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 static void btrfs_issue_discard(struct block_device *bdev,
1572                                 u64 start, u64 len)
1573 {
1574         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
1575 }
1576
1577 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1578                                 u64 num_bytes)
1579 {
1580         int ret;
1581         u64 map_length = num_bytes;
1582         struct btrfs_multi_bio *multi = NULL;
1583
1584         if (!btrfs_test_opt(root, DISCARD))
1585                 return 0;
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 }
1607
1608 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1609                          struct btrfs_root *root,
1610                          u64 bytenr, u64 num_bytes, u64 parent,
1611                          u64 root_objectid, u64 owner, u64 offset)
1612 {
1613         int ret;
1614         BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1615                root_objectid == BTRFS_TREE_LOG_OBJECTID);
1616
1617         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1618                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
1619                                         parent, root_objectid, (int)owner,
1620                                         BTRFS_ADD_DELAYED_REF, NULL);
1621         } else {
1622                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
1623                                         parent, root_objectid, owner, offset,
1624                                         BTRFS_ADD_DELAYED_REF, NULL);
1625         }
1626         return ret;
1627 }
1628
1629 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1630                                   struct btrfs_root *root,
1631                                   u64 bytenr, u64 num_bytes,
1632                                   u64 parent, u64 root_objectid,
1633                                   u64 owner, u64 offset, int refs_to_add,
1634                                   struct btrfs_delayed_extent_op *extent_op)
1635 {
1636         struct btrfs_path *path;
1637         struct extent_buffer *leaf;
1638         struct btrfs_extent_item *item;
1639         u64 refs;
1640         int ret;
1641         int err = 0;
1642
1643         path = btrfs_alloc_path();
1644         if (!path)
1645                 return -ENOMEM;
1646
1647         path->reada = 1;
1648         path->leave_spinning = 1;
1649         /* this will setup the path even if it fails to insert the back ref */
1650         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1651                                            path, bytenr, num_bytes, parent,
1652                                            root_objectid, owner, offset,
1653                                            refs_to_add, extent_op);
1654         if (ret == 0)
1655                 goto out;
1656
1657         if (ret != -EAGAIN) {
1658                 err = ret;
1659                 goto out;
1660         }
1661
1662         leaf = path->nodes[0];
1663         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1664         refs = btrfs_extent_refs(leaf, item);
1665         btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1666         if (extent_op)
1667                 __run_delayed_extent_op(extent_op, leaf, item);
1668
1669         btrfs_mark_buffer_dirty(leaf);
1670         btrfs_release_path(root->fs_info->extent_root, path);
1671
1672         path->reada = 1;
1673         path->leave_spinning = 1;
1674
1675         /* now insert the actual backref */
1676         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1677                                     path, bytenr, parent, root_objectid,
1678                                     owner, offset, refs_to_add);
1679         BUG_ON(ret);
1680 out:
1681         btrfs_free_path(path);
1682         return err;
1683 }
1684
1685 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
1686                                 struct btrfs_root *root,
1687                                 struct btrfs_delayed_ref_node *node,
1688                                 struct btrfs_delayed_extent_op *extent_op,
1689                                 int insert_reserved)
1690 {
1691         int ret = 0;
1692         struct btrfs_delayed_data_ref *ref;
1693         struct btrfs_key ins;
1694         u64 parent = 0;
1695         u64 ref_root = 0;
1696         u64 flags = 0;
1697
1698         ins.objectid = node->bytenr;
1699         ins.offset = node->num_bytes;
1700         ins.type = BTRFS_EXTENT_ITEM_KEY;
1701
1702         ref = btrfs_delayed_node_to_data_ref(node);
1703         if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1704                 parent = ref->parent;
1705         else
1706                 ref_root = ref->root;
1707
1708         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1709                 if (extent_op) {
1710                         BUG_ON(extent_op->update_key);
1711                         flags |= extent_op->flags_to_set;
1712                 }
1713                 ret = alloc_reserved_file_extent(trans, root,
1714                                                  parent, ref_root, flags,
1715                                                  ref->objectid, ref->offset,
1716                                                  &ins, node->ref_mod);
1717         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1718                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1719                                              node->num_bytes, parent,
1720                                              ref_root, ref->objectid,
1721                                              ref->offset, node->ref_mod,
1722                                              extent_op);
1723         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1724                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1725                                           node->num_bytes, parent,
1726                                           ref_root, ref->objectid,
1727                                           ref->offset, node->ref_mod,
1728                                           extent_op);
1729         } else {
1730                 BUG();
1731         }
1732         return ret;
1733 }
1734
1735 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1736                                     struct extent_buffer *leaf,
1737                                     struct btrfs_extent_item *ei)
1738 {
1739         u64 flags = btrfs_extent_flags(leaf, ei);
1740         if (extent_op->update_flags) {
1741                 flags |= extent_op->flags_to_set;
1742                 btrfs_set_extent_flags(leaf, ei, flags);
1743         }
1744
1745         if (extent_op->update_key) {
1746                 struct btrfs_tree_block_info *bi;
1747                 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1748                 bi = (struct btrfs_tree_block_info *)(ei + 1);
1749                 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1750         }
1751 }
1752
1753 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
1754                                  struct btrfs_root *root,
1755                                  struct btrfs_delayed_ref_node *node,
1756                                  struct btrfs_delayed_extent_op *extent_op)
1757 {
1758         struct btrfs_key key;
1759         struct btrfs_path *path;
1760         struct btrfs_extent_item *ei;
1761         struct extent_buffer *leaf;
1762         u32 item_size;
1763         int ret;
1764         int err = 0;
1765
1766         path = btrfs_alloc_path();
1767         if (!path)
1768                 return -ENOMEM;
1769
1770         key.objectid = node->bytenr;
1771         key.type = BTRFS_EXTENT_ITEM_KEY;
1772         key.offset = node->num_bytes;
1773
1774         path->reada = 1;
1775         path->leave_spinning = 1;
1776         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1777                                 path, 0, 1);
1778         if (ret < 0) {
1779                 err = ret;
1780                 goto out;
1781         }
1782         if (ret > 0) {
1783                 err = -EIO;
1784                 goto out;
1785         }
1786
1787         leaf = path->nodes[0];
1788         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1789 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1790         if (item_size < sizeof(*ei)) {
1791                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1792                                              path, (u64)-1, 0);
1793                 if (ret < 0) {
1794                         err = ret;
1795                         goto out;
1796                 }
1797                 leaf = path->nodes[0];
1798                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1799         }
1800 #endif
1801         BUG_ON(item_size < sizeof(*ei));
1802         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1803         __run_delayed_extent_op(extent_op, leaf, ei);
1804
1805         btrfs_mark_buffer_dirty(leaf);
1806 out:
1807         btrfs_free_path(path);
1808         return err;
1809 }
1810
1811 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
1812                                 struct btrfs_root *root,
1813                                 struct btrfs_delayed_ref_node *node,
1814                                 struct btrfs_delayed_extent_op *extent_op,
1815                                 int insert_reserved)
1816 {
1817         int ret = 0;
1818         struct btrfs_delayed_tree_ref *ref;
1819         struct btrfs_key ins;
1820         u64 parent = 0;
1821         u64 ref_root = 0;
1822
1823         ins.objectid = node->bytenr;
1824         ins.offset = node->num_bytes;
1825         ins.type = BTRFS_EXTENT_ITEM_KEY;
1826
1827         ref = btrfs_delayed_node_to_tree_ref(node);
1828         if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1829                 parent = ref->parent;
1830         else
1831                 ref_root = ref->root;
1832
1833         BUG_ON(node->ref_mod != 1);
1834         if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
1835                 BUG_ON(!extent_op || !extent_op->update_flags ||
1836                        !extent_op->update_key);
1837                 ret = alloc_reserved_tree_block(trans, root,
1838                                                 parent, ref_root,
1839                                                 extent_op->flags_to_set,
1840                                                 &extent_op->key,
1841                                                 ref->level, &ins);
1842         } else if (node->action == BTRFS_ADD_DELAYED_REF) {
1843                 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
1844                                              node->num_bytes, parent, ref_root,
1845                                              ref->level, 0, 1, extent_op);
1846         } else if (node->action == BTRFS_DROP_DELAYED_REF) {
1847                 ret = __btrfs_free_extent(trans, root, node->bytenr,
1848                                           node->num_bytes, parent, ref_root,
1849                                           ref->level, 0, 1, extent_op);
1850         } else {
1851                 BUG();
1852         }
1853         return ret;
1854 }
1855
1856
1857 /* helper function to actually process a single delayed ref entry */
1858 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
1859                                struct btrfs_root *root,
1860                                struct btrfs_delayed_ref_node *node,
1861                                struct btrfs_delayed_extent_op *extent_op,
1862                                int insert_reserved)
1863 {
1864         int ret;
1865         if (btrfs_delayed_ref_is_head(node)) {
1866                 struct btrfs_delayed_ref_head *head;
1867                 /*
1868                  * we've hit the end of the chain and we were supposed
1869                  * to insert this extent into the tree.  But, it got
1870                  * deleted before we ever needed to insert it, so all
1871                  * we have to do is clean up the accounting
1872                  */
1873                 BUG_ON(extent_op);
1874                 head = btrfs_delayed_node_to_head(node);
1875                 if (insert_reserved) {
1876                         int mark_free = 0;
1877                         struct extent_buffer *must_clean = NULL;
1878
1879                         ret = pin_down_bytes(trans, root, NULL,
1880                                              node->bytenr, node->num_bytes,
1881                                              head->is_data, 1, &must_clean);
1882                         if (ret > 0)
1883                                 mark_free = 1;
1884
1885                         if (must_clean) {
1886                                 clean_tree_block(NULL, root, must_clean);
1887                                 btrfs_tree_unlock(must_clean);
1888                                 free_extent_buffer(must_clean);
1889                         }
1890                         if (head->is_data) {
1891                                 ret = btrfs_del_csums(trans, root,
1892                                                       node->bytenr,
1893                                                       node->num_bytes);
1894                                 BUG_ON(ret);
1895                         }
1896                         if (mark_free) {
1897                                 ret = btrfs_free_reserved_extent(root,
1898                                                         node->bytenr,
1899                                                         node->num_bytes);
1900                                 BUG_ON(ret);
1901                         }
1902                 }
1903                 mutex_unlock(&head->mutex);
1904                 return 0;
1905         }
1906
1907         if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1908             node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1909                 ret = run_delayed_tree_ref(trans, root, node, extent_op,
1910                                            insert_reserved);
1911         else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1912                  node->type == BTRFS_SHARED_DATA_REF_KEY)
1913                 ret = run_delayed_data_ref(trans, root, node, extent_op,
1914                                            insert_reserved);
1915         else
1916                 BUG();
1917         return ret;
1918 }
1919
1920 static noinline struct btrfs_delayed_ref_node *
1921 select_delayed_ref(struct btrfs_delayed_ref_head *head)
1922 {
1923         struct rb_node *node;
1924         struct btrfs_delayed_ref_node *ref;
1925         int action = BTRFS_ADD_DELAYED_REF;
1926 again:
1927         /*
1928          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
1929          * this prevents ref count from going down to zero when
1930          * there still are pending delayed ref.
1931          */
1932         node = rb_prev(&head->node.rb_node);
1933         while (1) {
1934                 if (!node)
1935                         break;
1936                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1937                                 rb_node);
1938                 if (ref->bytenr != head->node.bytenr)
1939                         break;
1940                 if (ref->action == action)
1941                         return ref;
1942                 node = rb_prev(node);
1943         }
1944         if (action == BTRFS_ADD_DELAYED_REF) {
1945                 action = BTRFS_DROP_DELAYED_REF;
1946                 goto again;
1947         }
1948         return NULL;
1949 }
1950
1951 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
1952                                        struct btrfs_root *root,
1953                                        struct list_head *cluster)
1954 {
1955         struct btrfs_delayed_ref_root *delayed_refs;
1956         struct btrfs_delayed_ref_node *ref;
1957         struct btrfs_delayed_ref_head *locked_ref = NULL;
1958         struct btrfs_delayed_extent_op *extent_op;
1959         int ret;
1960         int count = 0;
1961         int must_insert_reserved = 0;
1962
1963         delayed_refs = &trans->transaction->delayed_refs;
1964         while (1) {
1965                 if (!locked_ref) {
1966                         /* pick a new head ref from the cluster list */
1967                         if (list_empty(cluster))
1968                                 break;
1969
1970                         locked_ref = list_entry(cluster->next,
1971                                      struct btrfs_delayed_ref_head, cluster);
1972
1973                         /* grab the lock that says we are going to process
1974                          * all the refs for this head */
1975                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
1976
1977                         /*
1978                          * we may have dropped the spin lock to get the head
1979                          * mutex lock, and that might have given someone else
1980                          * time to free the head.  If that's true, it has been
1981                          * removed from our list and we can move on.
1982                          */
1983                         if (ret == -EAGAIN) {
1984                                 locked_ref = NULL;
1985                                 count++;
1986                                 continue;
1987                         }
1988                 }
1989
1990                 /*
1991                  * record the must insert reserved flag before we
1992                  * drop the spin lock.
