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