199                                    struct btrfs_path *path,
6200                                    struct walk_control *wc)
6201 {
6202         int level = wc->level;
6203         int lookup_info = 1;
6204         int ret;
6205
6206         while (level >= 0) {
6207                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
6208                 if (ret > 0)
6209                         break;
6210
6211                 if (level == 0)
6212                         break;
6213
6214                 if (path->slots[level] >=
6215                     btrfs_header_nritems(path->nodes[level]))
6216                         break;
6217
6218                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
6219                 if (ret > 0) {
6220                         path->slots[level]++;
6221                         continue;
6222                 } else if (ret < 0)
6223                         return ret;
6224                 level = wc->level;
6225         }
6226         return 0;
6227 }
6228
6229 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
6230                                  struct btrfs_root *root,
6231                                  struct btrfs_path *path,
6232                                  struct walk_control *wc, int max_level)
6233 {
6234         int level = wc->level;
6235         int ret;
6236
6237         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
6238         while (level < max_level && path->nodes[level]) {
6239                 wc->level = level;
6240                 if (path->slots[level] + 1 <
6241                     btrfs_header_nritems(path->nodes[level])) {
6242                         path->slots[level]++;
6243                         return 0;
6244                 } else {
6245                         ret = walk_up_proc(trans, root, path, wc);
6246                         if (ret > 0)
6247                                 return 0;
6248
6249                         if (path->locks[level]) {
6250                                 btrfs_tree_unlock(path->nodes[level]);
6251                                 path->locks[level] = 0;
6252                         }
6253                         free_extent_buffer(path->nodes[level]);
6254                         path->nodes[level] = NULL;
6255                         level++;
6256                 }
6257         }
6258         return 1;
6259 }
6260
6261 /*
6262  * drop a subvolume tree.
6263  *
6264  * this function traverses the tree freeing any blocks that only
6265  * referenced by the tree.
6266  *
6267  * when a shared tree block is found. this function decreases its
6268  * reference count by one. if update_ref is true, this function
6269  * also make sure backrefs for the shared block and all lower level
6270  * blocks are properly updated.
6271  */
6272 int btrfs_drop_snapshot(struct btrfs_root *root,
6273                         struct btrfs_block_rsv *block_rsv, int update_ref)
6274 {
6275         struct btrfs_path *path;
6276         struct btrfs_trans_handle *trans;
6277         struct btrfs_root *tree_root = root->fs_info->tree_root;
6278         struct btrfs_root_item *root_item = &root->root_item;
6279         struct walk_control *wc;
6280         struct btrfs_key key;
6281         int err = 0;
6282         int ret;
6283         int level;
6284
6285         path = btrfs_alloc_path();
6286         BUG_ON(!path);
6287
6288         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6289         BUG_ON(!wc);
6290
6291         trans = btrfs_start_transaction(tree_root, 0);
6292         BUG_ON(IS_ERR(trans));
6293
6294         if (block_rsv)
6295                 trans->block_rsv = block_rsv;
6296
6297         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
6298                 level = btrfs_header_level(root->node);
6299                 path->nodes[level] = btrfs_lock_root_node(root);
6300                 btrfs_set_lock_blocking(path->nodes[level]);
6301                 path->slots[level] = 0;
6302                 path->locks[level] = 1;
6303                 memset(&wc->update_progress, 0,
6304                        sizeof(wc->update_progress));
6305         } else {
6306                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
6307                 memcpy(&wc->update_progress, &key,
6308                        sizeof(wc->update_progress));
6309
6310                 level = root_item->drop_level;
6311                 BUG_ON(level == 0);
6312                 path->lowest_level = level;
6313                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6314                 path->lowest_level = 0;
6315                 if (ret < 0) {
6316                         err = ret;
6317                         goto out;
6318                 }
6319                 WARN_ON(ret > 0);
6320
6321                 /*
6322                  * unlock our path, this is safe because only this
6323                  * function is allowed to delete this snapshot
6324                  */
6325                 btrfs_unlock_up_safe(path, 0);
6326
6327                 level = btrfs_header_level(root->node);
6328                 while (1) {
6329                         btrfs_tree_lock(path->nodes[level]);
6330                         btrfs_set_lock_blocking(path->nodes[level]);
6331
6332                         ret = btrfs_lookup_extent_info(trans, root,
6333                                                 path->nodes[level]->start,
6334                                                 path->nodes[level]->len,
6335                                                 &wc->refs[level],
6336                                                 &wc->flags[level]);
6337                         BUG_ON(ret);
6338                         BUG_ON(wc->refs[level] == 0);
6339
6340                         if (level == root_item->drop_level)
6341                                 break;
6342
6343                         btrfs_tree_unlock(path->nodes[level]);
6344                         WARN_ON(wc->refs[level] != 1);
6345                         level--;
6346                 }
6347         }
6348
6349         wc->level = level;
6350         wc->shared_level = -1;
6351         wc->stage = DROP_REFERENCE;
6352         wc->update_ref = update_ref;
6353         wc->keep_locks = 0;
6354         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6355
6356         while (1) {
6357                 ret = walk_down_tree(trans, root, path, wc);
6358                 if (ret < 0) {
6359                         err = ret;
6360                         break;
6361                 }
6362
6363                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
6364                 if (ret < 0) {
6365                         err = ret;
6366                         break;
6367                 }
6368
6369                 if (ret > 0) {
6370                         BUG_ON(wc->stage != DROP_REFERENCE);
6371                         break;
6372                 }
6373
6374                 if (wc->stage == DROP_REFERENCE) {
6375                         level = wc->level;
6376                         btrfs_node_key(path->nodes[level],
6377                                        &root_item->drop_progress,
6378                                        path->slots[level]);
6379                         root_item->drop_level = level;
6380                 }
6381
6382                 BUG_ON(wc->level == 0);
6383                 if (btrfs_should_end_transaction(trans, tree_root)) {
6384                         ret = btrfs_update_root(trans, tree_root,
6385                                                 &root->root_key,
6386                                                 root_item);
6387                         BUG_ON(ret);
6388
6389                         btrfs_end_transaction_throttle(trans, tree_root);
6390                         trans = btrfs_start_transaction(tree_root, 0);
6391                         BUG_ON(IS_ERR(trans));
6392                         if (block_rsv)
6393                                 trans->block_rsv = block_rsv;
6394                 }
6395         }
6396         btrfs_release_path(root, path);
6397         BUG_ON(err);
6398
6399         ret = btrfs_del_root(trans, tree_root, &root->root_key);
6400         BUG_ON(ret);
6401
6402         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6403                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
6404                                            NULL, NULL);
6405                 BUG_ON(ret < 0);
6406                 if (ret > 0) {
6407                         /* if we fail to delete the orphan item this time
6408                          * around, it'll get picked up the next time.
6409                          *
6410                          * The most common failure here is just -ENOENT.
6411                          */
6412                         btrfs_del_orphan_item(trans, tree_root,
6413                                               root->root_key.objectid);
6414                 }
6415         }
6416
6417         if (root->in_radix) {
6418                 btrfs_free_fs_root(tree_root->fs_info, root);
6419         } else {
6420                 free_extent_buffer(root->node);
6421                 free_extent_buffer(root->commit_root);
6422                 kfree(root);
6423         }
6424 out:
6425         btrfs_end_transaction_throttle(trans, tree_root);
6426         kfree(wc);
6427         btrfs_free_path(path);
6428         return err;
6429 }
6430
6431 /*
6432  * drop subtree rooted at tree block 'node'.