1993                  */
1994                 must_insert_reserved = locked_ref->must_insert_reserved;
1995                 locked_ref->must_insert_reserved = 0;
1996
1997                 extent_op = locked_ref->extent_op;
1998                 locked_ref->extent_op = NULL;
1999
2000                 /*
2001                  * locked_ref is the head node, so we have to go one
2002                  * node back for any delayed ref updates
2003                  */
2004                 ref = select_delayed_ref(locked_ref);
2005                 if (!ref) {
2006                         /* All delayed refs have been processed, Go ahead
2007                          * and send the head node to run_one_delayed_ref,
2008                          * so that any accounting fixes can happen
2009                          */
2010                         ref = &locked_ref->node;
2011
2012                         if (extent_op && must_insert_reserved) {
2013                                 kfree(extent_op);
2014                                 extent_op = NULL;
2015                         }
2016
2017                         if (extent_op) {
2018                                 spin_unlock(&delayed_refs->lock);
2019
2020                                 ret = run_delayed_extent_op(trans, root,
2021                                                             ref, extent_op);
2022                                 BUG_ON(ret);
2023                                 kfree(extent_op);
2024
2025                                 cond_resched();
2026                                 spin_lock(&delayed_refs->lock);
2027                                 continue;
2028                         }
2029
2030                         list_del_init(&locked_ref->cluster);
2031                         locked_ref = NULL;
2032                 }
2033
2034                 ref->in_tree = 0;
2035                 rb_erase(&ref->rb_node, &delayed_refs->root);
2036                 delayed_refs->num_entries--;
2037
2038                 spin_unlock(&delayed_refs->lock);
2039
2040                 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2041                                           must_insert_reserved);
2042                 BUG_ON(ret);
2043
2044                 btrfs_put_delayed_ref(ref);
2045                 kfree(extent_op);
2046                 count++;
2047
2048                 cond_resched();
2049                 spin_lock(&delayed_refs->lock);
2050         }
2051         return count;
2052 }
2053
2054 /*
2055  * this starts processing the delayed reference count updates and
2056  * extent insertions we have queued up so far.  count can be
2057  * 0, which means to process everything in the tree at the start
2058  * of the run (but not newly added entries), or it can be some target
2059  * number you'd like to process.
2060  */
2061 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2062                            struct btrfs_root *root, unsigned long count)
2063 {
2064         struct rb_node *node;
2065         struct btrfs_delayed_ref_root *delayed_refs;
2066         struct btrfs_delayed_ref_node *ref;
2067         struct list_head cluster;
2068         int ret;
2069         int run_all = count == (unsigned long)-1;
2070         int run_most = 0;
2071
2072         if (root == root->fs_info->extent_root)
2073                 root = root->fs_info->tree_root;
2074
2075         delayed_refs = &trans->transaction->delayed_refs;
2076         INIT_LIST_HEAD(&cluster);
2077 again:
2078         spin_lock(&delayed_refs->lock);
2079         if (count == 0) {
2080                 count = delayed_refs->num_entries * 2;
2081                 run_most = 1;
2082         }
2083         while (1) {
2084                 if (!(run_all || run_most) &&
2085                     delayed_refs->num_heads_ready < 64)
2086                         break;
2087
2088                 /*
2089                  * go find something we can process in the rbtree.  We start at
2090                  * the beginning of the tree, and then build a cluster
2091                  * of refs to process starting at the first one we are able to
2092                  * lock
2093                  */
2094                 ret = btrfs_find_ref_cluster(trans, &cluster,
2095                                              delayed_refs->run_delayed_start);
2096                 if (ret)
2097                         break;
2098
2099                 ret = run_clustered_refs(trans, root, &cluster);
2100                 BUG_ON(ret < 0);
2101
2102                 count -= min_t(unsigned long, ret, count);
2103
2104                 if (count == 0)
2105                         break;
2106         }
2107
2108         if (run_all) {
2109                 node = rb_first(&delayed_refs->root);
2110                 if (!node)
2111                         goto out;
2112                 count = (unsigned long)-1;
2113
2114                 while (node) {
2115                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
2116                                        rb_node);
2117                         if (btrfs_delayed_ref_is_head(ref)) {
2118                                 struct btrfs_delayed_ref_head *head;
2119
2120                                 head = btrfs_delayed_node_to_head(ref);
2121                                 atomic_inc(&ref->refs);
2122
2123                                 spin_unlock(&delayed_refs->lock);
2124                                 mutex_lock(&head->mutex);
2125                                 mutex_unlock(&head->mutex);
2126
2127                                 btrfs_put_delayed_ref(ref);
2128                                 cond_resched();
2129                                 goto again;
2130                         }
2131                         node = rb_next(node);
2132                 }
2133                 spin_unlock(&delayed_refs->lock);
2134                 schedule_timeout(1);
2135                 goto again;
2136         }
2137 out:
2138         spin_unlock(&delayed_refs->lock);
2139         return 0;
2140 }
2141
2142 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2143                                 struct btrfs_root *root,
2144                                 u64 bytenr, u64 num_bytes, u64 flags,
2145                                 int is_data)
2146 {
2147         struct btrfs_delayed_extent_op *extent_op;
2148         int ret;
2149
2150         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2151         if (!extent_op)
2152                 return -ENOMEM;
2153
2154         extent_op->flags_to_set = flags;
2155         extent_op->update_flags = 1;
2156         extent_op->update_key = 0;
2157         extent_op->is_data = is_data ? 1 : 0;
2158
2159         ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
2160         if (ret)
2161                 kfree(extent_op);
2162         return ret;
2163 }
2164
2165 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2166                                       struct btrfs_root *root,
2167                                       struct btrfs_path *path,
2168                                       u64 objectid, u64 offset, u64 bytenr)
2169 {
2170         struct btrfs_delayed_ref_head *head;
2171         struct btrfs_delayed_ref_node *ref;
2172         struct btrfs_delayed_data_ref *data_ref;
2173         struct btrfs_delayed_ref_root *delayed_refs;
2174         struct rb_node *node;
2175         int ret = 0;
2176
2177         ret = -ENOENT;
2178         delayed_refs = &trans->transaction->delayed_refs;
2179         spin_lock(&delayed_refs->lock);
2180         head = btrfs_find_delayed_ref_head(trans, bytenr);
2181         if (!head)
2182                 goto out;
2183
2184         if (!mutex_trylock(&head->mutex)) {
2185                 atomic_inc(&head->node.refs);
2186                 spin_unlock(&delayed_refs->lock);
2187
2188                 btrfs_release_path(root->fs_info->extent_root, path);
2189
2190                 mutex_lock(&head->mutex);
2191                 mutex_unlock(&head->mutex);
2192                 btrfs_put_delayed_ref(&head->node);
2193                 return -EAGAIN;
2194         }
2195
2196         node = rb_prev(&head->node.rb_node);
2197         if (!node)
2198                 goto out_unlock;
2199
2200         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2201
2202         if (ref->bytenr != bytenr)
2203                 goto out_unlock;
2204
2205         ret = 1;
2206         if (ref->type != BTRFS_EXTENT_DATA_REF_KEY)
2207                 goto out_unlock;
2208
2209         data_ref = btrfs_delayed_node_to_data_ref(ref);
2210
2211         node = rb_prev(node);
2212         if (node) {
2213                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2214                 if (ref->bytenr == bytenr)
2215                         goto out_unlock;
2216         }
2217
2218         if (data_ref->root != root->root_key.objectid ||
2219             data_ref->objectid != objectid || data_ref->offset != offset)
2220                 goto out_unlock;
2221
2222         ret = 0;
2223 out_unlock:
2224         mutex_unlock(&head->mutex);
2225 out:
2226         spin_unlock(&delayed_refs->lock);
2227         return ret;
2228 }
2229
2230 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2231                                         struct btrfs_root *root,
2232                                         struct btrfs_path *path,
2233                                         u64 objectid, u64 offset, u64 bytenr)
2234 {
2235         struct btrfs_root *extent_root = root->fs_info->extent_root;
2236         struct extent_buffer *leaf;
2237         struct btrfs_extent_data_ref *ref;
2238         struct btrfs_extent_inline_ref *iref;
2239         struct btrfs_extent_item *ei;
2240         struct btrfs_key key;
2241         u32 item_size;
2242         int ret;
2243
2244         key.objectid = bytenr;
2245         key.offset = (u64)-1;
2246         key.type = BTRFS_EXTENT_ITEM_KEY;
2247
2248         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2249         if (ret < 0)
2250                 goto out;
2251         BUG_ON(ret == 0);
2252
2253         ret = -ENOENT;
2254         if (path->slots[0] == 0)
2255                 goto out;
2256
2257         path->slots[0]--;
2258         leaf = path->nodes[0];
2259         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2260
2261         if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2262                 goto out;
2263
2264         ret = 1;
2265         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2266 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2267         if (item_size < sizeof(*ei)) {
2268                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2269                 goto out;
2270         }
2271 #endif
2272         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2273
2274         if (item_size != sizeof(*ei) +
2275             btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2276                 goto out;
2277
2278         if (btrfs_extent_generation(leaf, ei) <=
2279             btrfs_root_last_snapshot(&root->root_item))
2280                 goto out;
2281
2282         iref = (struct btrfs_extent_inline_ref *)(ei + 1);
2283         if (btrfs_extent_inline_ref_type(leaf, iref) !=
2284             BTRFS_EXTENT_DATA_REF_KEY)
2285                 goto out;
2286
2287         ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2288         if (btrfs_extent_refs(leaf, ei) !=
2289             btrfs_extent_data_ref_count(leaf, ref) ||
2290             btrfs_extent_data_ref_root(leaf, ref) !=
2291             root->root_key.objectid ||
2292             btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2293             btrfs_extent_data_ref_offset(leaf, ref) != offset)
2294                 goto out;
2295
2296         ret = 0;
2297 out:
2298         return ret;
2299 }
2300
2301 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
2302                           struct btrfs_root *root,
2303                           u64 objectid, u64 offset, u64 bytenr)
2304 {
2305         struct btrfs_path *path;
2306         int ret;
2307         int ret2;
2308
2309         path = btrfs_alloc_path();
2310         if (!path)
2311                 return -ENOENT;
2312
2313         do {
2314                 ret = check_committed_ref(trans, root, path, objectid,
2315                                           offset, bytenr);
2316                 if (ret && ret != -ENOENT)
2317                         goto out;
2318
2319                 ret2 = check_delayed_ref(trans, root, path, objectid,
2320                                          offset, bytenr);
2321         } while (ret2 == -EAGAIN);
2322
2323         if (ret2 && ret2 != -ENOENT) {
2324                 ret = ret2;
2325                 goto out;
2326         }
2327
2328         if (ret != -ENOENT || ret2 != -ENOENT)
2329                 ret = 0;
2330 out:
2331         btrfs_free_path(path);
2332         return ret;
2333 }
2334
2335 #if 0
2336 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2337                     struct extent_buffer *buf, u32 nr_extents)
2338 {
2339         struct btrfs_key key;
2340         struct btrfs_file_extent_item *fi;
2341         u64 root_gen;
2342         u32 nritems;
2343         int i;
2344         int level;
2345         int ret = 0;
2346         int shared = 0;
2347
2348         if (!root->ref_cows)
2349                 return 0;
2350
2351         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2352                 shared = 0;
2353                 root_gen = root->root_key.offset;
2354         } else {
2355                 shared = 1;
2356                 root_gen = trans->transid - 1;
2357         }
2358
2359         level = btrfs_header_level(buf);
2360         nritems = btrfs_header_nritems(buf);
2361
2362         if (level == 0) {
2363                 struct btrfs_leaf_ref *ref;
2364                 struct btrfs_extent_info *info;
2365
2366                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
2367                 if (!ref) {
2368                         ret = -ENOMEM;
2369                         goto out;
2370                 }
2371
2372                 ref->root_gen = root_gen;
2373                 ref->bytenr = buf->start;
2374                 ref->owner = btrfs_header_owner(buf);
2375                 ref->generation = btrfs_header_generation(buf);
2376                 ref->nritems = nr_extents;
2377                 info = ref->extents;
2378
2379                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
2380                         u64 disk_bytenr;
2381                         btrfs_item_key_to_cpu(buf, &key, i);
2382                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2383                                 continue;
2384                         fi = btrfs_item_ptr(buf, i,
2385                                             struct btrfs_file_extent_item);
2386                         if (btrfs_file_extent_type(buf, fi) ==
2387                             BTRFS_FILE_EXTENT_INLINE)
2388                                 continue;
2389                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2390                         if (disk_bytenr == 0)
2391                                 continue;
2392
2393                         info->bytenr = disk_bytenr;
2394                         info->num_bytes =
2395                                 btrfs_file_extent_disk_num_bytes(buf, fi);
2396                         info->objectid = key.objectid;
2397                         info->offset = key.offset;
2398                         info++;
2399                 }
2400
2401                 ret = btrfs_add_leaf_ref(root, ref, shared);
2402                 if (ret == -EEXIST && shared) {
2403                         struct btrfs_leaf_ref *old;
2404                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
2405                         BUG_ON(!old);
2406                         btrfs_remove_leaf_ref(root, old);
2407                         btrfs_free_leaf_ref(root, old);
2408                         ret = btrfs_add_leaf_ref(root, ref, shared);
2409                 }
2410                 WARN_ON(ret);
2411                 btrfs_free_leaf_ref(root, ref);
2412         }
2413 out:
2414         return ret;
2415 }
2416
2417 /* when a block goes through cow, we update the reference counts of
2418  * everything that block points to.  The internal pointers of the block
2419  * can be in just about any order, and it is likely to have clusters of
2420  * things that are close together and clusters of things that are not.
2421  *
2422  * To help reduce the seeks that come with updating all of these reference
2423  * counts, sort them by byte number before actual updates are done.
2424  *
2425  * struct refsort is used to match byte number to slot in the btree block.
2426  * we sort based on the byte number and then use the slot to actually
2427  * find the item.
2428  *
2429  * struct refsort is smaller than strcut btrfs_item and smaller than
2430  * struct btrfs_key_ptr.  Since we're currently limited to the page size
2431  * for a btree block, there's no way for a kmalloc of refsorts for a
2432  * single node to be bigger than a page.
2433  */
2434 struct refsort {
2435         u64 bytenr;
2436         u32 slot;
2437 };
2438
2439 /*
2440  * for passing into sort()
2441  */
2442 static int refsort_cmp(const void *a_void, const void *b_void)
2443 {
2444         const struct refsort *a = a_void;
2445         const struct refsort *b = b_void;
2446
2447         if (a->bytenr < b->bytenr)
2448                 return -1;
2449         if (a->bytenr > b->bytenr)
2450                 return 1;
2451         return 0;
2452 }
2453 #endif
2454
2455 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
2456                            struct btrfs_root *root,
2457                            struct extent_buffer *buf,
2458                            int full_backref, int inc)
2459 {
2460         u64 bytenr;
2461         u64 num_bytes;
2462         u64 parent;
2463         u64 ref_root;
2464         u32 nritems;
2465         struct btrfs_key key;
2466         struct btrfs_file_extent_item *fi;
2467         int i;
2468         int level;
2469         int ret = 0;
2470         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
2471                             u64, u64, u64, u64, u64, u64);
2472
2473         ref_root = btrfs_header_owner(buf);
2474         nritems = btrfs_header_nritems(buf);
2475         level = btrfs_header_level(buf);
2476
2477         if (!root->ref_cows && level == 0)
2478                 return 0;
2479
2480         if (inc)
2481                 process_func = btrfs_inc_extent_ref;
2482         else
2483                 process_func = btrfs_free_extent;
2484
2485         if (full_backref)
2486                 parent = buf->start;
2487         else
2488                 parent = 0;
2489
2490         for (i = 0; i < nritems; i++) {
2491                 if (level == 0) {
2492                         btrfs_item_key_to_cpu(buf, &key, i);
2493                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2494                                 continue;
2495                         fi = btrfs_item_ptr(buf, i,
2496                                             struct btrfs_file_extent_item);
2497                         if (btrfs_file_extent_type(buf, fi) ==
2498                             BTRFS_FILE_EXTENT_INLINE)
2499                                 continue;
2500                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2501                         if (bytenr == 0)
2502                                 continue;
2503
2504                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2505                         key.offset -= btrfs_file_extent_offset(buf, fi);
2506                         ret = process_func(trans, root, bytenr, num_bytes,
2507                                            parent, ref_root, key.objectid,
2508                                            key.offset);
2509                         if (ret)
2510                                 goto fail;
2511                 } else {
2512                         bytenr = btrfs_node_blockptr(buf, i);
2513                         num_bytes = btrfs_level_size(root, level - 1);
2514                         ret = process_func(trans, root, bytenr, num_bytes,
2515                                            parent, ref_root, level - 1, 0);
2516                         if (ret)
2517                                 goto fail;
2518                 }
2519         }
2520         return 0;
2521 fail:
2522         BUG();
2523         return ret;
2524 }
2525
2526 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2527                   struct extent_buffer *buf, int full_backref)
2528 {
2529         return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
2530 }
2531
2532 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2533                   struct extent_buffer *buf, int full_backref)
2534 {
2535         return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
2536 }
2537
2538 static int write_one_cache_group(struct btrfs_trans_handle *trans,
2539                                  struct btrfs_root *root,
2540                                  struct btrfs_path *path,
2541                                  struct btrfs_block_group_cache *cache)
2542 {
2543         int ret;
2544         struct btrfs_root *extent_root = root->fs_info->extent_root;
2545         unsigned long bi;
2546         struct extent_buffer *leaf;
2547
2548         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
2549         if (ret < 0)
2550                 goto fail;
2551         BUG_ON(ret);
2552
2553         leaf = path->nodes[0];
2554         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
2555         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
2556         btrfs_mark_buffer_dirty(leaf);
2557         btrfs_release_path(extent_root, path);
2558 fail:
2559         if (ret)
2560                 return ret;
2561         return 0;
2562
2563 }
2564
2565 static struct btrfs_block_group_cache *
2566 next_block_group(struct btrfs_root *root,
2567                  struct btrfs_block_group_cache *cache)
2568 {
2569         struct rb_node *node;
2570         spin_lock(&root->fs_info->block_group_cache_lock);
2571         node = rb_next(&cache->cache_node);
2572         btrfs_put_block_group(cache);
2573         if (node) {
2574                 cache = rb_entry(node, struct btrfs_block_group_cache,
2575                                  cache_node);
2576                 atomic_inc(&cache->count);
2577         } else
2578                 cache = NULL;
2579         spin_unlock(&root->fs_info->block_group_cache_lock);
2580         return cache;
2581 }
2582
2583 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
2584                                    struct btrfs_root *root)
2585 {
2586         struct btrfs_block_group_cache *cache;
2587         int err = 0;
2588         struct btrfs_path *path;
2589         u64 last = 0;
2590
2591         path = btrfs_alloc_path();
2592         if (!path)
2593                 return -ENOMEM;
2594
2595         while (1) {
2596                 if (last == 0) {
2597                         err = btrfs_run_delayed_refs(trans, root,
2598                                                      (unsigned long)-1);
2599                         BUG_ON(err);
2600                 }
2601
2602                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
2603                 while (cache) {
2604                         if (cache->dirty)
2605                                 break;
2606                         cache = next_block_group(root, cache);
2607                 }
2608                 if (!cache) {
2609                         if (last == 0)
2610                                 break;
2611                         last = 0;
2612                         continue;
2613                 }
2614
2615                 cache->dirty = 0;
2616                 last = cache->key.objectid + cache->key.offset;
2617
2618                 err = write_one_cache_group(trans, root, path, cache);
2619                 BUG_ON(err);
2620                 btrfs_put_block_group(cache);
2621         }
2622
2623         btrfs_free_path(path);
2624         return 0;
2625 }
2626
2627 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
2628 {
2629         struct btrfs_block_group_cache *block_group;
2630         int readonly = 0;
2631
2632         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
2633         if (!block_group || block_group->ro)
2634                 readonly = 1;
2635         if (block_group)
2636                 btrfs_put_block_group(block_group);
2637         return readonly;
2638 }
2639
2640 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
2641                              u64 total_bytes, u64 bytes_used,
2642                              struct btrfs_space_info **space_info)
2643 {
2644         struct btrfs_space_info *found;
2645
2646         found = __find_space_info(info, flags);
2647         if (found) {
2648                 spin_lock(&found->lock);
2649                 found->total_bytes += total_bytes;
2650                 found->bytes_used += bytes_used;
2651                 found->full = 0;
2652                 spin_unlock(&found->lock);
2653                 *space_info = found;
2654                 return 0;
2655         }
2656         found = kzalloc(sizeof(*found), GFP_NOFS);
2657         if (!found)
2658                 return -ENOMEM;
2659
2660         INIT_LIST_HEAD(&found->block_groups);
2661         init_rwsem(&found->groups_sem);
2662         spin_lock_init(&found->lock);
2663         found->flags = flags;
2664         found->total_bytes = total_bytes;
2665         found->bytes_used = bytes_used;
2666         found->bytes_pinned = 0;
2667         found->bytes_reserved = 0;
2668         found->bytes_readonly = 0;
2669         found->bytes_delalloc = 0;
2670         found->full = 0;
2671         found->force_alloc = 0;
2672         *space_info = found;
2673         list_add_rcu(&found->list, &info->space_info);
2674         atomic_set(&found->caching_threads, 0);
2675         return 0;
2676 }
2677
2678 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
2679 {
2680         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
2681                                    BTRFS_BLOCK_GROUP_RAID1 |
2682                                    BTRFS_BLOCK_GROUP_RAID10 |
2683                                    BTRFS_BLOCK_GROUP_DUP);
2684         if (extra_flags) {
2685                 if (flags & BTRFS_BLOCK_GROUP_DATA)
2686                         fs_info->avail_data_alloc_bits |= extra_flags;
2687                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
2688                         fs_info->avail_metadata_alloc_bits |= extra_flags;
2689                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
2690                         fs_info->avail_system_alloc_bits |= extra_flags;
2691         }
2692 }
2693
2694 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
2695 {
2696         spin_lock(&cache->space_info->lock);
2697         spin_lock(&cache->lock);
2698         if (!cache->ro) {
2699                 cache->space_info->bytes_readonly += cache->key.offset -
2700                                         btrfs_block_group_used(&cache->item);
2701                 cache->ro = 1;
2702         }
2703         spin_unlock(&cache->lock);
2704         spin_unlock(&cache->space_info->lock);
2705 }
2706
2707 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
2708 {
2709         u64 num_devices = root->fs_info->fs_devices->rw_devices;
2710
2711         if (num_devices == 1)
2712                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
2713         if (num_devices < 4)
2714                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
2715
2716         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
2717             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
2718                       BTRFS_BLOCK_GROUP_RAID10))) {
2719                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
2720         }
2721
2722         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
2723             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
2724                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
2725         }
2726
2727         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
2728             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
2729              (flags & BTRFS_BLOCK_GROUP_RAID10) |
2730              (flags & BTRFS_BLOCK_GROUP_DUP)))
2731                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
2732         return flags;
2733 }
2734
2735 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
2736 {
2737         struct btrfs_fs_info *info = root->fs_info;
2738         u64 alloc_profile;
2739
2740         if (data) {
2741                 alloc_profile = info->avail_data_alloc_bits &
2742                         info->data_alloc_profile;
2743                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2744         } else if (root == root->fs_info->chunk_root) {
2745                 alloc_profile = info->avail_system_alloc_bits &
2746                         info->system_alloc_profile;
2747                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2748         } else {
2749                 alloc_profile = info->avail_metadata_alloc_bits &
2750                         info->metadata_alloc_profile;
2751                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2752         }
2753
2754         return btrfs_reduce_alloc_profile(root, data);
2755 }
2756
2757 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
2758 {
2759         u64 alloc_target;
2760
2761         alloc_target = btrfs_get_alloc_profile(root, 1);
2762         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
2763                                                        alloc_target);
2764 }
2765
2766 static u64 calculate_bytes_needed(struct btrfs_root *root, int num_items)
2767 {
2768         u64 num_bytes;
2769         int level;
2770
2771         level = BTRFS_MAX_LEVEL - 2;
2772         /*
2773          * NOTE: these calculations are absolutely the worst possible case.
2774          * This assumes that _every_ item we insert will require a new leaf, and
2775          * that the tree has grown to its maximum level size.
2776          */
2777
2778         /*
2779          * for every item we insert we could insert both an extent item and a
2780          * extent ref item.  Then for ever item we insert, we will need to cow
2781          * both the original leaf, plus the leaf to the left and right of it.
2782          *
2783          * Unless we are talking about the extent root, then we just want the
2784          * number of items * 2, since we just need the extent item plus its ref.
2785          */
2786         if (root == root->fs_info->extent_root)
2787                 num_bytes = num_items * 2;
2788         else
2789                 num_bytes = (num_items + (2 * num_items)) * 3;
2790
2791         /*
2792          * num_bytes is total number of leaves we could need times the leaf
2793          * size, and then for every leaf we could end up cow'ing 2 nodes per
2794          * level, down to the leaf level.
2795          */
2796         num_bytes = (num_bytes * root->leafsize) +
2797                 (num_bytes * (level * 2)) * root->nodesize;
2798
2799         return num_bytes;
2800 }
2801
2802 /*
2803  * Unreserve metadata space for delalloc.  If we have less reserved credits than
2804  * we have extents, this function does nothing.
2805  */
2806 int btrfs_unreserve_metadata_for_delalloc(struct btrfs_root *root,
2807                                           struct inode *inode, int num_items)
2808 {
2809         struct btrfs_fs_info *info = root->fs_info;
2810         struct btrfs_space_info *meta_sinfo;
2811         u64 num_bytes;
2812         u64 alloc_target;
2813         bool bug = false;
2814
2815         /* get the space info for where the metadata will live */
2816         alloc_target = btrfs_get_alloc_profile(root, 0);
2817         meta_sinfo = __find_space_info(info, alloc_target);
2818
2819         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
2820                                            num_items);
2821
2822         spin_lock(&meta_sinfo->lock);
2823         spin_lock(&BTRFS_I(inode)->accounting_lock);
2824         if (BTRFS_I(inode)->reserved_extents <=
2825             BTRFS_I(inode)->outstanding_extents) {
2826                 spin_unlock(&BTRFS_I(inode)->accounting_lock);
2827                 spin_unlock(&meta_sinfo->lock);
2828                 return 0;
2829         }
2830         spin_unlock(&BTRFS_I(inode)->accounting_lock);
2831
2832         BTRFS_I(inode)->reserved_extents--;
2833         BUG_ON(BTRFS_I(inode)->reserved_extents < 0);
2834
2835         if (meta_sinfo->bytes_delalloc < num_bytes) {
2836                 bug = true;
2837                 meta_sinfo->bytes_delalloc = 0;
2838         } else {
2839                 meta_sinfo->bytes_delalloc -= num_bytes;
2840         }
2841         spin_unlock(&meta_sinfo->lock);
2842
2843         BUG_ON(bug);
2844
2845         return 0;
2846 }
2847
2848 static void check_force_delalloc(struct btrfs_space_info *meta_sinfo)
2849 {
2850         u64 thresh;
2851
2852         thresh = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
2853                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
2854                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
2855                 meta_sinfo->bytes_may_use;
2856
2857         thresh = meta_sinfo->total_bytes - thresh;
2858         thresh *= 80;
2859         do_div(thresh, 100);
2860         if (thresh <= meta_sinfo->bytes_delalloc)
2861                 meta_sinfo->force_delalloc = 1;
2862         else
2863                 meta_sinfo->force_delalloc = 0;
2864 }
2865
2866 struct async_flush {
2867         struct btrfs_root *root;
2868         struct btrfs_space_info *info;
2869         struct btrfs_work work;
2870 };
2871
2872 static noinline void flush_delalloc_async(struct btrfs_work *work)
2873 {
2874         struct async_flush *async;
2875         struct btrfs_root *root;
2876         struct btrfs_space_info *info;
2877
2878         async = container_of(work, struct async_flush, work);
2879         root = async->root;
2880         info = async->info;
2881
2882         btrfs_start_delalloc_inodes(root);
2883         wake_up(&info->flush_wait);
2884         btrfs_wait_ordered_extents(root, 0);
2885
2886         spin_lock(&info->lock);
2887         info->flushing = 0;
2888         spin_unlock(&info->lock);
2889         wake_up(&info->flush_wait);
2890
2891         kfree(async);
2892 }
2893
2894 static void wait_on_flush(struct btrfs_space_info *info)
2895 {
2896         DEFINE_WAIT(wait);
2897         u64 used;
2898
2899         while (1) {
2900                 prepare_to_wait(&info->flush_wait, &wait,
2901                                 TASK_UNINTERRUPTIBLE);
2902                 spin_lock(&info->lock);
2903                 if (!info->flushing) {
2904                         spin_unlock(&info->lock);
2905                         break;
2906                 }
2907
2908                 used = info->bytes_used + info->bytes_reserved +
2909                         info->bytes_pinned + info->bytes_readonly +
2910                         info->bytes_super + info->bytes_root +
2911                         info->bytes_may_use + info->bytes_delalloc;
2912                 if (used < info->total_bytes) {
2913                         spin_unlock(&info->lock);
2914                         break;
2915                 }
2916                 spin_unlock(&info->lock);
2917                 schedule();
2918         }
2919         finish_wait(&info->flush_wait, &wait);
2920 }
2921
2922 static void flush_delalloc(struct btrfs_root *root,
2923                                  struct btrfs_space_info *info)
2924 {
2925         struct async_flush *async;
2926         bool wait = false;
2927
2928         spin_lock(&info->lock);
2929
2930         if (!info->flushing) {
2931                 info->flushing = 1;
2932                 init_waitqueue_head(&info->flush_wait);
2933         } else {
2934                 wait = true;
2935         }
2936
2937         spin_unlock(&info->lock);
2938
2939         if (wait) {
2940                 wait_on_flush(info);
2941                 return;
2942         }
2943
2944         async = kzalloc(sizeof(*async), GFP_NOFS);
2945         if (!async)
2946                 goto flush;
2947
2948         async->root = root;
2949         async->info = info;
2950         async->work.func = flush_delalloc_async;
2951
2952         btrfs_queue_worker(&root->fs_info->enospc_workers,
2953                            &async->work);
2954         wait_on_flush(info);
2955         return;
2956
2957 flush:
2958         btrfs_start_delalloc_inodes(root);
2959         btrfs_wait_ordered_extents(root, 0);
2960
2961         spin_lock(&info->lock);
2962         info->flushing = 0;
2963         spin_unlock(&info->lock);
2964         wake_up(&info->flush_wait);
2965 }
2966
2967 static int maybe_allocate_chunk(struct btrfs_root *root,
2968                                  struct btrfs_space_info *info)
2969 {
2970         struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
2971         struct btrfs_trans_handle *trans;
2972         bool wait = false;
2973         int ret = 0;
2974         u64 min_metadata;
2975         u64 free_space;
2976
2977         free_space = btrfs_super_total_bytes(disk_super);
2978         /*
2979          * we allow the metadata to grow to a max of either 5gb or 5% of the
2980          * space in the volume.