6433  *
6434  * NOTE: this function will unlock and release tree block 'node'
6435  */
6436 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
6437                         struct btrfs_root *root,
6438                         struct extent_buffer *node,
6439                         struct extent_buffer *parent)
6440 {
6441         struct btrfs_path *path;
6442         struct walk_control *wc;
6443         int level;
6444         int parent_level;
6445         int ret = 0;
6446         int wret;
6447
6448         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6449
6450         path = btrfs_alloc_path();
6451         BUG_ON(!path);
6452
6453         wc = kzalloc(sizeof(*wc), GFP_NOFS);
6454         BUG_ON(!wc);
6455
6456         btrfs_assert_tree_locked(parent);
6457         parent_level = btrfs_header_level(parent);
6458         extent_buffer_get(parent);
6459         path->nodes[parent_level] = parent;
6460         path->slots[parent_level] = btrfs_header_nritems(parent);
6461
6462         btrfs_assert_tree_locked(node);
6463         level = btrfs_header_level(node);
6464         path->nodes[level] = node;
6465         path->slots[level] = 0;
6466         path->locks[level] = 1;
6467
6468         wc->refs[parent_level] = 1;
6469         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
6470         wc->level = level;
6471         wc->shared_level = -1;
6472         wc->stage = DROP_REFERENCE;
6473         wc->update_ref = 0;
6474         wc->keep_locks = 1;
6475         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
6476
6477         while (1) {
6478                 wret = walk_down_tree(trans, root, path, wc);
6479                 if (wret < 0) {
6480                         ret = wret;
6481                         break;
6482                 }
6483
6484                 wret = walk_up_tree(trans, root, path, wc, parent_level);
6485                 if (wret < 0)
6486                         ret = wret;
6487                 if (wret != 0)
6488                         break;
6489         }
6490
6491         kfree(wc);
6492         btrfs_free_path(path);
6493         return ret;
6494 }
6495
6496 #if 0
6497 static unsigned long calc_ra(unsigned long start, unsigned long last,
6498                              unsigned long nr)
6499 {
6500         return min(last, start + nr - 1);
6501 }
6502
6503 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
6504                                          u64 len)
6505 {
6506         u64 page_start;
6507         u64 page_end;
6508         unsigned long first_index;
6509         unsigned long last_index;
6510         unsigned long i;
6511         struct page *page;
6512         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6513         struct file_ra_state *ra;
6514         struct btrfs_ordered_extent *ordered;
6515         unsigned int total_read = 0;
6516         unsigned int total_dirty = 0;
6517         int ret = 0;
6518
6519         ra = kzalloc(sizeof(*ra), GFP_NOFS);
6520         if (!ra)
6521                 return -ENOMEM;
6522
6523         mutex_lock(&inode->i_mutex);
6524         first_index = start >> PAGE_CACHE_SHIFT;
6525         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
6526
6527         /* make sure the dirty trick played by the caller work */
6528         ret = invalidate_inode_pages2_range(inode->i_mapping,
6529                                             first_index, last_index);
6530         if (ret)
6531                 goto out_unlock;
6532
6533         file_ra_state_init(ra, inode->i_mapping);
6534
6535         for (i = first_index ; i <= last_index; i++) {
6536                 if (total_read % ra->ra_pages == 0) {
6537                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
6538                                        calc_ra(i, last_index, ra->ra_pages));
6539                 }
6540                 total_read++;
6541 again:
6542                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
6543                         BUG_ON(1);
6544                 page = grab_cache_page(inode->i_mapping, i);
6545                 if (!page) {
6546                         ret = -ENOMEM;
6547                         goto out_unlock;
6548                 }
6549                 if (!PageUptodate(page)) {
6550                         btrfs_readpage(NULL, page);
6551                         lock_page(page);
6552                         if (!PageUptodate(page)) {
6553                                 unlock_page(page);
6554                                 page_cache_release(page);
6555                                 ret = -EIO;
6556                                 goto out_unlock;
6557                         }
6558                 }
6559                 wait_on_page_writeback(page);
6560
6561                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
6562                 page_end = page_start + PAGE_CACHE_SIZE - 1;
6563                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
6564
6565                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6566                 if (ordered) {
6567                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6568                         unlock_page(page);
6569                         page_cache_release(page);
6570                         btrfs_start_ordered_extent(inode, ordered, 1);
6571                         btrfs_put_ordered_extent(ordered);
6572                         goto again;
6573                 }
6574                 set_page_extent_mapped(page);
6575
6576                 if (i == first_index)
6577                         set_extent_bits(io_tree, page_start, page_end,
6578                                         EXTENT_BOUNDARY, GFP_NOFS);
6579                 btrfs_set_extent_delalloc(inode, page_start, page_end);
6580
6581                 set_page_dirty(page);
6582                 total_dirty++;
6583
6584                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
6585                 unlock_page(page);
6586                 page_cache_release(page);
6587         }
6588
6589 out_unlock:
6590         kfree(ra);
6591         mutex_unlock(&inode->i_mutex);
6592         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
6593         return ret;
6594 }
6595
6596 static noinline int relocate_data_extent(struct inode *reloc_inode,
6597                                          struct btrfs_key *extent_key,
6598                                          u64 offset)
6599 {
6600         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6601         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
6602         struct extent_map *em;
6603         u64 start = extent_key->objectid - offset;
6604         u64 end = start + extent_key->offset - 1;
6605
6606         em = alloc_extent_map(GFP_NOFS);
6607         BUG_ON(!em);
6608
6609         em->start = start;
6610         em->len = extent_key->offset;
6611         em->block_len = extent_key->offset;
6612         em->block_start = extent_key->objectid;
6613         em->bdev = root->fs_info->fs_devices->latest_bdev;
6614         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6615
6616         /* setup extent map to cheat btrfs_readpage */
6617         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6618         while (1) {
6619                 int ret;
6620                 write_lock(&em_tree->lock);
6621                 ret = add_extent_mapping(em_tree, em);
6622                 write_unlock(&em_tree->lock);
6623                 if (ret != -EEXIST) {
6624                         free_extent_map(em);
6625                         break;
6626                 }
6627                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
6628         }
6629         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
6630
6631         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
6632 }
6633
6634 struct btrfs_ref_path {
6635         u64 extent_start;
6636         u64 nodes[BTRFS_MAX_LEVEL];
6637         u64 root_objectid;
6638         u64 root_generation;
6639         u64 owner_objectid;
6640         u32 num_refs;
6641         int lowest_level;
6642         int current_level;
6643         int shared_level;
6644
6645         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
6646         u64 new_nodes[BTRFS_MAX_LEVEL];
6647 };
6648
6649 struct disk_extent {
6650         u64 ram_bytes;
6651         u64 disk_bytenr;
6652         u64 disk_num_bytes;
6653         u64 offset;
6654         u64 num_bytes;
6655         u8 compression;
6656         u8 encryption;
6657         u16 other_encoding;
6658 };
6659
6660 static int is_cowonly_root(u64 root_objectid)
6661 {
6662         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
6663             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6664             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
6665             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
6666             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
6667             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
6668                 return 1;
6669         return 0;
6670 }
6671
6672 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
6673                                     struct btrfs_root *extent_root,
6674                                     struct btrfs_ref_path *ref_path,
6675                                     int first_time)
6676 {
6677         struct extent_buffer *leaf;
6678         struct btrfs_path *path;
6679         struct btrfs_extent_ref *ref;
6680         struct btrfs_key key;
6681         struct btrfs_key found_key;
6682         u64 bytenr;
6683         u32 nritems;
6684         int level;
6685         int ret = 1;
6686
6687         path = btrfs_alloc_path();
6688         if (!path)
6689                 return -ENOMEM;
6690
6691         if (first_time) {
6692                 ref_path->lowest_level = -1;
6693                 ref_path->current_level = -1;
6694                 ref_path->shared_level = -1;
6695                 goto walk_up;
6696         }
6697 walk_down:
6698         level = ref_path->current_level - 1;
6699         while (level >= -1) {
6700                 u64 parent;
6701                 if (level < ref_path->lowest_level)
6702                         break;
6703
6704                 if (level >= 0)
6705                         bytenr = ref_path->nodes[level];
6706                 else
6707                         bytenr = ref_path->extent_start;
6708                 BUG_ON(bytenr == 0);
6709
6710                 parent = ref_path->nodes[level + 1];
6711                 ref_path->nodes[level + 1] = 0;
6712                 ref_path->current_level = level;
6713                 BUG_ON(parent == 0);
6714
6715                 key.objectid = bytenr;
6716                 key.offset = parent + 1;
6717                 key.type = BTRFS_EXTENT_REF_KEY;
6718
6719                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6720                 if (ret < 0)
6721                         goto out;
6722                 BUG_ON(ret == 0);
6723
6724                 leaf = path->nodes[0];
6725                 nritems = btrfs_header_nritems(leaf);
6726                 if (path->slots[0] >= nritems) {
6727                         ret = btrfs_next_leaf(extent_root, path);
6728                         if (ret < 0)
6729                                 goto out;
6730                         if (ret > 0)
6731                                 goto next;
6732                         leaf = path->nodes[0];
6733                 }
6734
6735                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6736                 if (found_key.objectid == bytenr &&
6737                     found_key.type == BTRFS_EXTENT_REF_KEY) {
6738                         if (level < ref_path->shared_level)
6739                                 ref_path->shared_level = level;
6740                         goto found;
6741                 }
6742 next:
6743                 level--;
6744                 btrfs_release_path(extent_root, path);
6745                 cond_resched();
6746         }
6747         /* reached lowest level */
6748         ret = 1;
6749         goto out;
6750 walk_up:
6751         level = ref_path->current_level;
6752         while (level < BTRFS_MAX_LEVEL - 1) {
6753                 u64 ref_objectid;
6754
6755                 if (level >= 0)
6756                         bytenr = ref_path->nodes[level];
6757                 else
6758                         bytenr = ref_path->extent_start;
6759
6760                 BUG_ON(bytenr == 0);
6761
6762                 key.objectid = bytenr;
6763                 key.offset = 0;
6764                 key.type = BTRFS_EXTENT_REF_KEY;
6765
6766                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
6767                 if (ret < 0)
6768                         goto out;
6769
6770                 leaf = path->nodes[0];
6771                 nritems = btrfs_header_nritems(leaf);
6772                 if (path->slots[0] >= nritems) {
6773                         ret = btrfs_next_leaf(extent_root, path);
6774                         if (ret < 0)
6775                                 goto out;
6776                         if (ret > 0) {
6777                                 /* the extent was freed by someone */
6778                                 if (ref_path->lowest_level == level)
6779                                         goto out;
6780                                 btrfs_release_path(extent_root, path);
6781                                 goto walk_down;
6782                         }
6783                         leaf = path->nodes[0];
6784                 }
6785
6786                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6787                 if (found_key.objectid != bytenr ||
6788                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
6789                         /* the extent was freed by someone */
6790                         if (ref_path->lowest_level == level) {
6791                                 ret = 1;
6792                                 goto out;
6793                         }
6794                         btrfs_release_path(extent_root, path);
6795                         goto walk_down;
6796                 }
6797 found:
6798                 ref = btrfs_item_ptr(leaf, path->slots[0],
6799                                 struct btrfs_extent_ref);
6800                 ref_objectid = btrfs_ref_objectid(leaf, ref);
6801                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
6802                         if (first_time) {
6803                                 level = (int)ref_objectid;
6804                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
6805                                 ref_path->lowest_level = level;
6806                                 ref_path->current_level = level;
6807                                 ref_path->nodes[level] = bytenr;
6808                         } else {
6809                                 WARN_ON(ref_objectid != level);
6810                         }
6811                 } else {
6812                         WARN_ON(level != -1);
6813                 }
6814                 first_time = 0;
6815
6816                 if (ref_path->lowest_level == level) {
6817                         ref_path->owner_objectid = ref_objectid;
6818                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
6819                 }
6820
6821                 /*
6822                  * the block is tree root or the block isn't in reference
6823                  * counted tree.
6824                  */
6825                 if (found_key.objectid == found_key.offset ||
6826                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
6827                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6828                         ref_path->root_generation =
6829                                 btrfs_ref_generation(leaf, ref);
6830                         if (level < 0) {
6831                                 /* special reference from the tree log */
6832                                 ref_path->nodes[0] = found_key.offset;
6833                                 ref_path->current_level = 0;
6834                         }
6835                         ret = 0;
6836                         goto out;
6837                 }
6838
6839                 level++;
6840                 BUG_ON(ref_path->nodes[level] != 0);
6841                 ref_path->nodes[level] = found_key.offset;
6842                 ref_path->current_level = level;
6843
6844                 /*
6845                  * the reference was created in the running transaction,
6846                  * no need to continue walking up.