2981          */
2982         min_metadata = min((u64)5 * 1024 * 1024 * 1024,
2983                              div64_u64(free_space * 5, 100));
2984         if (info->total_bytes >= min_metadata) {
2985                 spin_unlock(&info->lock);
2986                 return 0;
2987         }
2988
2989         if (info->full) {
2990                 spin_unlock(&info->lock);
2991                 return 0;
2992         }
2993
2994         if (!info->allocating_chunk) {
2995                 info->force_alloc = 1;
2996                 info->allocating_chunk = 1;
2997                 init_waitqueue_head(&info->allocate_wait);
2998         } else {
2999                 wait = true;
3000         }
3001
3002         spin_unlock(&info->lock);
3003
3004         if (wait) {
3005                 wait_event(info->allocate_wait,
3006                            !info->allocating_chunk);
3007                 return 1;
3008         }
3009
3010         trans = btrfs_start_transaction(root, 1);
3011         if (!trans) {
3012                 ret = -ENOMEM;
3013                 goto out;
3014         }
3015
3016         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3017                              4096 + 2 * 1024 * 1024,
3018                              info->flags, 0);
3019         btrfs_end_transaction(trans, root);
3020         if (ret)
3021                 goto out;
3022 out:
3023         spin_lock(&info->lock);
3024         info->allocating_chunk = 0;
3025         spin_unlock(&info->lock);
3026         wake_up(&info->allocate_wait);
3027
3028         if (ret)
3029                 return 0;
3030         return 1;
3031 }
3032
3033 /*
3034  * Reserve metadata space for delalloc.
3035  */
3036 int btrfs_reserve_metadata_for_delalloc(struct btrfs_root *root,
3037                                         struct inode *inode, int num_items)
3038 {
3039         struct btrfs_fs_info *info = root->fs_info;
3040         struct btrfs_space_info *meta_sinfo;
3041         u64 num_bytes;
3042         u64 used;
3043         u64 alloc_target;
3044         int flushed = 0;
3045         int force_delalloc;
3046
3047         /* get the space info for where the metadata will live */
3048         alloc_target = btrfs_get_alloc_profile(root, 0);
3049         meta_sinfo = __find_space_info(info, alloc_target);
3050
3051         num_bytes = calculate_bytes_needed(root->fs_info->extent_root,
3052                                            num_items);
3053 again:
3054         spin_lock(&meta_sinfo->lock);
3055
3056         force_delalloc = meta_sinfo->force_delalloc;
3057
3058         if (unlikely(!meta_sinfo->bytes_root))
3059                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3060
3061         if (!flushed)
3062                 meta_sinfo->bytes_delalloc += num_bytes;
3063
3064         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3065                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3066                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3067                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3068
3069         if (used > meta_sinfo->total_bytes) {
3070                 flushed++;
3071
3072                 if (flushed == 1) {
3073                         if (maybe_allocate_chunk(root, meta_sinfo))
3074                                 goto again;
3075                         flushed++;
3076                 } else {
3077                         spin_unlock(&meta_sinfo->lock);
3078                 }
3079
3080                 if (flushed == 2) {
3081                         filemap_flush(inode->i_mapping);
3082                         goto again;
3083                 } else if (flushed == 3) {
3084                         flush_delalloc(root, meta_sinfo);
3085                         goto again;
3086                 }
3087                 spin_lock(&meta_sinfo->lock);
3088                 meta_sinfo->bytes_delalloc -= num_bytes;
3089                 spin_unlock(&meta_sinfo->lock);
3090                 printk(KERN_ERR "enospc, has %d, reserved %d\n",
3091                        BTRFS_I(inode)->outstanding_extents,
3092                        BTRFS_I(inode)->reserved_extents);
3093                 dump_space_info(meta_sinfo, 0, 0);
3094                 return -ENOSPC;
3095         }
3096
3097         BTRFS_I(inode)->reserved_extents++;
3098         check_force_delalloc(meta_sinfo);
3099         spin_unlock(&meta_sinfo->lock);
3100
3101         if (!flushed && force_delalloc)
3102                 filemap_flush(inode->i_mapping);
3103
3104         return 0;
3105 }
3106
3107 /*
3108  * unreserve num_items number of items worth of metadata space.  This needs to
3109  * be paired with btrfs_reserve_metadata_space.
3110  *
3111  * NOTE: if you have the option, run this _AFTER_ you do a
3112  * btrfs_end_transaction, since btrfs_end_transaction will run delayed ref
3113  * oprations which will result in more used metadata, so we want to make sure we
3114  * can do that without issue.
3115  */
3116 int btrfs_unreserve_metadata_space(struct btrfs_root *root, int num_items)
3117 {
3118         struct btrfs_fs_info *info = root->fs_info;
3119         struct btrfs_space_info *meta_sinfo;
3120         u64 num_bytes;
3121         u64 alloc_target;
3122         bool bug = false;
3123
3124         /* get the space info for where the metadata will live */
3125         alloc_target = btrfs_get_alloc_profile(root, 0);
3126         meta_sinfo = __find_space_info(info, alloc_target);
3127
3128         num_bytes = calculate_bytes_needed(root, num_items);
3129
3130         spin_lock(&meta_sinfo->lock);
3131         if (meta_sinfo->bytes_may_use < num_bytes) {
3132                 bug = true;
3133                 meta_sinfo->bytes_may_use = 0;
3134         } else {
3135                 meta_sinfo->bytes_may_use -= num_bytes;
3136         }
3137         spin_unlock(&meta_sinfo->lock);
3138
3139         BUG_ON(bug);
3140
3141         return 0;
3142 }
3143
3144 /*
3145  * Reserve some metadata space for use.  We'll calculate the worste case number
3146  * of bytes that would be needed to modify num_items number of items.  If we
3147  * have space, fantastic, if not, you get -ENOSPC.  Please call
3148  * btrfs_unreserve_metadata_space when you are done for the _SAME_ number of
3149  * items you reserved, since whatever metadata you needed should have already
3150  * been allocated.
3151  *
3152  * This will commit the transaction to make more space if we don't have enough
3153  * metadata space.  THe only time we don't do this is if we're reserving space
3154  * inside of a transaction, then we will just return -ENOSPC and it is the
3155  * callers responsibility to handle it properly.
3156  */
3157 int btrfs_reserve_metadata_space(struct btrfs_root *root, int num_items)
3158 {
3159         struct btrfs_fs_info *info = root->fs_info;
3160         struct btrfs_space_info *meta_sinfo;
3161         u64 num_bytes;
3162         u64 used;
3163         u64 alloc_target;
3164         int retries = 0;
3165
3166         /* get the space info for where the metadata will live */
3167         alloc_target = btrfs_get_alloc_profile(root, 0);
3168         meta_sinfo = __find_space_info(info, alloc_target);
3169
3170         num_bytes = calculate_bytes_needed(root, num_items);
3171 again:
3172         spin_lock(&meta_sinfo->lock);
3173
3174         if (unlikely(!meta_sinfo->bytes_root))
3175                 meta_sinfo->bytes_root = calculate_bytes_needed(root, 6);
3176
3177         if (!retries)
3178                 meta_sinfo->bytes_may_use += num_bytes;
3179
3180         used = meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
3181                 meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly +
3182                 meta_sinfo->bytes_super + meta_sinfo->bytes_root +
3183                 meta_sinfo->bytes_may_use + meta_sinfo->bytes_delalloc;
3184
3185         if (used > meta_sinfo->total_bytes) {
3186                 retries++;
3187                 if (retries == 1) {
3188                         if (maybe_allocate_chunk(root, meta_sinfo))
3189                                 goto again;
3190                         retries++;
3191                 } else {
3192                         spin_unlock(&meta_sinfo->lock);
3193                 }
3194
3195                 if (retries == 2) {
3196                         flush_delalloc(root, meta_sinfo);
3197                         goto again;
3198                 }
3199                 spin_lock(&meta_sinfo->lock);
3200                 meta_sinfo->bytes_may_use -= num_bytes;
3201                 spin_unlock(&meta_sinfo->lock);
3202
3203                 dump_space_info(meta_sinfo, 0, 0);
3204                 return -ENOSPC;
3205         }
3206
3207         check_force_delalloc(meta_sinfo);
3208         spin_unlock(&meta_sinfo->lock);
3209
3210         return 0;
3211 }
3212
3213 /*
3214  * This will check the space that the inode allocates from to make sure we have
3215  * enough space for bytes.
3216  */
3217 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
3218                                 u64 bytes)
3219 {
3220         struct btrfs_space_info *data_sinfo;
3221         int ret = 0, committed = 0;
3222
3223         /* make sure bytes are sectorsize aligned */
3224         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3225
3226         data_sinfo = BTRFS_I(inode)->space_info;
3227         if (!data_sinfo)
3228                 goto alloc;
3229
3230 again:
3231         /* make sure we have enough space to handle the data first */
3232         spin_lock(&data_sinfo->lock);
3233         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
3234             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
3235             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
3236             data_sinfo->bytes_may_use - data_sinfo->bytes_super < bytes) {
3237                 struct btrfs_trans_handle *trans;
3238
3239                 /*
3240                  * if we don't have enough free bytes in this space then we need
3241                  * to alloc a new chunk.
3242                  */
3243                 if (!data_sinfo->full) {
3244                         u64 alloc_target;
3245
3246                         data_sinfo->force_alloc = 1;
3247                         spin_unlock(&data_sinfo->lock);
3248 alloc:
3249                         alloc_target = btrfs_get_alloc_profile(root, 1);
3250                         trans = btrfs_start_transaction(root, 1);
3251                         if (!trans)
3252                                 return -ENOMEM;
3253
3254                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3255                                              bytes + 2 * 1024 * 1024,
3256                                              alloc_target, 0);
3257                         btrfs_end_transaction(trans, root);
3258                         if (ret)
3259                                 return ret;
3260
3261                         if (!data_sinfo) {
3262                                 btrfs_set_inode_space_info(root, inode);
3263                                 data_sinfo = BTRFS_I(inode)->space_info;
3264                         }
3265                         goto again;
3266                 }
3267                 spin_unlock(&data_sinfo->lock);
3268
3269                 /* commit the current transaction and try again */
3270                 if (!committed && !root->fs_info->open_ioctl_trans) {
3271                         committed = 1;
3272                         trans = btrfs_join_transaction(root, 1);
3273                         if (!trans)
3274                                 return -ENOMEM;
3275                         ret = btrfs_commit_transaction(trans, root);
3276                         if (ret)
3277                                 return ret;
3278                         goto again;
3279                 }
3280
3281                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
3282                        ", %llu bytes_used, %llu bytes_reserved, "
3283                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
3284                        "%llu total\n", (unsigned long long)bytes,
3285                        (unsigned long long)data_sinfo->bytes_delalloc,
3286                        (unsigned long long)data_sinfo->bytes_used,
3287                        (unsigned long long)data_sinfo->bytes_reserved,
3288                        (unsigned long long)data_sinfo->bytes_pinned,
3289                        (unsigned long long)data_sinfo->bytes_readonly,
3290                        (unsigned long long)data_sinfo->bytes_may_use,
3291                        (unsigned long long)data_sinfo->total_bytes);
3292                 return -ENOSPC;
3293         }
3294         data_sinfo->bytes_may_use += bytes;
3295         BTRFS_I(inode)->reserved_bytes += bytes;
3296         spin_unlock(&data_sinfo->lock);
3297
3298         return 0;
3299 }
3300
3301 /*
3302  * if there was an error for whatever reason after calling
3303  * btrfs_check_data_free_space, call this so we can cleanup the counters.
3304  */
3305 void btrfs_free_reserved_data_space(struct btrfs_root *root,
3306                                     struct inode *inode, u64 bytes)
3307 {
3308         struct btrfs_space_info *data_sinfo;
3309
3310         /* make sure bytes are sectorsize aligned */
3311         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
3312
3313         data_sinfo = BTRFS_I(inode)->space_info;
3314         spin_lock(&data_sinfo->lock);
3315         data_sinfo->bytes_may_use -= bytes;
3316         BTRFS_I(inode)->reserved_bytes -= bytes;
3317         spin_unlock(&data_sinfo->lock);
3318 }
3319
3320 /* called when we are adding a delalloc extent to the inode's io_tree */
3321 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
3322                                   u64 bytes)
3323 {
3324         struct btrfs_space_info *data_sinfo;
3325
3326         /* get the space info for where this inode will be storing its data */
3327         data_sinfo = BTRFS_I(inode)->space_info;
3328
3329         /* make sure we have enough space to handle the data first */
3330         spin_lock(&data_sinfo->lock);
3331         data_sinfo->bytes_delalloc += bytes;
3332
3333         /*
3334          * we are adding a delalloc extent without calling
3335          * btrfs_check_data_free_space first.  This happens on a weird
3336          * writepage condition, but shouldn't hurt our accounting
3337          */
3338         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
3339                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
3340                 BTRFS_I(inode)->reserved_bytes = 0;
3341         } else {
3342                 data_sinfo->bytes_may_use -= bytes;
3343                 BTRFS_I(inode)->reserved_bytes -= bytes;
3344         }
3345
3346         spin_unlock(&data_sinfo->lock);
3347 }
3348
3349 /* called when we are clearing an delalloc extent from the inode's io_tree */
3350 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
3351                               u64 bytes)
3352 {
3353         struct btrfs_space_info *info;
3354
3355         info = BTRFS_I(inode)->space_info;
3356
3357         spin_lock(&info->lock);
3358         info->bytes_delalloc -= bytes;
3359         spin_unlock(&info->lock);
3360 }
3361
3362 static void force_metadata_allocation(struct btrfs_fs_info *info)
3363 {
3364         struct list_head *head = &info->space_info;
3365         struct btrfs_space_info *found;
3366
3367         rcu_read_lock();
3368         list_for_each_entry_rcu(found, head, list) {
3369                 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3370                         found->force_alloc = 1;
3371         }
3372         rcu_read_unlock();
3373 }
3374
3375 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3376                           struct btrfs_root *extent_root, u64 alloc_bytes,
3377                           u64 flags, int force)
3378 {
3379         struct btrfs_space_info *space_info;
3380         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3381         u64 thresh;
3382         int ret = 0;
3383
3384         mutex_lock(&fs_info->chunk_mutex);
3385
3386         flags = btrfs_reduce_alloc_profile(extent_root, flags);
3387
3388         space_info = __find_space_info(extent_root->fs_info, flags);
3389         if (!space_info) {
3390                 ret = update_space_info(extent_root->fs_info, flags,
3391                                         0, 0, &space_info);
3392                 BUG_ON(ret);
3393         }
3394         BUG_ON(!space_info);
3395
3396         spin_lock(&space_info->lock);
3397         if (space_info->force_alloc)
3398                 force = 1;
3399         if (space_info->full) {
3400                 spin_unlock(&space_info->lock);
3401                 goto out;
3402         }
3403
3404         thresh = space_info->total_bytes - space_info->bytes_readonly;
3405         thresh = div_factor(thresh, 8);
3406         if (!force &&
3407            (space_info->bytes_used + space_info->bytes_pinned +
3408             space_info->bytes_reserved + alloc_bytes) < thresh) {
3409                 spin_unlock(&space_info->lock);
3410                 goto out;
3411         }
3412         spin_unlock(&space_info->lock);
3413
3414         /*
3415          * if we're doing a data chunk, go ahead and make sure that
3416          * we keep a reasonable number of metadata chunks allocated in the
3417          * FS as well.
3418          */
3419         if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3420                 fs_info->data_chunk_allocations++;
3421                 if (!(fs_info->data_chunk_allocations %
3422                       fs_info->metadata_ratio))
3423                         force_metadata_allocation(fs_info);
3424         }
3425
3426         ret = btrfs_alloc_chunk(trans, extent_root, flags);
3427         spin_lock(&space_info->lock);
3428         if (ret)
3429                 space_info->full = 1;
3430         space_info->force_alloc = 0;
3431         spin_unlock(&space_info->lock);
3432 out:
3433         mutex_unlock(&extent_root->fs_info->chunk_mutex);
3434         return ret;
3435 }
3436
3437 static int update_block_group(struct btrfs_trans_handle *trans,
3438                               struct btrfs_root *root,
3439                               u64 bytenr, u64 num_bytes, int alloc,
3440                               int mark_free)
3441 {
3442         struct btrfs_block_group_cache *cache;
3443         struct btrfs_fs_info *info = root->fs_info;
3444         u64 total = num_bytes;
3445         u64 old_val;
3446         u64 byte_in_group;
3447
3448         /* block accounting for super block */
3449         spin_lock(&info->delalloc_lock);
3450         old_val = btrfs_super_bytes_used(&info->super_copy);
3451         if (alloc)
3452                 old_val += num_bytes;
3453         else
3454                 old_val -= num_bytes;
3455         btrfs_set_super_bytes_used(&info->super_copy, old_val);
3456
3457         /* block accounting for root item */
3458         old_val = btrfs_root_used(&root->root_item);
3459         if (alloc)
3460                 old_val += num_bytes;
3461         else
3462                 old_val -= num_bytes;
3463         btrfs_set_root_used(&root->root_item, old_val);
3464         spin_unlock(&info->delalloc_lock);
3465
3466         while (total) {
3467                 cache = btrfs_lookup_block_group(info, bytenr);
3468                 if (!cache)
3469                         return -1;
3470                 byte_in_group = bytenr - cache->key.objectid;
3471                 WARN_ON(byte_in_group > cache->key.offset);
3472
3473                 spin_lock(&cache->space_info->lock);
3474                 spin_lock(&cache->lock);
3475                 cache->dirty = 1;
3476                 old_val = btrfs_block_group_used(&cache->item);
3477                 num_bytes = min(total, cache->key.offset - byte_in_group);
3478                 if (alloc) {
3479                         old_val += num_bytes;
3480                         btrfs_set_block_group_used(&cache->item, old_val);
3481                         cache->reserved -= num_bytes;
3482                         cache->space_info->bytes_used += num_bytes;
3483                         cache->space_info->bytes_reserved -= num_bytes;
3484                         if (cache->ro)
3485                                 cache->space_info->bytes_readonly -= num_bytes;
3486                         spin_unlock(&cache->lock);
3487                         spin_unlock(&cache->space_info->lock);
3488                 } else {
3489                         old_val -= num_bytes;
3490                         cache->space_info->bytes_used -= num_bytes;
3491                         if (cache->ro)
3492                                 cache->space_info->bytes_readonly += num_bytes;
3493                         btrfs_set_block_group_used(&cache->item, old_val);
3494                         spin_unlock(&cache->lock);
3495                         spin_unlock(&cache->space_info->lock);
3496                         if (mark_free) {
3497                                 int ret;
3498
3499                                 ret = btrfs_discard_extent(root, bytenr,
3500                                                            num_bytes);
3501                                 WARN_ON(ret);
3502
3503                                 ret = btrfs_add_free_space(cache, bytenr,
3504                                                            num_bytes);
3505                                 WARN_ON(ret);
3506                         }
3507                 }
3508                 btrfs_put_block_group(cache);
3509                 total -= num_bytes;
3510                 bytenr += num_bytes;
3511         }
3512         return 0;
3513 }
3514
3515 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
3516 {
3517         struct btrfs_block_group_cache *cache;
3518         u64 bytenr;
3519
3520         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
3521         if (!cache)
3522                 return 0;
3523
3524         bytenr = cache->key.objectid;
3525         btrfs_put_block_group(cache);
3526
3527         return bytenr;
3528 }
3529
3530 /*
3531  * this function must be called within transaction
3532  */
3533 int btrfs_pin_extent(struct btrfs_root *root,
3534                      u64 bytenr, u64 num_bytes, int reserved)
3535 {
3536         struct btrfs_fs_info *fs_info = root->fs_info;
3537         struct btrfs_block_group_cache *cache;
3538
3539         cache = btrfs_lookup_block_group(fs_info, bytenr);
3540         BUG_ON(!cache);
3541
3542         spin_lock(&cache->space_info->lock);
3543         spin_lock(&cache->lock);
3544         cache->pinned += num_bytes;
3545         cache->space_info->bytes_pinned += num_bytes;
3546         if (reserved) {
3547                 cache->reserved -= num_bytes;
3548                 cache->space_info->bytes_reserved -= num_bytes;
3549         }
3550         spin_unlock(&cache->lock);
3551         spin_unlock(&cache->space_info->lock);
3552
3553         btrfs_put_block_group(cache);
3554
3555         set_extent_dirty(fs_info->pinned_extents,
3556                          bytenr, bytenr + num_bytes - 1, GFP_NOFS);
3557         return 0;
3558 }
3559
3560 static int update_reserved_extents(struct btrfs_block_group_cache *cache,
3561                                    u64 num_bytes, int reserve)
3562 {
3563         spin_lock(&cache->space_info->lock);
3564         spin_lock(&cache->lock);
3565         if (reserve) {
3566                 cache->reserved += num_bytes;
3567                 cache->space_info->bytes_reserved += num_bytes;
3568         } else {
3569                 cache->reserved -= num_bytes;
3570                 cache->space_info->bytes_reserved -= num_bytes;
3571         }
3572         spin_unlock(&cache->lock);
3573         spin_unlock(&cache->space_info->lock);
3574         return 0;
3575 }
3576
3577 int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
3578                                 struct btrfs_root *root)
3579 {
3580         struct btrfs_fs_info *fs_info = root->fs_info;
3581         struct btrfs_caching_control *next;
3582         struct btrfs_caching_control *caching_ctl;
3583         struct btrfs_block_group_cache *cache;
3584
3585         down_write(&fs_info->extent_commit_sem);
3586
3587         list_for_each_entry_safe(caching_ctl, next,
3588                                  &fs_info->caching_block_groups, list) {
3589                 cache = caching_ctl->block_group;
3590                 if (block_group_cache_done(cache)) {
3591                         cache->last_byte_to_unpin = (u64)-1;
3592                         list_del_init(&caching_ctl->list);
3593                         put_caching_control(caching_ctl);
3594                 } else {
3595                         cache->last_byte_to_unpin = caching_ctl->progress;
3596                 }
3597         }
3598
3599         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3600                 fs_info->pinned_extents = &fs_info->freed_extents[1];
3601         else
3602                 fs_info->pinned_extents = &fs_info->freed_extents[0];
3603
3604         up_write(&fs_info->extent_commit_sem);
3605         return 0;
3606 }
3607
3608 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
3609 {
3610         struct btrfs_fs_info *fs_info = root->fs_info;
3611         struct btrfs_block_group_cache *cache = NULL;
3612         u64 len;
3613
3614         while (start <= end) {
3615                 if (!cache ||
3616                     start >= cache->key.objectid + cache->key.offset) {
3617                         if (cache)
3618                                 btrfs_put_block_group(cache);
3619                         cache = btrfs_lookup_block_group(fs_info, start);
3620                         BUG_ON(!cache);
3621                 }
3622
3623                 len = cache->key.objectid + cache->key.offset - start;
3624                 len = min(len, end + 1 - start);
3625
3626                 if (start < cache->last_byte_to_unpin) {
3627                         len = min(len, cache->last_byte_to_unpin - start);
3628                         btrfs_add_free_space(cache, start, len);
3629                 }
3630
3631                 spin_lock(&cache->space_info->lock);
3632                 spin_lock(&cache->lock);
3633                 cache->pinned -= len;
3634                 cache->space_info->bytes_pinned -= len;
3635                 spin_unlock(&cache->lock);
3636                 spin_unlock(&cache->space_info->lock);
3637
3638                 start += len;
3639         }
3640
3641         if (cache)
3642                 btrfs_put_block_group(cache);
3643         return 0;
3644 }
3645
3646 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
3647                                struct btrfs_root *root)
3648 {
3649         struct btrfs_fs_info *fs_info = root->fs_info;
3650         struct extent_io_tree *unpin;
3651         u64 start;
3652         u64 end;
3653         int ret;
3654
3655         if (fs_info->pinned_extents == &fs_info->freed_extents[0])
3656                 unpin = &fs_info->freed_extents[1];
3657         else
3658                 unpin = &fs_info->freed_extents[0];
3659
3660         while (1) {
3661                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3662                                             EXTENT_DIRTY);
3663                 if (ret)
3664                         break;
3665
3666                 ret = btrfs_discard_extent(root, start, end + 1 - start);
3667
3668                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3669                 unpin_extent_range(root, start, end);
3670                 cond_resched();
3671         }
3672
3673         return ret;
3674 }
3675
3676 static int pin_down_bytes(struct btrfs_trans_handle *trans,
3677                           struct btrfs_root *root,
3678                           struct btrfs_path *path,
3679                           u64 bytenr, u64 num_bytes,
3680                           int is_data, int reserved,
3681                           struct extent_buffer **must_clean)
3682 {
3683         int err = 0;
3684         struct extent_buffer *buf;
3685
3686         if (is_data)
3687                 goto pinit;
3688
3689         /*
3690          * discard is sloooow, and so triggering discards on
3691          * individual btree blocks isn't a good plan.  Just
3692          * pin everything in discard mode.