6847                  */
6848                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6849                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6850                         ref_path->root_generation =
6851                                 btrfs_ref_generation(leaf, ref);
6852                         ret = 0;
6853                         goto out;
6854                 }
6855
6856                 btrfs_release_path(extent_root, path);
6857                 cond_resched();
6858         }
6859         /* reached max tree level, but no tree root found. */
6860         BUG();
6861 out:
6862         btrfs_free_path(path);
6863         return ret;
6864 }
6865
6866 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6867                                 struct btrfs_root *extent_root,
6868                                 struct btrfs_ref_path *ref_path,
6869                                 u64 extent_start)
6870 {
6871         memset(ref_path, 0, sizeof(*ref_path));
6872         ref_path->extent_start = extent_start;
6873
6874         return __next_ref_path(trans, extent_root, ref_path, 1);
6875 }
6876
6877 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6878                                struct btrfs_root *extent_root,
6879                                struct btrfs_ref_path *ref_path)
6880 {
6881         return __next_ref_path(trans, extent_root, ref_path, 0);
6882 }
6883
6884 static noinline int get_new_locations(struct inode *reloc_inode,
6885                                       struct btrfs_key *extent_key,
6886                                       u64 offset, int no_fragment,
6887                                       struct disk_extent **extents,
6888                                       int *nr_extents)
6889 {
6890         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6891         struct btrfs_path *path;
6892         struct btrfs_file_extent_item *fi;
6893         struct extent_buffer *leaf;
6894         struct disk_extent *exts = *extents;
6895         struct btrfs_key found_key;
6896         u64 cur_pos;
6897         u64 last_byte;
6898         u32 nritems;
6899         int nr = 0;
6900         int max = *nr_extents;
6901         int ret;
6902
6903         WARN_ON(!no_fragment && *extents);
6904         if (!exts) {
6905                 max = 1;
6906                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6907                 if (!exts)
6908                         return -ENOMEM;
6909         }
6910
6911         path = btrfs_alloc_path();
6912         BUG_ON(!path);
6913
6914         cur_pos = extent_key->objectid - offset;
6915         last_byte = extent_key->objectid + extent_key->offset;
6916         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6917                                        cur_pos, 0);
6918         if (ret < 0)
6919                 goto out;
6920         if (ret > 0) {
6921                 ret = -ENOENT;
6922                 goto out;
6923         }
6924
6925         while (1) {
6926                 leaf = path->nodes[0];
6927                 nritems = btrfs_header_nritems(leaf);
6928                 if (path->slots[0] >= nritems) {
6929                         ret = btrfs_next_leaf(root, path);
6930                         if (ret < 0)
6931                                 goto out;
6932                         if (ret > 0)
6933                                 break;
6934                         leaf = path->nodes[0];
6935                 }
6936
6937                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6938                 if (found_key.offset != cur_pos ||
6939                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6940                     found_key.objectid != reloc_inode->i_ino)
6941                         break;
6942
6943                 fi = btrfs_item_ptr(leaf, path->slots[0],
6944                                     struct btrfs_file_extent_item);
6945                 if (btrfs_file_extent_type(leaf, fi) !=
6946                     BTRFS_FILE_EXTENT_REG ||
6947                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6948                         break;
6949
6950                 if (nr == max) {
6951                         struct disk_extent *old = exts;
6952                         max *= 2;
6953                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6954                         memcpy(exts, old, sizeof(*exts) * nr);
6955                         if (old != *extents)
6956                                 kfree(old);
6957                 }
6958
6959                 exts[nr].disk_bytenr =
6960                         btrfs_file_extent_disk_bytenr(leaf, fi);
6961                 exts[nr].disk_num_bytes =
6962                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6963                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6964                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6965                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6966                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6967                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6968                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6969                                                                            fi);
6970                 BUG_ON(exts[nr].offset > 0);
6971                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6972                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6973
6974                 cur_pos += exts[nr].num_bytes;
6975                 nr++;
6976
6977                 if (cur_pos + offset >= last_byte)
6978                         break;
6979
6980                 if (no_fragment) {
6981                         ret = 1;
6982                         goto out;
6983                 }
6984                 path->slots[0]++;
6985         }
6986
6987         BUG_ON(cur_pos + offset > last_byte);
6988         if (cur_pos + offset < last_byte) {
6989                 ret = -ENOENT;
6990                 goto out;
6991         }
6992         ret = 0;
6993 out:
6994         btrfs_free_path(path);
6995         if (ret) {
6996                 if (exts != *extents)
6997                         kfree(exts);
6998         } else {
6999                 *extents = exts;
7000                 *nr_extents = nr;
7001         }
7002         return ret;
7003 }
7004
7005 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
7006                                         struct btrfs_root *root,
7007                                         struct btrfs_path *path,
7008                                         struct btrfs_key *extent_key,
7009                                         struct btrfs_key *leaf_key,
7010                                         struct btrfs_ref_path *ref_path,
7011                                         struct disk_extent *new_extents,
7012                                         int nr_extents)
7013 {
7014         struct extent_buffer *leaf;
7015         struct btrfs_file_extent_item *fi;
7016         struct inode *inode = NULL;
7017         struct btrfs_key key;
7018         u64 lock_start = 0;
7019         u64 lock_end = 0;
7020         u64 num_bytes;
7021         u64 ext_offset;
7022         u64 search_end = (u64)-1;
7023         u32 nritems;
7024         int nr_scaned = 0;
7025         int extent_locked = 0;
7026         int extent_type;
7027         int ret;
7028
7029         memcpy(&key, leaf_key, sizeof(key));
7030         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7031                 if (key.objectid < ref_path->owner_objectid ||
7032                     (key.objectid == ref_path->owner_objectid &&
7033                      key.type < BTRFS_EXTENT_DATA_KEY)) {
7034                         key.objectid = ref_path->owner_objectid;
7035                         key.type = BTRFS_EXTENT_DATA_KEY;
7036                         key.offset = 0;
7037                 }
7038         }
7039
7040         while (1) {
7041                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7042                 if (ret < 0)
7043                         goto out;
7044
7045                 leaf = path->nodes[0];
7046                 nritems = btrfs_header_nritems(leaf);
7047 next:
7048                 if (extent_locked && ret > 0) {
7049                         /*
7050                          * the file extent item was modified by someone
7051                          * before the extent got locked.
7052                          */
7053                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7054                                       lock_end, GFP_NOFS);
7055                         extent_locked = 0;
7056                 }
7057
7058                 if (path->slots[0] >= nritems) {
7059                         if (++nr_scaned > 2)
7060                                 break;
7061
7062                         BUG_ON(extent_locked);
7063                         ret = btrfs_next_leaf(root, path);
7064                         if (ret < 0)
7065                                 goto out;
7066                         if (ret > 0)
7067                                 break;
7068                         leaf = path->nodes[0];
7069                         nritems = btrfs_header_nritems(leaf);
7070                 }
7071
7072                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7073
7074                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
7075                         if ((key.objectid > ref_path->owner_objectid) ||
7076                             (key.objectid == ref_path->owner_objectid &&
7077                              key.type > BTRFS_EXTENT_DATA_KEY) ||
7078                             key.offset >= search_end)
7079                                 break;
7080                 }
7081
7082                 if (inode && key.objectid != inode->i_ino) {
7083                         BUG_ON(extent_locked);
7084                         btrfs_release_path(root, path);
7085                         mutex_unlock(&inode->i_mutex);
7086                         iput(inode);
7087                         inode = NULL;
7088                         continue;
7089                 }
7090
7091                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
7092                         path->slots[0]++;
7093                         ret = 1;
7094                         goto next;
7095                 }
7096                 fi = btrfs_item_ptr(leaf, path->slots[0],
7097                                     struct btrfs_file_extent_item);
7098                 extent_type = btrfs_file_extent_type(leaf, fi);
7099                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
7100                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
7101                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
7102                      extent_key->objectid)) {
7103                         path->slots[0]++;
7104                         ret = 1;
7105                         goto next;
7106                 }
7107
7108                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7109                 ext_offset = btrfs_file_extent_offset(leaf, fi);
7110
7111                 if (search_end == (u64)-1) {
7112                         search_end = key.offset - ext_offset +
7113                                 btrfs_file_extent_ram_bytes(leaf, fi);
7114                 }
7115
7116                 if (!extent_locked) {
7117                         lock_start = key.offset;
7118                         lock_end = lock_start + num_bytes - 1;
7119                 } else {
7120                         if (lock_start > key.offset ||
7121                             lock_end + 1 < key.offset + num_bytes) {
7122                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7123                                               lock_start, lock_end, GFP_NOFS);
7124                                 extent_locked = 0;
7125                         }
7126                 }
7127
7128                 if (!inode) {
7129                         btrfs_release_path(root, path);
7130
7131                         inode = btrfs_iget_locked(root->fs_info->sb,
7132                                                   key.objectid, root);
7133                         if (inode->i_state & I_NEW) {
7134                                 BTRFS_I(inode)->root = root;
7135                                 BTRFS_I(inode)->location.objectid =
7136                                         key.objectid;
7137                                 BTRFS_I(inode)->location.type =
7138                                         BTRFS_INODE_ITEM_KEY;
7139                                 BTRFS_I(inode)->location.offset = 0;
7140                                 btrfs_read_locked_inode(inode);
7141                                 unlock_new_inode(inode);
7142                         }
7143                         /*
7144                          * some code call btrfs_commit_transaction while
7145                          * holding the i_mutex, so we can't use mutex_lock
7146                          * here.