3693          */
3694         if (btrfs_test_opt(root, DISCARD))
3695                 goto pinit;
3696
3697         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
3698         if (!buf)
3699                 goto pinit;
3700
3701         /* we can reuse a block if it hasn't been written
3702          * and it is from this transaction.  We can't
3703          * reuse anything from the tree log root because
3704          * it has tiny sub-transactions.
3705          */
3706         if (btrfs_buffer_uptodate(buf, 0) &&
3707             btrfs_try_tree_lock(buf)) {
3708                 u64 header_owner = btrfs_header_owner(buf);
3709                 u64 header_transid = btrfs_header_generation(buf);
3710                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
3711                     header_transid == trans->transid &&
3712                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
3713                         *must_clean = buf;
3714                         return 1;
3715                 }
3716                 btrfs_tree_unlock(buf);
3717         }
3718         free_extent_buffer(buf);
3719 pinit:
3720         if (path)
3721                 btrfs_set_path_blocking(path);
3722         /* unlocks the pinned mutex */
3723         btrfs_pin_extent(root, bytenr, num_bytes, reserved);
3724
3725         BUG_ON(err < 0);
3726         return 0;
3727 }
3728
3729 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
3730                                 struct btrfs_root *root,
3731                                 u64 bytenr, u64 num_bytes, u64 parent,
3732                                 u64 root_objectid, u64 owner_objectid,
3733                                 u64 owner_offset, int refs_to_drop,
3734                                 struct btrfs_delayed_extent_op *extent_op)
3735 {
3736         struct btrfs_key key;
3737         struct btrfs_path *path;
3738         struct btrfs_fs_info *info = root->fs_info;
3739         struct btrfs_root *extent_root = info->extent_root;
3740         struct extent_buffer *leaf;
3741         struct btrfs_extent_item *ei;
3742         struct btrfs_extent_inline_ref *iref;
3743         int ret;
3744         int is_data;
3745         int extent_slot = 0;
3746         int found_extent = 0;
3747         int num_to_del = 1;
3748         u32 item_size;
3749         u64 refs;
3750
3751         path = btrfs_alloc_path();
3752         if (!path)
3753                 return -ENOMEM;
3754
3755         path->reada = 1;
3756         path->leave_spinning = 1;
3757
3758         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
3759         BUG_ON(!is_data && refs_to_drop != 1);
3760
3761         ret = lookup_extent_backref(trans, extent_root, path, &iref,
3762                                     bytenr, num_bytes, parent,
3763                                     root_objectid, owner_objectid,
3764                                     owner_offset);
3765         if (ret == 0) {
3766                 extent_slot = path->slots[0];
3767                 while (extent_slot >= 0) {
3768                         btrfs_item_key_to_cpu(path->nodes[0], &key,
3769                                               extent_slot);
3770                         if (key.objectid != bytenr)
3771                                 break;
3772                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3773                             key.offset == num_bytes) {
3774                                 found_extent = 1;
3775                                 break;
3776                         }
3777                         if (path->slots[0] - extent_slot > 5)
3778                                 break;
3779                         extent_slot--;
3780                 }
3781 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3782                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
3783                 if (found_extent && item_size < sizeof(*ei))
3784                         found_extent = 0;
3785 #endif
3786                 if (!found_extent) {
3787                         BUG_ON(iref);
3788                         ret = remove_extent_backref(trans, extent_root, path,
3789                                                     NULL, refs_to_drop,
3790                                                     is_data);
3791                         BUG_ON(ret);
3792                         btrfs_release_path(extent_root, path);
3793                         path->leave_spinning = 1;
3794
3795                         key.objectid = bytenr;
3796                         key.type = BTRFS_EXTENT_ITEM_KEY;
3797                         key.offset = num_bytes;
3798
3799                         ret = btrfs_search_slot(trans, extent_root,
3800                                                 &key, path, -1, 1);
3801                         if (ret) {
3802                                 printk(KERN_ERR "umm, got %d back from search"
3803                                        ", was looking for %llu\n", ret,
3804                                        (unsigned long long)bytenr);
3805                                 btrfs_print_leaf(extent_root, path->nodes[0]);
3806                         }
3807                         BUG_ON(ret);
3808                         extent_slot = path->slots[0];
3809                 }
3810         } else {
3811                 btrfs_print_leaf(extent_root, path->nodes[0]);
3812                 WARN_ON(1);
3813                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
3814                        "parent %llu root %llu  owner %llu offset %llu\n",
3815                        (unsigned long long)bytenr,
3816                        (unsigned long long)parent,
3817                        (unsigned long long)root_objectid,
3818                        (unsigned long long)owner_objectid,
3819                        (unsigned long long)owner_offset);
3820         }
3821
3822         leaf = path->nodes[0];
3823         item_size = btrfs_item_size_nr(leaf, extent_slot);
3824 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3825         if (item_size < sizeof(*ei)) {
3826                 BUG_ON(found_extent || extent_slot != path->slots[0]);
3827                 ret = convert_extent_item_v0(trans, extent_root, path,
3828                                              owner_objectid, 0);
3829                 BUG_ON(ret < 0);
3830
3831                 btrfs_release_path(extent_root, path);
3832                 path->leave_spinning = 1;
3833
3834                 key.objectid = bytenr;
3835                 key.type = BTRFS_EXTENT_ITEM_KEY;
3836                 key.offset = num_bytes;
3837
3838                 ret = btrfs_search_slot(trans, extent_root, &key, path,
3839                                         -1, 1);
3840                 if (ret) {
3841                         printk(KERN_ERR "umm, got %d back from search"
3842                                ", was looking for %llu\n", ret,
3843                                (unsigned long long)bytenr);
3844                         btrfs_print_leaf(extent_root, path->nodes[0]);
3845                 }
3846                 BUG_ON(ret);
3847                 extent_slot = path->slots[0];
3848                 leaf = path->nodes[0];
3849                 item_size = btrfs_item_size_nr(leaf, extent_slot);
3850         }
3851 #endif
3852         BUG_ON(item_size < sizeof(*ei));
3853         ei = btrfs_item_ptr(leaf, extent_slot,
3854                             struct btrfs_extent_item);
3855         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3856                 struct btrfs_tree_block_info *bi;
3857                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3858                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3859                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
3860         }
3861
3862         refs = btrfs_extent_refs(leaf, ei);
3863         BUG_ON(refs < refs_to_drop);
3864         refs -= refs_to_drop;
3865
3866         if (refs > 0) {
3867                 if (extent_op)
3868                         __run_delayed_extent_op(extent_op, leaf, ei);
3869                 /*
3870                  * In the case of inline back ref, reference count will
3871                  * be updated by remove_extent_backref
3872                  */
3873                 if (iref) {
3874                         BUG_ON(!found_extent);
3875                 } else {
3876                         btrfs_set_extent_refs(leaf, ei, refs);
3877                         btrfs_mark_buffer_dirty(leaf);
3878                 }
3879                 if (found_extent) {
3880                         ret = remove_extent_backref(trans, extent_root, path,
3881                                                     iref, refs_to_drop,
3882                                                     is_data);
3883                         BUG_ON(ret);
3884                 }
3885         } else {
3886                 int mark_free = 0;
3887                 struct extent_buffer *must_clean = NULL;
3888
3889                 if (found_extent) {
3890                         BUG_ON(is_data && refs_to_drop !=
3891                                extent_data_ref_count(root, path, iref));
3892                         if (iref) {
3893                                 BUG_ON(path->slots[0] != extent_slot);
3894                         } else {
3895                                 BUG_ON(path->slots[0] != extent_slot + 1);
3896                                 path->slots[0] = extent_slot;
3897                                 num_to_del = 2;
3898                         }
3899                 }
3900
3901                 ret = pin_down_bytes(trans, root, path, bytenr,
3902                                      num_bytes, is_data, 0, &must_clean);
3903                 if (ret > 0)
3904                         mark_free = 1;
3905                 BUG_ON(ret < 0);
3906                 /*
3907                  * it is going to be very rare for someone to be waiting
3908                  * on the block we're freeing.  del_items might need to
3909                  * schedule, so rather than get fancy, just force it
3910                  * to blocking here
3911                  */
3912                 if (must_clean)
3913                         btrfs_set_lock_blocking(must_clean);
3914
3915                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3916                                       num_to_del);
3917                 BUG_ON(ret);
3918                 btrfs_release_path(extent_root, path);
3919
3920                 if (must_clean) {
3921                         clean_tree_block(NULL, root, must_clean);
3922                         btrfs_tree_unlock(must_clean);
3923                         free_extent_buffer(must_clean);
3924                 }
3925
3926                 if (is_data) {
3927                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
3928                         BUG_ON(ret);
3929                 } else {
3930                         invalidate_mapping_pages(info->btree_inode->i_mapping,
3931                              bytenr >> PAGE_CACHE_SHIFT,
3932                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
3933                 }
3934
3935                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
3936                                          mark_free);
3937                 BUG_ON(ret);
3938         }
3939         btrfs_free_path(path);
3940         return ret;
3941 }
3942
3943 /*
3944  * when we free an extent, it is possible (and likely) that we free the last
3945  * delayed ref for that extent as well.  This searches the delayed ref tree for
3946  * a given extent, and if there are no other delayed refs to be processed, it
3947  * removes it from the tree.
3948  */
3949 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
3950                                       struct btrfs_root *root, u64 bytenr)
3951 {
3952         struct btrfs_delayed_ref_head *head;
3953         struct btrfs_delayed_ref_root *delayed_refs;
3954         struct btrfs_delayed_ref_node *ref;
3955         struct rb_node *node;
3956         int ret;
3957
3958         delayed_refs = &trans->transaction->delayed_refs;
3959         spin_lock(&delayed_refs->lock);
3960         head = btrfs_find_delayed_ref_head(trans, bytenr);
3961         if (!head)
3962                 goto out;
3963
3964         node = rb_prev(&head->node.rb_node);
3965         if (!node)
3966                 goto out;
3967
3968         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3969
3970         /* there are still entries for this ref, we can't drop it */
3971         if (ref->bytenr == bytenr)
3972                 goto out;
3973
3974         if (head->extent_op) {
3975                 if (!head->must_insert_reserved)
3976                         goto out;
3977                 kfree(head->extent_op);
3978                 head->extent_op = NULL;
3979         }
3980
3981         /*
3982          * waiting for the lock here would deadlock.  If someone else has it
3983          * locked they are already in the process of dropping it anyway
3984          */
3985         if (!mutex_trylock(&head->mutex))
3986                 goto out;
3987
3988         /*
3989          * at this point we have a head with no other entries.  Go
3990          * ahead and process it.
3991          */
3992         head->node.in_tree = 0;
3993         rb_erase(&head->node.rb_node, &delayed_refs->root);
3994
3995         delayed_refs->num_entries--;
3996
3997         /*
3998          * we don't take a ref on the node because we're removing it from the
3999          * tree, so we just steal the ref the tree was holding.
4000          */
4001         delayed_refs->num_heads--;
4002         if (list_empty(&head->cluster))
4003                 delayed_refs->num_heads_ready--;
4004
4005         list_del_init(&head->cluster);
4006         spin_unlock(&delayed_refs->lock);
4007
4008         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
4009                                   &head->node, head->extent_op,
4010                                   head->must_insert_reserved);
4011         BUG_ON(ret);
4012         btrfs_put_delayed_ref(&head->node);
4013         return 0;
4014 out:
4015         spin_unlock(&delayed_refs->lock);
4016         return 0;
4017 }
4018
4019 int btrfs_free_extent(struct btrfs_trans_handle *trans,
4020                       struct btrfs_root *root,
4021                       u64 bytenr, u64 num_bytes, u64 parent,
4022                       u64 root_objectid, u64 owner, u64 offset)
4023 {
4024         int ret;
4025
4026         /*
4027          * tree log blocks never actually go into the extent allocation
4028          * tree, just update pinning info and exit early.
4029          */
4030         if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
4031                 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
4032                 /* unlocks the pinned mutex */
4033                 btrfs_pin_extent(root, bytenr, num_bytes, 1);
4034                 ret = 0;
4035         } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
4036                 ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
4037                                         parent, root_objectid, (int)owner,
4038                                         BTRFS_DROP_DELAYED_REF, NULL);
4039                 BUG_ON(ret);
4040                 ret = check_ref_cleanup(trans, root, bytenr);
4041                 BUG_ON(ret);
4042         } else {
4043                 ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
4044                                         parent, root_objectid, owner,
4045                                         offset, BTRFS_DROP_DELAYED_REF, NULL);
4046                 BUG_ON(ret);
4047         }
4048         return ret;
4049 }
4050
4051 static u64 stripe_align(struct btrfs_root *root, u64 val)
4052 {
4053         u64 mask = ((u64)root->stripesize - 1);
4054         u64 ret = (val + mask) & ~mask;
4055         return ret;
4056 }
4057
4058 /*
4059  * when we wait for progress in the block group caching, its because
4060  * our allocation attempt failed at least once.  So, we must sleep
4061  * and let some progress happen before we try again.
4062  *
4063  * This function will sleep at least once waiting for new free space to
4064  * show up, and then it will check the block group free space numbers
4065  * for our min num_bytes.  Another option is to have it go ahead
4066  * and look in the rbtree for a free extent of a given size, but this
4067  * is a good start.