7147                          */
7148                         if (is_bad_inode(inode) ||
7149                             !mutex_trylock(&inode->i_mutex)) {
7150                                 iput(inode);
7151                                 inode = NULL;
7152                                 key.offset = (u64)-1;
7153                                 goto skip;
7154                         }
7155                 }
7156
7157                 if (!extent_locked) {
7158                         struct btrfs_ordered_extent *ordered;
7159
7160                         btrfs_release_path(root, path);
7161
7162                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7163                                     lock_end, GFP_NOFS);
7164                         ordered = btrfs_lookup_first_ordered_extent(inode,
7165                                                                     lock_end);
7166                         if (ordered &&
7167                             ordered->file_offset <= lock_end &&
7168                             ordered->file_offset + ordered->len > lock_start) {
7169                                 unlock_extent(&BTRFS_I(inode)->io_tree,
7170                                               lock_start, lock_end, GFP_NOFS);
7171                                 btrfs_start_ordered_extent(inode, ordered, 1);
7172                                 btrfs_put_ordered_extent(ordered);
7173                                 key.offset += num_bytes;
7174                                 goto skip;
7175                         }
7176                         if (ordered)
7177                                 btrfs_put_ordered_extent(ordered);
7178
7179                         extent_locked = 1;
7180                         continue;
7181                 }
7182
7183                 if (nr_extents == 1) {
7184                         /* update extent pointer in place */
7185                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
7186                                                 new_extents[0].disk_bytenr);
7187                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7188                                                 new_extents[0].disk_num_bytes);
7189                         btrfs_mark_buffer_dirty(leaf);
7190
7191                         btrfs_drop_extent_cache(inode, key.offset,
7192                                                 key.offset + num_bytes - 1, 0);
7193
7194                         ret = btrfs_inc_extent_ref(trans, root,
7195                                                 new_extents[0].disk_bytenr,
7196                                                 new_extents[0].disk_num_bytes,
7197                                                 leaf->start,
7198                                                 root->root_key.objectid,
7199                                                 trans->transid,
7200                                                 key.objectid);
7201                         BUG_ON(ret);
7202
7203                         ret = btrfs_free_extent(trans, root,
7204                                                 extent_key->objectid,
7205                                                 extent_key->offset,
7206                                                 leaf->start,
7207                                                 btrfs_header_owner(leaf),
7208                                                 btrfs_header_generation(leaf),
7209                                                 key.objectid, 0);
7210                         BUG_ON(ret);
7211
7212                         btrfs_release_path(root, path);
7213                         key.offset += num_bytes;
7214                 } else {
7215                         BUG_ON(1);
7216 #if 0
7217                         u64 alloc_hint;
7218                         u64 extent_len;
7219                         int i;
7220                         /*
7221                          * drop old extent pointer at first, then insert the
7222                          * new pointers one bye one
7223                          */
7224                         btrfs_release_path(root, path);
7225                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
7226                                                  key.offset + num_bytes,
7227                                                  key.offset, &alloc_hint);
7228                         BUG_ON(ret);
7229
7230                         for (i = 0; i < nr_extents; i++) {
7231                                 if (ext_offset >= new_extents[i].num_bytes) {
7232                                         ext_offset -= new_extents[i].num_bytes;
7233                                         continue;
7234                                 }
7235                                 extent_len = min(new_extents[i].num_bytes -
7236                                                  ext_offset, num_bytes);
7237
7238                                 ret = btrfs_insert_empty_item(trans, root,
7239                                                               path, &key,
7240                                                               sizeof(*fi));
7241                                 BUG_ON(ret);
7242
7243                                 leaf = path->nodes[0];
7244                                 fi = btrfs_item_ptr(leaf, path->slots[0],
7245                                                 struct btrfs_file_extent_item);
7246                                 btrfs_set_file_extent_generation(leaf, fi,
7247                                                         trans->transid);
7248                                 btrfs_set_file_extent_type(leaf, fi,
7249                                                         BTRFS_FILE_EXTENT_REG);
7250                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7251                                                 new_extents[i].disk_bytenr);
7252                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7253                                                 new_extents[i].disk_num_bytes);
7254                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
7255                                                 new_extents[i].ram_bytes);
7256
7257                                 btrfs_set_file_extent_compression(leaf, fi,
7258                                                 new_extents[i].compression);
7259                                 btrfs_set_file_extent_encryption(leaf, fi,
7260                                                 new_extents[i].encryption);
7261                                 btrfs_set_file_extent_other_encoding(leaf, fi,
7262                                                 new_extents[i].other_encoding);
7263
7264                                 btrfs_set_file_extent_num_bytes(leaf, fi,
7265                                                         extent_len);
7266                                 ext_offset += new_extents[i].offset;
7267                                 btrfs_set_file_extent_offset(leaf, fi,
7268                                                         ext_offset);
7269                                 btrfs_mark_buffer_dirty(leaf);
7270
7271                                 btrfs_drop_extent_cache(inode, key.offset,
7272                                                 key.offset + extent_len - 1, 0);
7273
7274                                 ret = btrfs_inc_extent_ref(trans, root,
7275                                                 new_extents[i].disk_bytenr,
7276                                                 new_extents[i].disk_num_bytes,
7277                                                 leaf->start,
7278                                                 root->root_key.objectid,
7279                                                 trans->transid, key.objectid);
7280                                 BUG_ON(ret);
7281                                 btrfs_release_path(root, path);
7282
7283                                 inode_add_bytes(inode, extent_len);
7284
7285                                 ext_offset = 0;
7286                                 num_bytes -= extent_len;
7287                                 key.offset += extent_len;
7288
7289                                 if (num_bytes == 0)
7290                                         break;
7291                         }
7292                         BUG_ON(i >= nr_extents);
7293 #endif
7294                 }
7295
7296                 if (extent_locked) {
7297                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7298                                       lock_end, GFP_NOFS);
7299                         extent_locked = 0;
7300                 }
7301 skip:
7302                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
7303                     key.offset >= search_end)
7304                         break;
7305
7306                 cond_resched();
7307         }
7308         ret = 0;
7309 out:
7310         btrfs_release_path(root, path);
7311         if (inode) {
7312                 mutex_unlock(&inode->i_mutex);
7313                 if (extent_locked) {
7314                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
7315                                       lock_end, GFP_NOFS);
7316                 }
7317                 iput(inode);
7318         }
7319         return ret;
7320 }
7321
7322 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
7323                                struct btrfs_root *root,
7324                                struct extent_buffer *buf, u64 orig_start)
7325 {
7326         int level;
7327         int ret;
7328
7329         BUG_ON(btrfs_header_generation(buf) != trans->transid);
7330         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
7331
7332         level = btrfs_header_level(buf);
7333         if (level == 0) {
7334                 struct btrfs_leaf_ref *ref;
7335                 struct btrfs_leaf_ref *orig_ref;
7336
7337                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
7338                 if (!orig_ref)
7339                         return -ENOENT;
7340
7341                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
7342                 if (!ref) {
7343                         btrfs_free_leaf_ref(root, orig_ref);
7344                         return -ENOMEM;
7345                 }
7346
7347                 ref->nritems = orig_ref->nritems;
7348                 memcpy(ref->extents, orig_ref->extents,
7349                         sizeof(ref->extents[0]) * ref->nritems);
7350
7351                 btrfs_free_leaf_ref(root, orig_ref);
7352
7353                 ref->root_gen = trans->transid;
7354                 ref->bytenr = buf->start;
7355                 ref->owner = btrfs_header_owner(buf);
7356                 ref->generation = btrfs_header_generation(buf);
7357
7358                 ret = btrfs_add_leaf_ref(root, ref, 0);
7359                 WARN_ON(ret);
7360                 btrfs_free_leaf_ref(root, ref);
7361         }
7362         return 0;
7363 }
7364
7365 static noinline int invalidate_extent_cache(struct btrfs_root *root,
7366                                         struct extent_buffer *leaf,
7367                                         struct btrfs_block_group_cache *group,
7368                                         struct btrfs_root *target_root)
7369 {
7370         struct btrfs_key key;
7371         struct inode *inode = NULL;
7372         struct btrfs_file_extent_item *fi;
7373         struct extent_state *cached_state = NULL;
7374         u64 num_bytes;
7375         u64 skip_objectid = 0;
7376         u32 nritems;
7377         u32 i;
7378
7379         nritems = btrfs_header_nritems(leaf);
7380         for (i = 0; i < nritems; i++) {
7381                 btrfs_item_key_to_cpu(leaf, &key, i);
7382                 if (key.objectid == skip_objectid ||
7383                     key.type != BTRFS_EXTENT_DATA_KEY)
7384                         continue;
7385                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7386                 if (btrfs_file_extent_type(leaf, fi) ==
7387                     BTRFS_FILE_EXTENT_INLINE)
7388                         continue;
7389                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
7390                         continue;
7391                 if (!inode || inode->i_ino != key.objectid) {
7392                         iput(inode);
7393                         inode = btrfs_ilookup(target_root->fs_info->sb,
7394                                               key.objectid, target_root, 1);
7395                 }
7396                 if (!inode) {
7397                         skip_objectid = key.objectid;
7398                         continue;
7399                 }
7400                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
7401
7402                 lock_extent_bits(&BTRFS_I(inode)->io_tree, key.offset,
7403                                  key.offset + num_bytes - 1, 0, &cached_state,
7404                                  GFP_NOFS);
7405                 btrfs_drop_extent_cache(inode, key.offset,
7406                                         key.offset + num_bytes - 1, 1);
7407                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, key.offset,
7408                                      key.offset + num_bytes - 1, &cached_state,
7409                                      GFP_NOFS);
7410                 cond_resched();
7411         }
7412         iput(inode);
7413         return 0;
7414 }
7415
7416 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
7417                                         struct btrfs_root *root,
7418                                         struct extent_buffer *leaf,
7419                                         struct btrfs_block_group_cache *group,
7420                                         struct inode *reloc_inode)
7421 {
7422         struct btrfs_key key;
7423         struct btrfs_key extent_key;
7424         struct btrfs_file_extent_item *fi;
7425         struct btrfs_leaf_ref *ref;
7426         struct disk_extent *new_extent;
7427         u64 bytenr;
7428         u64 num_bytes;
7429         u32 nritems;
7430         u32 i;
7431         int ext_index;
7432         int nr_extent;
7433         int ret;
7434
7435         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
7436         BUG_ON(!new_extent);
7437
7438         ref = btrfs_lookup_leaf_ref(root, leaf->start);
7439         BUG_ON(!ref);
7440
7441         ext_index = -1;
7442         nritems = btrfs_header_nritems(leaf);
7443         for (i = 0; i < nritems; i++) {
7444                 btrfs_item_key_to_cpu(leaf, &key, i);
7445                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
7446                         continue;
7447                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
7448                 if (btrfs_file_extent_type(leaf, fi) ==
7449                     BTRFS_FILE_EXTENT_INLINE)
7450                         continue;
7451                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7452                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
7453                 if (bytenr == 0)
7454                         continue;
7455
7456                 ext_index++;
7457                 if (bytenr >= group->key.objectid + group->key.offset ||
7458                     bytenr + num_bytes <= group->key.objectid)
7459                         continue;
7460
7461                 extent_key.objectid = bytenr;
7462                 extent_key.offset = num_bytes;
7463                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
7464                 nr_extent = 1;
7465                 ret = get_new_locations(reloc_inode, &extent_key,
7466                                         group->key.objectid, 1,
7467                                         &new_extent, &nr_extent);
7468                 if (ret > 0)
7469                         continue;
7470                 BUG_ON(ret < 0);
7471
7472                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
7473                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
7474                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
7475                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
7476
7477                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
7478                                                 new_extent->disk_bytenr);
7479                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
7480                                                 new_extent->disk_num_bytes);
7481                 btrfs_mark_buffer_dirty(leaf);
7482
7483                 ret = btrfs_inc_extent_ref(trans, root,
7484                                         new_extent->disk_bytenr,
7485                                         new_extent->disk_num_bytes,
7486                                         leaf->start,
7487                                         root->root_key.objectid,
7488                                         trans->transid, key.objectid);
7489                 BUG_ON(ret);
7490
7491                 ret = btrfs_free_extent(trans, root,
7492                                         bytenr, num_bytes, leaf->start,
7493                                         btrfs_header_owner(leaf),
7494                                         btrfs_header_generation(leaf),
7495                                         key.objectid, 0);
7496                 BUG_ON(ret);
7497                 cond_resched();
7498         }
7499         kfree(new_extent);
7500         BUG_ON(ext_index + 1 != ref->nritems);
7501         btrfs_free_leaf_ref(root, ref);
7502         return 0;
7503 }
7504
7505 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
7506                           struct btrfs_root *root)
7507 {
7508         struct btrfs_root *reloc_root;
7509         int ret;
7510
7511         if (root->reloc_root) {
7512                 reloc_root = root->reloc_root;
7513                 root->reloc_root = NULL;
7514                 list_add(&reloc_root->dead_list,
7515                          &root->fs_info->dead_reloc_roots);
7516
7517                 btrfs_set_root_bytenr(&reloc_root->root_item,
7518                                       reloc_root->node->start);
7519                 btrfs_set_root_level(&root->root_item,
7520                                      btrfs_header_level(reloc_root->node));
7521                 memset(&reloc_root->root_item.drop_progress, 0,
7522                         sizeof(struct btrfs_disk_key));
7523                 reloc_root->root_item.drop_level = 0;
7524
7525                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7526                                         &reloc_root->root_key,
7527                                         &reloc_root->root_item);
7528                 BUG_ON(ret);
7529         }
7530         return 0;
7531 }
7532
7533 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
7534 {
7535         struct btrfs_trans_handle *trans;
7536         struct btrfs_root *reloc_root;
7537         struct btrfs_root *prev_root = NULL;
7538         struct list_head dead_roots;
7539         int ret;
7540         unsigned long nr;
7541
7542         INIT_LIST_HEAD(&dead_roots);
7543         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
7544
7545         while (!list_empty(&dead_roots)) {
7546                 reloc_root = list_entry(dead_roots.prev,
7547                                         struct btrfs_root, dead_list);
7548                 list_del_init(&reloc_root->dead_list);
7549
7550                 BUG_ON(reloc_root->commit_root != NULL);
7551                 while (1) {
7552                         trans = btrfs_join_transaction(root, 1);
7553                         BUG_ON(IS_ERR(trans));
7554
7555                         mutex_lock(&root->fs_info->drop_mutex);
7556                         ret = btrfs_drop_snapshot(trans, reloc_root);
7557                         if (ret != -EAGAIN)
7558                                 break;
7559                         mutex_unlock(&root->fs_info->drop_mutex);
7560
7561                         nr = trans->blocks_used;
7562                         ret = btrfs_end_transaction(trans, root);
7563                         BUG_ON(ret);
7564                         btrfs_btree_balance_dirty(root, nr);
7565                 }
7566
7567                 free_extent_buffer(reloc_root->node);
7568
7569                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
7570                                      &reloc_root->root_key);
7571                 BUG_ON(ret);
7572                 mutex_unlock(&root->fs_info->drop_mutex);
7573
7574                 nr = trans->blocks_used;
7575                 ret = btrfs_end_transaction(trans, root);
7576                 BUG_ON(ret);
7577                 btrfs_btree_balance_dirty(root, nr);
7578
7579                 kfree(prev_root);
7580                 prev_root = reloc_root;
7581         }
7582         if (prev_root) {
7583                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
7584                 kfree(prev_root);
7585         }
7586         return 0;
7587 }
7588
7589 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
7590 {
7591         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
7592         return 0;
7593 }
7594
7595 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
7596 {
7597         struct btrfs_root *reloc_root;
7598         struct btrfs_trans_handle *trans;
7599         struct btrfs_key location;
7600         int found;
7601         int ret;
7602
7603         mutex_lock(&root->fs_info->tree_reloc_mutex);
7604         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
7605         BUG_ON(ret);
7606         found = !list_empty(&root->fs_info->dead_reloc_roots);
7607         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7608
7609         if (found) {
7610                 trans = btrfs_start_transaction(root, 1);
7611                 BUG_ON(IS_ERR(trans));
7612                 ret = btrfs_commit_transaction(trans, root);
7613                 BUG_ON(ret);
7614         }
7615
7616         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7617         location.offset = (u64)-1;
7618         location.type = BTRFS_ROOT_ITEM_KEY;
7619
7620         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
7621         BUG_ON(!reloc_root);
7622         ret = btrfs_orphan_cleanup(reloc_root);
7623         BUG_ON(ret);
7624         return 0;
7625 }
7626
7627 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
7628                                     struct btrfs_root *root)
7629 {
7630         struct btrfs_root *reloc_root;
7631         struct extent_buffer *eb;
7632         struct btrfs_root_item *root_item;
7633         struct btrfs_key root_key;
7634         int ret;
7635
7636         BUG_ON(!root->ref_cows);
7637         if (root->reloc_root)
7638                 return 0;
7639
7640         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
7641         BUG_ON(!root_item);
7642
7643         ret = btrfs_copy_root(trans, root, root->commit_root,
7644                               &eb, BTRFS_TREE_RELOC_OBJECTID);
7645         BUG_ON(ret);
7646
7647         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7648         root_key.offset = root->root_key.objectid;
7649         root_key.type = BTRFS_ROOT_ITEM_KEY;
7650
7651         memcpy(root_item, &root->root_item, sizeof(root_item));
7652         btrfs_set_root_refs(root_item, 0);
7653         btrfs_set_root_bytenr(root_item, eb->start);
7654         btrfs_set_root_level(root_item, btrfs_header_level(eb));
7655         btrfs_set_root_generation(root_item, trans->transid);
7656
7657         btrfs_tree_unlock(eb);
7658         free_extent_buffer(eb);
7659
7660         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
7661                                 &root_key, root_item);
7662         BUG_ON(ret);
7663         kfree(root_item);
7664
7665         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
7666                                                  &root_key);
7667         BUG_ON(!reloc_root);
7668         reloc_root->last_trans = trans->transid;
7669         reloc_root->commit_root = NULL;
7670         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
7671
7672         root->reloc_root = reloc_root;
7673         return 0;
7674 }
7675
7676 /*
7677  * Core function of space balance.
7678  *
7679  * The idea is using reloc trees to relocate tree blocks in reference
7680  * counted roots. There is one reloc tree for each subvol, and all
7681  * reloc trees share same root key objectid. Reloc trees are snapshots
7682  * of the latest committed roots of subvols (root->commit_root).
7683  *
7684  * To relocate a tree block referenced by a subvol, there are two steps.
7685  * COW the block through subvol's reloc tree, then update block pointer
7686  * in the subvol to point to the new block. Since all reloc trees share
7687  * same root key objectid, doing special handing for tree blocks owned
7688  * by them is easy. Once a tree block has been COWed in one reloc tree,
7689  * we can use the resulting new block directly when the same block is
7690  * required to COW again through other reloc trees. By this way, relocated
7691  * tree blocks are shared between reloc trees, so they are also shared
7692  * between subvols.
7693  */
7694 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
7695                                       struct btrfs_root *root,
7696                                       struct btrfs_path *path,
7697                                       struct btrfs_key *first_key,
7698                                       struct btrfs_ref_path *ref_path,
7699                                       struct btrfs_block_group_cache *group,
7700                                       struct inode *reloc_inode)
7701 {
7702         struct btrfs_root *reloc_root;
7703         struct extent_buffer *eb = NULL;
7704         struct btrfs_key *keys;
7705         u64 *nodes;
7706         int level;
7707         int shared_level;
7708         int lowest_level = 0;
7709         int ret;
7710
7711         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
7712                 lowest_level = ref_path->owner_objectid;
7713
7714         if (!root->ref_cows) {
7715                 path->lowest_level = lowest_level;
7716                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
7717                 BUG_ON(ret < 0);
7718                 path->lowest_level = 0;
7719                 btrfs_release_path(root, path);
7720                 return 0;
7721         }
7722
7723         mutex_lock(&root->fs_info->tree_reloc_mutex);
7724         ret = init_reloc_tree(trans, root);
7725         BUG_ON(ret);
7726         reloc_root = root->reloc_root;
7727
7728         shared_level = ref_path->shared_level;
7729         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
7730
7731         keys = ref_path->node_keys;
7732         nodes = ref_path->new_nodes;
7733         memset(&keys[shared_level + 1], 0,
7734                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
7735         memset(&nodes[shared_level + 1], 0,
7736                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
7737
7738         if (nodes[lowest_level] == 0) {
7739                 path->lowest_level = lowest_level;
7740                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7741                                         0, 1);
7742                 BUG_ON(ret);
7743                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
7744                         eb = path->nodes[level];
7745                         if (!eb || eb == reloc_root->node)
7746                                 break;
7747                         nodes[level] = eb->start;
7748                         if (level == 0)
7749                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
7750                         else
7751                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
7752                 }
7753                 if (nodes[0] &&
7754                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7755                         eb = path->nodes[0];
7756                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
7757                                                       group, reloc_inode);
7758                         BUG_ON(ret);
7759                 }
7760                 btrfs_release_path(reloc_root, path);
7761         } else {
7762                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
7763                                        lowest_level);
7764                 BUG_ON(ret);
7765         }
7766
7767         /*
7768          * replace tree blocks in the fs tree with tree blocks in
7769          * the reloc tree.