4068  */
4069 static noinline int
4070 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
4071                                 u64 num_bytes)
4072 {
4073         struct btrfs_caching_control *caching_ctl;
4074         DEFINE_WAIT(wait);
4075
4076         caching_ctl = get_caching_control(cache);
4077         if (!caching_ctl)
4078                 return 0;
4079
4080         wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
4081                    (cache->free_space >= num_bytes));
4082
4083         put_caching_control(caching_ctl);
4084         return 0;
4085 }
4086
4087 static noinline int
4088 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
4089 {
4090         struct btrfs_caching_control *caching_ctl;
4091         DEFINE_WAIT(wait);
4092
4093         caching_ctl = get_caching_control(cache);
4094         if (!caching_ctl)
4095                 return 0;
4096
4097         wait_event(caching_ctl->wait, block_group_cache_done(cache));
4098
4099         put_caching_control(caching_ctl);
4100         return 0;
4101 }
4102
4103 enum btrfs_loop_type {
4104         LOOP_FIND_IDEAL = 0,
4105         LOOP_CACHING_NOWAIT = 1,
4106         LOOP_CACHING_WAIT = 2,
4107         LOOP_ALLOC_CHUNK = 3,
4108         LOOP_NO_EMPTY_SIZE = 4,
4109 };
4110
4111 /*
4112  * walks the btree of allocated extents and find a hole of a given size.
4113  * The key ins is changed to record the hole:
4114  * ins->objectid == block start
4115  * ins->flags = BTRFS_EXTENT_ITEM_KEY
4116  * ins->offset == number of blocks
4117  * Any available blocks before search_start are skipped.
4118  */
4119 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
4120                                      struct btrfs_root *orig_root,
4121                                      u64 num_bytes, u64 empty_size,
4122                                      u64 search_start, u64 search_end,
4123                                      u64 hint_byte, struct btrfs_key *ins,
4124                                      u64 exclude_start, u64 exclude_nr,
4125                                      int data)
4126 {
4127         int ret = 0;
4128         struct btrfs_root *root = orig_root->fs_info->extent_root;
4129         struct btrfs_free_cluster *last_ptr = NULL;
4130         struct btrfs_block_group_cache *block_group = NULL;
4131         int empty_cluster = 2 * 1024 * 1024;
4132         int allowed_chunk_alloc = 0;
4133         int done_chunk_alloc = 0;
4134         struct btrfs_space_info *space_info;
4135         int last_ptr_loop = 0;
4136         int loop = 0;
4137         bool found_uncached_bg = false;
4138         bool failed_cluster_refill = false;
4139         bool failed_alloc = false;
4140         u64 ideal_cache_percent = 0;
4141         u64 ideal_cache_offset = 0;
4142
4143         WARN_ON(num_bytes < root->sectorsize);
4144         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
4145         ins->objectid = 0;
4146         ins->offset = 0;
4147
4148         space_info = __find_space_info(root->fs_info, data);
4149
4150         if (orig_root->ref_cows || empty_size)
4151                 allowed_chunk_alloc = 1;
4152
4153         if (data & BTRFS_BLOCK_GROUP_METADATA) {
4154                 last_ptr = &root->fs_info->meta_alloc_cluster;
4155                 if (!btrfs_test_opt(root, SSD))
4156                         empty_cluster = 64 * 1024;
4157         }
4158
4159         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
4160                 last_ptr = &root->fs_info->data_alloc_cluster;
4161         }
4162
4163         if (last_ptr) {
4164                 spin_lock(&last_ptr->lock);
4165                 if (last_ptr->block_group)
4166                         hint_byte = last_ptr->window_start;
4167                 spin_unlock(&last_ptr->lock);
4168         }
4169
4170         search_start = max(search_start, first_logical_byte(root, 0));
4171         search_start = max(search_start, hint_byte);
4172
4173         if (!last_ptr)
4174                 empty_cluster = 0;
4175
4176         if (search_start == hint_byte) {
4177 ideal_cache:
4178                 block_group = btrfs_lookup_block_group(root->fs_info,
4179                                                        search_start);
4180                 /*
4181                  * we don't want to use the block group if it doesn't match our
4182                  * allocation bits, or if its not cached.
4183                  *
4184                  * However if we are re-searching with an ideal block group
4185                  * picked out then we don't care that the block group is cached.
4186                  */
4187                 if (block_group && block_group_bits(block_group, data) &&
4188                     (block_group->cached != BTRFS_CACHE_NO ||
4189                      search_start == ideal_cache_offset)) {
4190                         down_read(&space_info->groups_sem);
4191                         if (list_empty(&block_group->list) ||
4192                             block_group->ro) {
4193                                 /*
4194                                  * someone is removing this block group,
4195                                  * we can't jump into the have_block_group
4196                                  * target because our list pointers are not
4197                                  * valid
4198                                  */
4199                                 btrfs_put_block_group(block_group);
4200                                 up_read(&space_info->groups_sem);
4201                         } else {
4202                                 goto have_block_group;
4203                         }
4204                 } else if (block_group) {
4205                         btrfs_put_block_group(block_group);
4206                 }
4207         }
4208 search:
4209         down_read(&space_info->groups_sem);
4210         list_for_each_entry(block_group, &space_info->block_groups, list) {
4211                 u64 offset;
4212                 int cached;
4213
4214                 atomic_inc(&block_group->count);
4215                 search_start = block_group->key.objectid;
4216
4217 have_block_group:
4218                 if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
4219                         u64 free_percent;
4220
4221                         free_percent = btrfs_block_group_used(&block_group->item);
4222                         free_percent *= 100;
4223                         free_percent = div64_u64(free_percent,
4224                                                  block_group->key.offset);
4225                         free_percent = 100 - free_percent;
4226                         if (free_percent > ideal_cache_percent &&
4227                             likely(!block_group->ro)) {
4228                                 ideal_cache_offset = block_group->key.objectid;
4229                                 ideal_cache_percent = free_percent;
4230                         }
4231
4232                         /*
4233                          * We only want to start kthread caching if we are at
4234                          * the point where we will wait for caching to make
4235                          * progress, or if our ideal search is over and we've
4236                          * found somebody to start caching.
4237                          */
4238                         if (loop > LOOP_CACHING_NOWAIT ||
4239                             (loop > LOOP_FIND_IDEAL &&
4240                              atomic_read(&space_info->caching_threads) < 2)) {
4241                                 ret = cache_block_group(block_group);
4242                                 BUG_ON(ret);
4243                         }
4244                         found_uncached_bg = true;
4245
4246                         /*
4247                          * If loop is set for cached only, try the next block
4248                          * group.
4249                          */
4250                         if (loop == LOOP_FIND_IDEAL)
4251                                 goto loop;
4252                 }
4253
4254                 cached = block_group_cache_done(block_group);
4255                 if (unlikely(!cached))
4256                         found_uncached_bg = true;
4257
4258                 if (unlikely(block_group->ro))
4259                         goto loop;
4260
4261                 /*
4262                  * Ok we want to try and use the cluster allocator, so lets look
4263                  * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
4264                  * have tried the cluster allocator plenty of times at this
4265                  * point and not have found anything, so we are likely way too
4266                  * fragmented for the clustering stuff to find anything, so lets
4267                  * just skip it and let the allocator find whatever block it can
4268                  * find
4269                  */
4270                 if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) {
4271                         /*
4272                          * the refill lock keeps out other
4273                          * people trying to start a new cluster
4274                          */
4275                         spin_lock(&last_ptr->refill_lock);
4276                         if (last_ptr->block_group &&
4277                             (last_ptr->block_group->ro ||
4278                             !block_group_bits(last_ptr->block_group, data))) {
4279                                 offset = 0;
4280                                 goto refill_cluster;
4281                         }
4282
4283                         offset = btrfs_alloc_from_cluster(block_group, last_ptr,
4284                                                  num_bytes, search_start);
4285                         if (offset) {
4286                                 /* we have a block, we're done */
4287                                 spin_unlock(&last_ptr->refill_lock);
4288                                 goto checks;
4289                         }
4290
4291                         spin_lock(&last_ptr->lock);
4292                         /*
4293                          * whoops, this cluster doesn't actually point to
4294                          * this block group.  Get a ref on the block
4295                          * group is does point to and try again
4296                          */
4297                         if (!last_ptr_loop && last_ptr->block_group &&
4298                             last_ptr->block_group != block_group) {
4299
4300                                 btrfs_put_block_group(block_group);
4301                                 block_group = last_ptr->block_group;
4302                                 atomic_inc(&block_group->count);
4303                                 spin_unlock(&last_ptr->lock);
4304                                 spin_unlock(&last_ptr->refill_lock);
4305
4306                                 last_ptr_loop = 1;
4307                                 search_start = block_group->key.objectid;
4308                                 /*
4309                                  * we know this block group is properly
4310                                  * in the list because
4311                                  * btrfs_remove_block_group, drops the
4312                                  * cluster before it removes the block
4313                                  * group from the list
4314                                  */
4315                                 goto have_block_group;
4316                         }
4317                         spin_unlock(&last_ptr->lock);
4318 refill_cluster:
4319                         /*
4320                          * this cluster didn't work out, free it and
4321                          * start over
4322                          */
4323                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4324
4325                         last_ptr_loop = 0;
4326
4327                         /* allocate a cluster in this block group */
4328                         ret = btrfs_find_space_cluster(trans, root,
4329                                                block_group, last_ptr,
4330                                                offset, num_bytes,
4331                                                empty_cluster + empty_size);
4332                         if (ret == 0) {
4333                                 /*
4334                                  * now pull our allocation out of this
4335                                  * cluster
4336                                  */
4337                                 offset = btrfs_alloc_from_cluster(block_group,
4338                                                   last_ptr, num_bytes,
4339                                                   search_start);
4340                                 if (offset) {
4341                                         /* we found one, proceed */
4342                                         spin_unlock(&last_ptr->refill_lock);
4343                                         goto checks;
4344                                 }
4345                         } else if (!cached && loop > LOOP_CACHING_NOWAIT
4346                                    && !failed_cluster_refill) {
4347                                 spin_unlock(&last_ptr->refill_lock);
4348
4349                                 failed_cluster_refill = true;
4350                                 wait_block_group_cache_progress(block_group,
4351                                        num_bytes + empty_cluster + empty_size);
4352                                 goto have_block_group;
4353                         }
4354
4355                         /*
4356                          * at this point we either didn't find a cluster
4357                          * or we weren't able to allocate a block from our
4358                          * cluster.  Free the cluster we've been trying
4359                          * to use, and go to the next block group
4360                          */
4361                         btrfs_return_cluster_to_free_space(NULL, last_ptr);
4362                         spin_unlock(&last_ptr->refill_lock);
4363                         goto loop;
4364                 }
4365
4366                 offset = btrfs_find_space_for_alloc(block_group, search_start,
4367                                                     num_bytes, empty_size);
4368                 /*
4369                  * If we didn't find a chunk, and we haven't failed on this
4370                  * block group before, and this block group is in the middle of
4371                  * caching and we are ok with waiting, then go ahead and wait
4372                  * for progress to be made, and set failed_alloc to true.
4373                  *
4374                  * If failed_alloc is true then we've already waited on this
4375                  * block group once and should move on to the next block group.
4376                  */
4377                 if (!offset && !failed_alloc && !cached &&
4378                     loop > LOOP_CACHING_NOWAIT) {
4379                         wait_block_group_cache_progress(block_group,
4380                                                 num_bytes + empty_size);
4381                         failed_alloc = true;
4382                         goto have_block_group;
4383                 } else if (!offset) {
4384                         goto loop;
4385                 }
4386 checks:
4387                 search_start = stripe_align(root, offset);
4388                 /* move on to the next group */
4389                 if (search_start + num_bytes >= search_end) {
4390                         btrfs_add_free_space(block_group, offset, num_bytes);
4391                         goto loop;
4392                 }
4393
4394                 /* move on to the next group */
4395                 if (search_start + num_bytes >
4396                     block_group->key.objectid + block_group->key.offset) {
4397                         btrfs_add_free_space(block_group, offset, num_bytes);
4398                         goto loop;
4399                 }
4400
4401                 if (exclude_nr > 0 &&
4402                     (search_start + num_bytes > exclude_start &&
4403                      search_start < exclude_start + exclude_nr)) {
4404                         search_start = exclude_start + exclude_nr;
4405
4406                         btrfs_add_free_space(block_group, offset, num_bytes);
4407                         /*
4408                          * if search_start is still in this block group
4409                          * then we just re-search this block group
4410                          */
4411                         if (search_start >= block_group->key.objectid &&
4412                             search_start < (block_group->key.objectid +
4413                                             block_group->key.offset))
4414                                 goto have_block_group;
4415                         goto loop;
4416                 }
4417
4418                 ins->objectid = search_start;
4419                 ins->offset = num_bytes;
4420
4421                 if (offset < search_start)
4422                         btrfs_add_free_space(block_group, offset,
4423                                              search_start - offset);
4424                 BUG_ON(offset > search_start);
4425
4426                 update_reserved_extents(block_group, num_bytes, 1);
4427
4428                 /* we are all good, lets return */
4429                 break;
4430 loop:
4431                 failed_cluster_refill = false;
4432                 failed_alloc = false;
4433                 btrfs_put_block_group(block_group);
4434         }
4435         up_read(&space_info->groups_sem);
4436
4437         /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
4438          *                      for them to make caching progress.  Also
4439          *                      determine the best possible bg to cache
4440          * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
4441          *                      caching kthreads as we move along
4442          * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
4443          * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
4444          * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
4445          *                      again
4446          */
4447         if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
4448             (found_uncached_bg || empty_size || empty_cluster ||
4449              allowed_chunk_alloc)) {
4450                 if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
4451                         found_uncached_bg = false;
4452                         loop++;
4453                         if (!ideal_cache_percent &&
4454                             atomic_read(&space_info->caching_threads))
4455                                 goto search;
4456
4457                         /*
4458                          * 1 of the following 2 things have happened so far
4459                          *
4460                          * 1) We found an ideal block group for caching that
4461                          * is mostly full and will cache quickly, so we might
4462                          * as well wait for it.