7770          */
7771         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
7772         BUG_ON(ret < 0);
7773
7774         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7775                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
7776                                         0, 0);
7777                 BUG_ON(ret);
7778                 extent_buffer_get(path->nodes[0]);
7779                 eb = path->nodes[0];
7780                 btrfs_release_path(reloc_root, path);
7781                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
7782                 BUG_ON(ret);
7783                 free_extent_buffer(eb);
7784         }
7785
7786         mutex_unlock(&root->fs_info->tree_reloc_mutex);
7787         path->lowest_level = 0;
7788         return 0;
7789 }
7790
7791 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
7792                                         struct btrfs_root *root,
7793                                         struct btrfs_path *path,
7794                                         struct btrfs_key *first_key,
7795                                         struct btrfs_ref_path *ref_path)
7796 {
7797         int ret;
7798
7799         ret = relocate_one_path(trans, root, path, first_key,
7800                                 ref_path, NULL, NULL);
7801         BUG_ON(ret);
7802
7803         return 0;
7804 }
7805
7806 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
7807                                     struct btrfs_root *extent_root,
7808                                     struct btrfs_path *path,
7809                                     struct btrfs_key *extent_key)
7810 {
7811         int ret;
7812
7813         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
7814         if (ret)
7815                 goto out;
7816         ret = btrfs_del_item(trans, extent_root, path);
7817 out:
7818         btrfs_release_path(extent_root, path);
7819         return ret;
7820 }
7821
7822 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
7823                                                 struct btrfs_ref_path *ref_path)
7824 {
7825         struct btrfs_key root_key;
7826
7827         root_key.objectid = ref_path->root_objectid;
7828         root_key.type = BTRFS_ROOT_ITEM_KEY;
7829         if (is_cowonly_root(ref_path->root_objectid))
7830                 root_key.offset = 0;
7831         else
7832                 root_key.offset = (u64)-1;
7833
7834         return btrfs_read_fs_root_no_name(fs_info, &root_key);
7835 }
7836
7837 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
7838                                         struct btrfs_path *path,
7839                                         struct btrfs_key *extent_key,
7840                                         struct btrfs_block_group_cache *group,
7841                                         struct inode *reloc_inode, int pass)
7842 {
7843         struct btrfs_trans_handle *trans;
7844         struct btrfs_root *found_root;
7845         struct btrfs_ref_path *ref_path = NULL;
7846         struct disk_extent *new_extents = NULL;
7847         int nr_extents = 0;
7848         int loops;
7849         int ret;
7850         int level;
7851         struct btrfs_key first_key;
7852         u64 prev_block = 0;
7853
7854
7855         trans = btrfs_start_transaction(extent_root, 1);
7856         BUG_ON(IS_ERR(trans));
7857
7858         if (extent_key->objectid == 0) {
7859                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7860                 goto out;
7861         }
7862
7863         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7864         if (!ref_path) {
7865                 ret = -ENOMEM;
7866                 goto out;
7867         }
7868
7869         for (loops = 0; ; loops++) {
7870                 if (loops == 0) {
7871                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7872                                                    extent_key->objectid);
7873                 } else {
7874                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7875                 }
7876                 if (ret < 0)
7877                         goto out;
7878                 if (ret > 0)
7879                         break;
7880
7881                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7882                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7883                         continue;
7884
7885                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7886                 BUG_ON(!found_root);
7887                 /*
7888                  * for reference counted tree, only process reference paths
7889                  * rooted at the latest committed root.
7890                  */
7891                 if (found_root->ref_cows &&
7892                     ref_path->root_generation != found_root->root_key.offset)
7893                         continue;
7894
7895                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7896                         if (pass == 0) {
7897                                 /*
7898                                  * copy data extents to new locations
7899                                  */
7900                                 u64 group_start = group->key.objectid;
7901                                 ret = relocate_data_extent(reloc_inode,
7902                                                            extent_key,
7903                                                            group_start);
7904                                 if (ret < 0)
7905                                         goto out;
7906                                 break;
7907                         }
7908                         level = 0;
7909                 } else {
7910                         level = ref_path->owner_objectid;
7911                 }
7912
7913                 if (prev_block != ref_path->nodes[level]) {
7914                         struct extent_buffer *eb;
7915                         u64 block_start = ref_path->nodes[level];
7916                         u64 block_size = btrfs_level_size(found_root, level);
7917
7918                         eb = read_tree_block(found_root, block_start,
7919                                              block_size, 0);
7920                         btrfs_tree_lock(eb);
7921                         BUG_ON(level != btrfs_header_level(eb));
7922
7923                         if (level == 0)
7924                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7925                         else
7926                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7927
7928                         btrfs_tree_unlock(eb);
7929                         free_extent_buffer(eb);
7930                         prev_block = block_start;
7931                 }
7932
7933                 mutex_lock(&extent_root->fs_info->trans_mutex);
7934                 btrfs_record_root_in_trans(found_root);
7935                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7936                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7937                         /*
7938                          * try to update data extent references while
7939                          * keeping metadata shared between snapshots.
7940                          */
7941                         if (pass == 1) {
7942                                 ret = relocate_one_path(trans, found_root,
7943                                                 path, &first_key, ref_path,
7944                                                 group, reloc_inode);
7945                                 if (ret < 0)
7946                                         goto out;
7947                                 continue;
7948                         }
7949                         /*
7950                          * use fallback method to process the remaining
7951                          * references.
7952                          */
7953                         if (!new_extents) {
7954                                 u64 group_start = group->key.objectid;
7955                                 new_extents = kmalloc(sizeof(*new_extents),
7956                                                       GFP_NOFS);
7957                                 nr_extents = 1;
7958                                 ret = get_new_locations(reloc_inode,
7959                                                         extent_key,
7960                                                         group_start, 1,
7961                                                         &new_extents,
7962                                                         &nr_extents);
7963                                 if (ret)
7964                                         goto out;
7965                         }
7966                         ret = replace_one_extent(trans, found_root,
7967                                                 path, extent_key,
7968                                                 &first_key, ref_path,
7969                                                 new_extents, nr_extents);
7970                 } else {
7971                         ret = relocate_tree_block(trans, found_root, path,
7972                                                   &first_key, ref_path);
7973                 }
7974                 if (ret < 0)
7975                         goto out;
7976         }
7977         ret = 0;
7978 out:
7979         btrfs_end_transaction(trans, extent_root);
7980         kfree(new_extents);
7981         kfree(ref_path);
7982         return ret;
7983 }
7984 #endif
7985
7986 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7987 {
7988         u64 num_devices;
7989         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7990                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7991
7992         /*
7993          * we add in the count of missing devices because we want
7994          * to make sure that any RAID levels on a degraded FS
7995          * continue to be honored.
7996          */
7997         num_devices = root->fs_info->fs_devices->rw_devices +
7998                 root->fs_info->fs_devices->missing_devices;
7999
8000         if (num_devices == 1) {
8001                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8002                 stripped = flags & ~stripped;
8003
8004                 /* turn raid0 into single device chunks */
8005                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8006                         return stripped;
8007
8008                 /* turn mirroring into duplication */
8009                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8010                              BTRFS_BLOCK_GROUP_RAID10))
8011                         return stripped | BTRFS_BLOCK_GROUP_DUP;
8012                 return flags;
8013         } else {
8014                 /* they already had raid on here, just return */
8015                 if (flags & stripped)
8016                         return flags;
8017
8018                 stripped |= BTRFS_BLOCK_GROUP_DUP;
8019                 stripped = flags & ~stripped;
8020
8021                 /* switch duplicated blocks with raid1 */
8022                 if (flags & BTRFS_BLOCK_GROUP_DUP)
8023                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
8024
8025                 /* turn single device chunks into raid0 */
8026                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
8027         }
8028         return flags;
8029 }
8030
8031 static int set_block_group_ro(struct btrfs_block_group_cache *cache)
8032 {
8033         struct btrfs_space_info *sinfo = cache->space_info;
8034         u64 num_bytes;
8035         int ret = -ENOSPC;
8036
8037         if (cache->ro)
8038                 return 0;
8039
8040         spin_lock(&sinfo->lock);
8041         spin_lock(&cache->lock);
8042         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8043                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8044
8045         if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8046             sinfo->bytes_may_use + sinfo->bytes_readonly +
8047             cache->reserved_pinned + num_bytes <= sinfo->total_bytes) {
8048                 sinfo->bytes_readonly += num_bytes;
8049                 sinfo->bytes_reserved += cache->reserved_pinned;
8050                 cache->reserved_pinned = 0;
8051                 cache->ro = 1;
8052                 ret = 0;
8053         }
8054
8055         spin_unlock(&cache->lock);
8056         spin_unlock(&sinfo->lock);
8057         return ret;
8058 }
8059
8060 int btrfs_set_block_group_ro(struct btrfs_root *root,
8061                              struct btrfs_block_group_cache *cache)
8062
8063 {
8064         struct btrfs_trans_handle *trans;
8065         u64 alloc_flags;
8066         int ret;
8067
8068         BUG_ON(cache->ro);
8069
8070         trans = btrfs_join_transaction(root, 1);
8071         BUG_ON(IS_ERR(trans));
8072
8073         alloc_flags = update_block_group_flags(root, cache->flags);
8074         if (alloc_flags != cache->flags)
8075                 do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8076
8077         ret = set_block_group_ro(cache);
8078         if (!ret)
8079                 goto out;
8080         alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8081         ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8082         if (ret < 0)
8083                 goto out;
8084         ret = set_block_group_ro(cache);
8085 out:
8086         btrfs_end_transaction(trans, root);
8087         return ret;
8088 }
8089
8090 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8091                             struct btrfs_root *root, u64 type)
8092 {
8093         u64 alloc_flags = get_alloc_profile(root, type);
8094         return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, 1);
8095 }
8096
8097 /*
8098  * helper to account the unused space of all the readonly block group in the
8099  * list. takes mirrors into account.
8100  */
8101 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8102 {
8103         struct btrfs_block_group_cache *block_group;
8104         u64 free_bytes = 0;
8105         int factor;
8106
8107         list_for_each_entry(block_group, groups_list, list) {
8108                 spin_lock(&block_group->lock);
8109
8110                 if (!block_group->ro) {
8111                         spin_unlock(&block_group->lock);
8112                         continue;
8113                 }
8114
8115                 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8116                                           BTRFS_BLOCK_GROUP_RAID10 |
8117                                           BTRFS_BLOCK_GROUP_DUP))
8118                         factor = 2;
8119                 else
8120                         factor = 1;
8121
8122                 free_bytes += (block_group->key.offset -
8123                                btrfs_block_group_used(&block_group->item)) *
8124                                factor;
8125
8126                 spin_unlock(&block_group->lock);
8127         }
8128
8129         return free_bytes;
8130 }
8131
8132 /*
8133  * helper to account the unused space of all the readonly block group in the
8134  * space_info. takes mirrors into account.
8135  */
8136 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8137 {
8138         int i;
8139         u64 free_bytes = 0;
8140
8141         spin_lock(&sinfo->lock);
8142
8143         for(i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8144                 if (!list_empty(&sinfo->block_groups[i]))
8145                         free_bytes += __btrfs_get_ro_block_group_free_space(
8146                                                 &sinfo->block_groups[i]);
8147
8148         spin_unlock(&sinfo->lock);
8149
8150         return free_bytes;
8151 }
8152
8153 int btrfs_set_block_group_rw(struct btrfs_root *root,
8154                               struct btrfs_block_group_cache *cache)
8155 {
8156         struct btrfs_space_info *sinfo = cache->space_info;
8157         u64 num_bytes;
8158
8159         BUG_ON(!cache->ro);
8160
8161         spin_lock(&sinfo->lock);
8162         spin_lock(&cache->lock);
8163         num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8164                     cache->bytes_super - btrfs_block_group_used(&cache->item);
8165         sinfo->bytes_readonly -= num_bytes;
8166         cache->ro = 0;
8167         spin_unlock(&cache->lock);
8168         spin_unlock(&sinfo->lock);
8169         return 0;
8170 }
8171
8172 /*
8173  * checks to see if its even possible to relocate this block group.
8174  *
8175  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8176  * ok to go ahead and try.
8177  */
8178 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8179 {
8180         struct btrfs_block_group_cache *block_group;
8181         struct btrfs_space_info *space_info;
8182         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8183         struct btrfs_device *device;
8184         int full = 0;
8185         int ret = 0;
8186
8187         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8188
8189         /* odd, couldn't find the block group, leave it alone */
8190         if (!block_group)
8191                 return -1;
8192
8193         /* no bytes used, we're good */
8194         if (!btrfs_block_group_used(&block_group->item))
8195                 goto out;
8196
8197         space_info = block_group->space_info;
8198         spin_lock(&space_info->lock);
8199
8200         full = space_info->full;
8201
8202         /*
8203          * if this is the last block group we have in this space, we can't
8204          * relocate it unless we're able to allocate a new chunk below.
8205          *
8206          * Otherwise, we need to make sure we have room in the space to handle
8207          * all of the extents from this block group.  If we can, we're good
8208          */
8209         if ((space_info->total_bytes != block_group->key.offset) &&
8210            (space_info->bytes_used + space_info->bytes_reserved +
8211             space_info->bytes_pinned + space_info->bytes_readonly +
8212             btrfs_block_group_used(&block_group->item) <
8213             space_info->total_bytes)) {
8214                 spin_unlock(&space_info->lock);
8215                 goto out;
8216         }
8217         spin_unlock(&space_info->lock);
8218
8219         /*
8220          * ok we don't have enough space, but maybe we have free space on our
8221          * devices to allocate new chunks for relocation, so loop through our
8222          * alloc devices and guess if we have enough space.  However, if we
8223          * were marked as full, then we know there aren't enough chunks, and we
8224          * can just return.
8225          */
8226         ret = -1;
8227         if (full)
8228                 goto out;
8229
8230         mutex_lock(&root->fs_info->chunk_mutex);
8231         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8232                 u64 min_free = btrfs_block_group_used(&block_group->item);
8233                 u64 dev_offset;
8234
8235                 /*
8236                  * check to make sure we can actually find a chunk with enough
8237                  * space to fit our block group in.
8238                  */
8239                 if (device->total_bytes > device->bytes_used + min_free) {
8240                         ret = find_free_dev_extent(NULL, device, min_free,
8241                                                    &dev_offset, NULL);
8242                         if (!ret)
8243                                 break;
8244                         ret = -1;
8245                 }
8246         }
8247         mutex_unlock(&root->fs_info->chunk_mutex);
8248 out:
8249         btrfs_put_block_group(block_group);
8250         return ret;
8251 }
8252
8253 static int find_first_block_group(struct btrfs_root *root,
8254                 struct btrfs_path *path, struct btrfs_key *key)
8255 {
8256         int ret = 0;
8257         struct btrfs_key found_key;
8258         struct extent_buffer *leaf;
8259         int slot;
8260
8261         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8262         if (ret < 0)
8263                 goto out;
8264
8265         while (1) {
8266                 slot = path->slots[0];
8267                 leaf = path->nodes[0];
8268                 if (slot >= btrfs_header_nritems(leaf)) {
8269                         ret = btrfs_next_leaf(root, path);
8270                         if (ret == 0)
8271                                 continue;
8272                         if (ret < 0)
8273                                 goto out;
8274                         break;
8275                 }
8276                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8277
8278                 if (found_key.objectid >= key->objectid &&
8279                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8280                         ret = 0;
8281                         goto out;
8282                 }
8283                 path->slots[0]++;
8284         }
8285 out:
8286         return ret;
8287 }
8288
8289 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8290 {
8291         struct btrfs_block_group_cache *block_group;
8292         u64 last = 0;
8293
8294         while (1) {
8295                 struct inode *inode;
8296
8297                 block_group = btrfs_lookup_first_block_group(info, last);
8298                 while (block_group) {
8299                         spin_lock(&block_group->lock);
8300                         if (block_group->iref)
8301                                 break;
8302                         spin_unlock(&block_group->lock);
8303                         block_group = next_block_group(info->tree_root,
8304                                                        block_group);
8305                 }
8306                 if (!block_group) {
8307                         if (last == 0)
8308                                 break;
8309                         last = 0;
8310                         continue;
8311                 }
8312
8313                 inode = block_group->inode;
8314                 block_group->iref = 0;
8315                 block_group->inode = NULL;
8316                 spin_unlock(&block_group->lock);
8317                 iput(inode);
8318                 last = block_group->key.objectid + block_group->key.offset;
8319                 btrfs_put_block_group(block_group);
8320         }
8321 }
8322
8323 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8324 {
8325         struct btrfs_block_group_cache *block_group;
8326         struct btrfs_space_info *space_info;
8327         struct btrfs_caching_control *caching_ctl;
8328         struct rb_node *n;
8329
8330         down_write(&info->extent_commit_sem);
8331         while (!list_empty(&info->caching_block_groups)) {
8332                 caching_ctl = list_entry(info->caching_block_groups.next,
8333                                          struct btrfs_caching_control, list);
8334                 list_del(&caching_ctl->list);
8335                 put_caching_control(caching_ctl);
8336         }
8337         up_write(&info->extent_commit_sem);
8338
8339         spin_lock(&info->block_group_cache_lock);
8340         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8341                 block_group = rb_entry(n, struct btrfs_block_group_cache,
8342                                        cache_node);
8343                 rb_erase(&block_group->cache_node,
8344                          &info->block_group_cache_tree);
8345                 spin_unlock(&info->block_group_cache_lock);
8346
8347                 down_write(&block_group->space_info->groups_sem);
8348                 list_del(&block_group->list);
8349                 up_write(&block_group->space_info->groups_sem);
8350
8351                 if (block_group->cached == BTRFS_CACHE_STARTED)
8352                         wait_block_group_cache_done(block_group);
8353
8354                 /*
8355                  * We haven't cached this block group, which means we could
8356                  * possibly have excluded extents on this block group.
8357                  */
8358                 if (block_group->cached == BTRFS_CACHE_NO)
8359                         free_excluded_extents(info->extent_root, block_group);
8360
8361                 btrfs_remove_free_space_cache(block_group);
8362                 btrfs_put_block_group(block_group);
8363
8364                 spin_lock(&info->block_group_cache_lock);
8365         }
8366         spin_unlock(&info->block_group_cache_lock);
8367
8368         /* now that all the block groups are freed, go through and
8369          * free all the space_info structs.  This is only called during
8370          * the final stages of unmount, and so we know nobody is
8371          * using them.  We call synchronize_rcu() once before we start,
8372          * just to be on the safe side.
8373          */
8374         synchronize_rcu();
8375
8376         release_global_block_rsv(info);
8377
8378         while(!list_empty(&info->space_info)) {
8379                 space_info = list_entry(info->space_info.next,
8380                                         struct btrfs_space_info,
8381                                         list);
8382                 if (space_info->bytes_pinned > 0 ||
8383                     space_info->bytes_reserved > 0) {
8384                         WARN_ON(1);
8385                         dump_space_info(space_info, 0, 0);
8386                 }
8387                 list_del(&space_info->list);
8388                 kfree(space_info);
8389         }
8390         return 0;
8391 }
8392
8393 static void __link_block_group(struct btrfs_space_info *space_info,
8394                                struct btrfs_block_group_cache *cache)
8395 {
8396         int index = get_block_group_index(cache);
8397
8398         down_write(&space_info->groups_sem);
8399         list_add_tail(&cache->list, &space_info->block_groups[index]);
8400         up_write(&space_info->groups_sem);
8401 }
8402
8403 int btrfs_read_block_groups(struct btrfs_root *root)
8404 {
8405         struct btrfs_path *path;
8406         int ret;
8407         struct btrfs_block_group_cache *cache;
8408         struct btrfs_fs_info *info = root->fs_info;
8409         struct btrfs_space_info *space_info;
8410         struct btrfs_key key;
8411         struct btrfs_key found_key;
8412         struct extent_buffer *leaf;
8413         int need_clear = 0;
8414         u64 cache_gen;
8415
8416         root = info->extent_root;
8417         key.objectid = 0;
8418         key.offset = 0;
8419         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
8420         path = btrfs_alloc_path();
8421         if (!path)
8422                 return -ENOMEM;
8423
8424         cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy);
8425         if (cache_gen != 0 &&
8426             btrfs_super_generation(&root->fs_info->super_copy) != cache_gen)
8427                 need_clear = 1;
8428         if (btrfs_test_opt(root, CLEAR_CACHE))
8429                 need_clear = 1;
8430         if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen)
8431                 printk(KERN_INFO "btrfs: disk space caching is enabled\n");
8432
8433         while (1) {
8434                 ret = find_first_block_group(root, path, &key);
8435                 if (ret > 0)
8436                         break;
8437                 if (ret != 0)
8438                         goto error;
8439                 leaf = path->nodes[0];
8440                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8441                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8442                 if (!cache) {
8443                         ret = -ENOMEM;
8444                         goto error;
8445                 }
8446
8447                 atomic_set(&cache->count, 1);
8448                 spin_lock_init(&cache->lock);
8449                 spin_lock_init(&cache->tree_lock);
8450                 cache->fs_info = info;
8451                 INIT_LIST_HEAD(&cache->list);
8452                 INIT_LIST_HEAD(&cache->cluster_list);
8453
8454                 if (need_clear)
8455                         cache->disk_cache_state = BTRFS_DC_CLEAR;
8456
8457                 /*
8458                  * we only want to have 32k of ram per block group for keeping
8459                  * track of free space, and if we pass 1/2 of that we want to
8460                  * start converting things over to using bitmaps
8461                  */
8462                 cache->extents_thresh = ((1024 * 32) / 2) /
8463                         sizeof(struct btrfs_free_space);
8464
8465                 read_extent_buffer(leaf, &cache->item,
8466                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
8467                                    sizeof(cache->item));
8468                 memcpy(&cache->key, &found_key, sizeof(found_key));
8469
8470                 key.objectid = found_key.objectid + found_key.offset;
8471                 btrfs_release_path(root, path);
8472                 cache->flags = btrfs_block_group_flags(&cache->item);
8473                 cache->sectorsize = root->sectorsize;
8474
8475                 /*
8476                  * We need to exclude the super stripes now so that the space
8477                  * info has super bytes accounted for, otherwise we'll think
8478                  * we have more space than we actually do.
8479                  */
8480                 exclude_super_stripes(root, cache);
8481
8482                 /*
8483                  * check for two cases, either we are full, and therefore
8484                  * don't need to bother with the caching work since we won't
8485                  * find any space, or we are empty, and we can just add all
8486                  * the space in and be done with it.  This saves us _alot_ of
8487                  * time, particularly in the full case.
8488                  */
8489                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
8490                         cache->last_byte_to_unpin = (u64)-1;
8491                         cache->cached = BTRFS_CACHE_FINISHED;
8492                         free_excluded_extents(root, cache);
8493                 } else if (btrfs_block_group_used(&cache->item) == 0) {
8494                         cache->last_byte_to_unpin = (u64)-1;
8495                         cache->cached = BTRFS_CACHE_FINISHED;
8496                         add_new_free_space(cache, root->fs_info,
8497                                            found_key.objectid,
8498                                            found_key.objectid +
8499                                            found_key.offset);
8500                         free_excluded_extents(root, cache);
8501                 }
8502
8503                 ret = update_space_info(info, cache->flags, found_key.offset,
8504                                         btrfs_block_group_used(&cache->item),
8505                                         &space_info);
8506                 BUG_ON(ret);
8507                 cache->space_info = space_info;
8508                 spin_lock(&cache->space_info->lock);
8509                 cache->space_info->bytes_readonly += cache->bytes_super;
8510                 spin_unlock(&cache->space_info->lock);
8511
8512                 __link_block_group(space_info, cache);
8513
8514                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
8515                 BUG_ON(ret);
8516
8517                 set_avail_alloc_bits(root->fs_info, cache->flags);
8518                 if (btrfs_chunk_readonly(root, cache->key.objectid))
8519                         set_block_group_ro(cache);
8520         }
8521
8522         list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
8523                 if (!(get_alloc_profile(root, space_info->flags) &
8524                       (BTRFS_BLOCK_GROUP_RAID10 |
8525                        BTRFS_BLOCK_GROUP_RAID1 |
8526                        BTRFS_BLOCK_GROUP_DUP)))
8527                         continue;
8528                 /*
8529                  * avoid allocating from un-mirrored block group if there are
8530                  * mirrored block groups.
8531                  */
8532                 list_for_each_entry(cache, &space_info->block_groups[3], list)
8533                         set_block_group_ro(cache);
8534                 list_for_each_entry(cache, &space_info->block_groups[4], list)
8535                         set_block_group_ro(cache);
8536         }
8537
8538         init_global_block_rsv(info);
8539         ret = 0;
8540 error:
8541         btrfs_free_path(path);
8542         return ret;
8543 }
8544
8545 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
8546                            struct btrfs_root *root, u64 bytes_used,
8547                            u64 type, u64 chunk_objectid, u64 chunk_offset,
8548                            u64 size)
8549 {
8550         int ret;
8551         struct btrfs_root *extent_root;
8552         struct btrfs_block_group_cache *cache;
8553
8554         extent_root = root->fs_info->extent_root;
8555
8556         root->fs_info->last_trans_log_full_commit = trans->transid;
8557
8558         cache = kzalloc(sizeof(*cache), GFP_NOFS);
8559         if (!cache)
8560                 return -ENOMEM;
8561
8562         cache->key.objectid = chunk_offset;
8563         cache->key.offset = size;
8564         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8565         cache->sectorsize = root->sectorsize;
8566         cache->fs_info = root->fs_info;
8567
8568         /*
8569          * we only want to have 32k of ram per block group for keeping track
8570          * of free space, and if we pass 1/2 of that we want to start
8571          * converting things over to using bitmaps
8572          */
8573         cache->extents_thresh = ((1024 * 32) / 2) /
8574                 sizeof(struct btrfs_free_space);
8575         atomic_set(&cache->count, 1);
8576         spin_lock_init(&cache->lock);
8577         spin_lock_init(&cache->tree_lock);
8578         INIT_LIST_HEAD(&cache->list);
8579         INIT_LIST_HEAD(&cache->cluster_list);
8580
8581         btrfs_set_block_group_used(&cache->item, bytes_used);
8582         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
8583         cache->flags = type;
8584         btrfs_set_block_group_flags(&cache->item, type);
8585
8586         cache->last_byte_to_unpin = (u64)-1;
8587         cache->cached = BTRFS_CACHE_FINISHED;
8588         exclude_super_stripes(root, cache);
8589
8590         add_new_free_space(cache, root->fs_info, chunk_offset,
8591                            chunk_offset + size);
8592
8593         free_excluded_extents(root, cache);
8594
8595         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
8596                                 &cache->space_info);
8597         BUG_ON(ret);
8598
8599         spin_lock(&cache->space_info->lock);
8600         cache->space_info->bytes_readonly += cache->bytes_super;
8601         spin_unlock(&cache->space_info->lock);
8602
8603         __link_block_group(cache->space_info, cache);
8604
8605         ret = btrfs_add_block_group_cache(root->fs_info, cache);
8606         BUG_ON(ret);
8607
8608         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
8609                                 sizeof(cache->item));
8610         BUG_ON(ret);
8611
8612         set_avail_alloc_bits(extent_root->fs_info, type);
8613
8614         return 0;
8615 }
8616
8617 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
8618                              struct btrfs_root *root, u64 group_start)
8619 {
8620         struct btrfs_path *path;
8621         struct btrfs_block_group_cache *block_group;
8622         struct btrfs_free_cluster *cluster;
8623         struct btrfs_root *tree_root = root->fs_info->tree_root;
8624         struct btrfs_key key;
8625         struct inode *inode;
8626         int ret;
8627         int factor;
8628
8629         root = root->fs_info->extent_root;
8630
8631         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
8632         BUG_ON(!block_group);
8633         BUG_ON(!block_group->ro);
8634
8635         memcpy(&key, &block_group->key, sizeof(key));
8636         if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
8637                                   BTRFS_BLOCK_GROUP_RAID1 |
8638                                   BTRFS_BLOCK_GROUP_RAID10))
8639                 factor = 2;
8640         else
8641                 factor = 1;
8642
8643         /* make sure this block group isn't part of an allocation cluster */
8644         cluster = &root->fs_info->data_alloc_cluster;
8645         spin_lock(&cluster->refill_lock);
8646         btrfs_return_cluster_to_free_space(block_group, cluster);
8647         spin_unlock(&cluster->refill_lock);
8648
8649         /*
8650          * make sure this block group isn't part of a metadata
8651          * allocation cluster
8652          */
8653         cluster = &root->fs_info->meta_alloc_cluster;
8654         spin_lock(&cluster->refill_lock);
8655         btrfs_return_cluster_to_free_space(block_group, cluster);
8656         spin_unlock(&cluster->refill_lock);
8657
8658         path = btrfs_alloc_path();
8659         BUG_ON(!path);
8660
8661         inode = lookup_free_space_inode(root, block_group, path);
8662         if (!IS_ERR(inode)) {
8663                 btrfs_orphan_add(trans, inode);
8664                 clear_nlink(inode);
8665                 /* One for the block groups ref */
8666                 spin_lock(&block_group->lock);
8667                 if (block_group->iref) {
8668                         block_group->iref = 0;
8669                         block_group->inode = NULL;
8670                         spin_unlock(&block_group->lock);
8671                         iput(inode);
8672                 } else {
8673                         spin_unlock(&block_group->lock);
8674                 }
8675                 /* One for our lookup ref */
8676                 iput(inode);
8677         }
8678
8679         key.objectid = BTRFS_FREE_SPACE_OBJECTID;
8680         key.offset = block_group->key.objectid;
8681         key.type = 0;
8682
8683         ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
8684         if (ret < 0)
8685                 goto out;
8686         if (ret > 0)
8687                 btrfs_release_path(tree_root, path);
8688         if (ret == 0) {
8689                 ret = btrfs_del_item(trans, tree_root, path);
8690                 if (ret)
8691                         goto out;
8692                 btrfs_release_path(tree_root, path);
8693         }
8694
8695         spin_lock(&root->fs_info->block_group_cache_lock);
8696         rb_erase(&block_group->cache_node,
8697                  &root->fs_info->block_group_cache_tree);
8698         spin_unlock(&root->fs_info->block_group_cache_lock);
8699
8700         down_write(&block_group->space_info->groups_sem);
8701         /*
8702          * we must use list_del_init so people can check to see if they
8703          * are still on the list after taking the semaphore
8704          */
8705         list_del_init(&block_group->list);
8706         up_write(&block_group->space_info->groups_sem);
8707
8708         if (block_group->cached == BTRFS_CACHE_STARTED)
8709                 wait_block_group_cache_done(block_group);
8710
8711         btrfs_remove_free_space_cache(block_group);
8712
8713         spin_lock(&block_group->space_info->lock);
8714         block_group->space_info->total_bytes -= block_group->key.offset;
8715         block_group->space_info->bytes_readonly -= block_group->key.offset;
8716         block_group->space_info->disk_total -= block_group->key.offset * factor;
8717         spin_unlock(&block_group->space_info->lock);
8718
8719         memcpy(&key, &block_group->key, sizeof(key));
8720
8721         btrfs_clear_space_info_full(root->fs_info);
8722
8723         btrfs_put_block_group(block_group);
8724         btrfs_put_block_group(block_group);
8725
8726         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8727         if (ret > 0)
8728                 ret = -EIO;
8729         if (ret < 0)
8730                 goto out;
8731
8732         ret = btrfs_del_item(trans, root, path);
8733 out:
8734         btrfs_free_path(path);
8735         return ret;
8736 }
8737
8738 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
8739 {
8740         return unpin_extent_range(root, start, end);
8741 }
8742
8743 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
8744                                u64 num_bytes)
8745 {
8746         return btrfs_discard_extent(root, bytenr, num_bytes);
8747 }