4463                          *
4464                          * 2) We searched for cached only and we didn't find
4465                          * anything, and we didn't start any caching kthreads
4466                          * either, so chances are we will loop through and
4467                          * start a couple caching kthreads, and then come back
4468                          * around and just wait for them.  This will be slower
4469                          * because we will have 2 caching kthreads reading at
4470                          * the same time when we could have just started one
4471                          * and waited for it to get far enough to give us an
4472                          * allocation, so go ahead and go to the wait caching
4473                          * loop.
4474                          */
4475                         loop = LOOP_CACHING_WAIT;
4476                         search_start = ideal_cache_offset;
4477                         ideal_cache_percent = 0;
4478                         goto ideal_cache;
4479                 } else if (loop == LOOP_FIND_IDEAL) {
4480                         /*
4481                          * Didn't find a uncached bg, wait on anything we find
4482                          * next.
4483                          */
4484                         loop = LOOP_CACHING_WAIT;
4485                         goto search;
4486                 }
4487
4488                 if (loop < LOOP_CACHING_WAIT) {
4489                         loop++;
4490                         goto search;
4491                 }
4492
4493                 if (loop == LOOP_ALLOC_CHUNK) {
4494                         empty_size = 0;
4495                         empty_cluster = 0;
4496                 }
4497
4498                 if (allowed_chunk_alloc) {
4499                         ret = do_chunk_alloc(trans, root, num_bytes +
4500                                              2 * 1024 * 1024, data, 1);
4501                         allowed_chunk_alloc = 0;
4502                         done_chunk_alloc = 1;
4503                 } else if (!done_chunk_alloc) {
4504                         space_info->force_alloc = 1;
4505                 }
4506
4507                 if (loop < LOOP_NO_EMPTY_SIZE) {
4508                         loop++;
4509                         goto search;
4510                 }
4511                 ret = -ENOSPC;
4512         } else if (!ins->objectid) {
4513                 ret = -ENOSPC;
4514         }
4515
4516         /* we found what we needed */
4517         if (ins->objectid) {
4518                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
4519                         trans->block_group = block_group->key.objectid;
4520
4521                 btrfs_put_block_group(block_group);
4522                 ret = 0;
4523         }
4524
4525         return ret;
4526 }
4527
4528 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
4529                             int dump_block_groups)
4530 {
4531         struct btrfs_block_group_cache *cache;
4532
4533         spin_lock(&info->lock);
4534         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
4535                (unsigned long long)(info->total_bytes - info->bytes_used -
4536                                     info->bytes_pinned - info->bytes_reserved -
4537                                     info->bytes_super),
4538                (info->full) ? "" : "not ");
4539         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
4540                " may_use=%llu, used=%llu, root=%llu, super=%llu, reserved=%llu"
4541                "\n",
4542                (unsigned long long)info->total_bytes,
4543                (unsigned long long)info->bytes_pinned,
4544                (unsigned long long)info->bytes_delalloc,
4545                (unsigned long long)info->bytes_may_use,
4546                (unsigned long long)info->bytes_used,
4547                (unsigned long long)info->bytes_root,
4548                (unsigned long long)info->bytes_super,
4549                (unsigned long long)info->bytes_reserved);
4550         spin_unlock(&info->lock);
4551
4552         if (!dump_block_groups)
4553                 return;
4554
4555         down_read(&info->groups_sem);
4556         list_for_each_entry(cache, &info->block_groups, list) {
4557                 spin_lock(&cache->lock);
4558                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
4559                        "%llu pinned %llu reserved\n",
4560                        (unsigned long long)cache->key.objectid,
4561                        (unsigned long long)cache->key.offset,
4562                        (unsigned long long)btrfs_block_group_used(&cache->item),
4563                        (unsigned long long)cache->pinned,
4564                        (unsigned long long)cache->reserved);
4565                 btrfs_dump_free_space(cache, bytes);
4566                 spin_unlock(&cache->lock);
4567         }
4568         up_read(&info->groups_sem);
4569 }
4570
4571 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
4572                          struct btrfs_root *root,
4573                          u64 num_bytes, u64 min_alloc_size,
4574                          u64 empty_size, u64 hint_byte,
4575                          u64 search_end, struct btrfs_key *ins,
4576                          u64 data)
4577 {
4578         int ret;
4579         u64 search_start = 0;
4580         struct btrfs_fs_info *info = root->fs_info;
4581
4582         data = btrfs_get_alloc_profile(root, data);
4583 again:
4584         /*
4585          * the only place that sets empty_size is btrfs_realloc_node, which
4586          * is not called recursively on allocations
4587          */
4588         if (empty_size || root->ref_cows) {
4589                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
4590                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4591                                      2 * 1024 * 1024,
4592                                      BTRFS_BLOCK_GROUP_METADATA |
4593                                      (info->metadata_alloc_profile &
4594                                       info->avail_metadata_alloc_bits), 0);
4595                 }
4596                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4597                                      num_bytes + 2 * 1024 * 1024, data, 0);
4598         }
4599
4600         WARN_ON(num_bytes < root->sectorsize);
4601         ret = find_free_extent(trans, root, num_bytes, empty_size,
4602                                search_start, search_end, hint_byte, ins,
4603                                trans->alloc_exclude_start,
4604                                trans->alloc_exclude_nr, data);
4605
4606         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
4607                 num_bytes = num_bytes >> 1;
4608                 num_bytes = num_bytes & ~(root->sectorsize - 1);
4609                 num_bytes = max(num_bytes, min_alloc_size);
4610                 do_chunk_alloc(trans, root->fs_info->extent_root,
4611                                num_bytes, data, 1);
4612                 goto again;
4613         }
4614         if (ret == -ENOSPC) {
4615                 struct btrfs_space_info *sinfo;
4616
4617                 sinfo = __find_space_info(root->fs_info, data);
4618                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
4619                        "wanted %llu\n", (unsigned long long)data,
4620                        (unsigned long long)num_bytes);
4621                 dump_space_info(sinfo, num_bytes, 1);
4622         }
4623
4624         return ret;
4625 }
4626
4627 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
4628 {
4629         struct btrfs_block_group_cache *cache;
4630         int ret = 0;
4631
4632         cache = btrfs_lookup_block_group(root->fs_info, start);
4633         if (!cache) {
4634                 printk(KERN_ERR "Unable to find block group for %llu\n",
4635                        (unsigned long long)start);
4636                 return -ENOSPC;
4637         }
4638
4639         ret = btrfs_discard_extent(root, start, len);
4640
4641         btrfs_add_free_space(cache, start, len);
4642         update_reserved_extents(cache, len, 0);
4643         btrfs_put_block_group(cache);
4644
4645         return ret;
4646 }
4647
4648 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4649                                       struct btrfs_root *root,
4650                                       u64 parent, u64 root_objectid,
4651                                       u64 flags, u64 owner, u64 offset,
4652                                       struct btrfs_key *ins, int ref_mod)
4653 {
4654         int ret;
4655         struct btrfs_fs_info *fs_info = root->fs_info;
4656         struct btrfs_extent_item *extent_item;
4657         struct btrfs_extent_inline_ref *iref;
4658         struct btrfs_path *path;
4659         struct extent_buffer *leaf;
4660         int type;
4661         u32 size;
4662
4663         if (parent > 0)
4664                 type = BTRFS_SHARED_DATA_REF_KEY;
4665         else
4666                 type = BTRFS_EXTENT_DATA_REF_KEY;
4667
4668         size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
4669
4670         path = btrfs_alloc_path();
4671         BUG_ON(!path);
4672
4673         path->leave_spinning = 1;
4674         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4675                                       ins, size);
4676         BUG_ON(ret);
4677
4678         leaf = path->nodes[0];
4679         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4680                                      struct btrfs_extent_item);
4681         btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4682         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4683         btrfs_set_extent_flags(leaf, extent_item,
4684                                flags | BTRFS_EXTENT_FLAG_DATA);
4685
4686         iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4687         btrfs_set_extent_inline_ref_type(leaf, iref, type);
4688         if (parent > 0) {
4689                 struct btrfs_shared_data_ref *ref;
4690                 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4691                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4692                 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4693         } else {
4694                 struct btrfs_extent_data_ref *ref;
4695                 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4696                 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4697                 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4698                 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4699                 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4700         }
4701
4702         btrfs_mark_buffer_dirty(path->nodes[0]);
4703         btrfs_free_path(path);
4704
4705         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4706                                  1, 0);
4707         if (ret) {
4708                 printk(KERN_ERR "btrfs update block group failed for %llu "
4709                        "%llu\n", (unsigned long long)ins->objectid,
4710                        (unsigned long long)ins->offset);
4711                 BUG();
4712         }
4713         return ret;
4714 }
4715
4716 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
4717                                      struct btrfs_root *root,
4718                                      u64 parent, u64 root_objectid,
4719                                      u64 flags, struct btrfs_disk_key *key,
4720                                      int level, struct btrfs_key *ins)
4721 {
4722         int ret;
4723         struct btrfs_fs_info *fs_info = root->fs_info;
4724         struct btrfs_extent_item *extent_item;
4725         struct btrfs_tree_block_info *block_info;
4726         struct btrfs_extent_inline_ref *iref;
4727         struct btrfs_path *path;
4728         struct extent_buffer *leaf;
4729         u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
4730
4731         path = btrfs_alloc_path();
4732         BUG_ON(!path);
4733
4734         path->leave_spinning = 1;
4735         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4736                                       ins, size);
4737         BUG_ON(ret);
4738
4739         leaf = path->nodes[0];
4740         extent_item = btrfs_item_ptr(leaf, path->slots[0],
4741                                      struct btrfs_extent_item);
4742         btrfs_set_extent_refs(leaf, extent_item, 1);
4743         btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4744         btrfs_set_extent_flags(leaf, extent_item,
4745                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
4746         block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
4747
4748         btrfs_set_tree_block_key(leaf, block_info, key);
4749         btrfs_set_tree_block_level(leaf, block_info, level);
4750
4751         iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4752         if (parent > 0) {
4753                 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4754                 btrfs_set_extent_inline_ref_type(leaf, iref,
4755                                                  BTRFS_SHARED_BLOCK_REF_KEY);
4756                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4757         } else {
4758                 btrfs_set_extent_inline_ref_type(leaf, iref,
4759                                                  BTRFS_TREE_BLOCK_REF_KEY);
4760                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
4761         }
4762
4763         btrfs_mark_buffer_dirty(leaf);
4764         btrfs_free_path(path);
4765
4766         ret = update_block_group(trans, root, ins->objectid, ins->offset,
4767                                  1, 0);
4768         if (ret) {
4769                 printk(KERN_ERR "btrfs update block group failed for %llu "
4770                        "%llu\n", (unsigned long long)ins->objectid,
4771                        (unsigned long long)ins->offset);
4772                 BUG();
4773         }
4774         return ret;
4775 }
4776
4777 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
4778                                      struct btrfs_root *root,
4779                                      u64 root_objectid, u64 owner,
4780                                      u64 offset, struct btrfs_key *ins)
4781 {
4782         int ret;
4783
4784         BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
4785
4786         ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
4787                                          0, root_objectid, owner, offset,
4788                                          BTRFS_ADD_DELAYED_EXTENT, NULL);
4789         return ret;
4790 }
4791
4792 /*
4793  * this is used by the tree logging recovery code.  It records that
4794  * an extent has been allocated and makes sure to clear the free
4795  * space cache bits as well
4796  */
4797 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
4798                                    struct btrfs_root *root,
4799                                    u64 root_objectid, u64 owner, u64 offset,
4800                                    struct btrfs_key *ins)
4801 {
4802         int ret;
4803         struct btrfs_block_group_cache *block_group;
4804         struct btrfs_caching_control *caching_ctl;
4805         u64 start = ins->objectid;
4806         u64 num_bytes = ins->offset;
4807
4808         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
4809         cache_block_group(block_group);
4810         caching_ctl = get_caching_control(block_group);
4811
4812         if (!caching_ctl) {
4813                 BUG_ON(!block_group_cache_done(block_group));
4814                 ret = btrfs_remove_free_space(block_group, start, num_bytes);
4815                 BUG_ON(ret);
4816         } else {
4817                 mutex_lock(&caching_ctl->mutex);
4818
4819                 if (start >= caching_ctl->progress) {
4820                         ret = add_excluded_extent(root, start, num_bytes);
4821                         BUG_ON(ret);
4822                 } else if (start + num_bytes <= caching_ctl->progress) {
4823                         ret = btrfs_remove_free_space(block_group,
4824                                                       start, num_bytes);
4825                         BUG_ON(ret);
4826                 } else {
4827                         num_bytes = caching_ctl->progress - start;
4828                         ret = btrfs_remove_free_space(block_group,
4829                                                       start, num_bytes);
4830                         BUG_ON(ret);
4831
4832                         start = caching_ctl->progress;
4833                         num_bytes = ins->objectid + ins->offset -
4834                                     caching_ctl->progress;
4835                         ret = add_excluded_extent(root, start, num_bytes);
4836                         BUG_ON(ret);
4837                 }
4838
4839                 mutex_unlock(&caching_ctl->mutex);
4840                 put_caching_control(caching_ctl);
4841         }
4842
4843         update_reserved_extents(block_group, ins->offset, 1);
4844         btrfs_put_block_group(block_group);
4845         ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
4846                                          0, owner, offset, ins, 1);
4847         return ret;
4848 }
4849
4850 /*
4851  * finds a free extent and does all the dirty work required for allocation
4852  * returns the key for the extent through ins, and a tree buffer for
4853  * the first block of the extent through buf.
4854  *
4855  * returns 0 if everything worked, non-zero otherwise.
4856  */
4857 static int alloc_tree_block(struct btrfs_trans_handle *trans,
4858                             struct btrfs_root *root,
4859                             u64 num_bytes, u64 parent, u64 root_objectid,
4860                             struct btrfs_disk_key *key, int level,
4861                             u64 empty_size, u64 hint_byte, u64 search_end,
4862                             struct btrfs_key *ins)
4863 {
4864         int ret;
4865         u64 flags = 0;
4866
4867         ret = btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
4868                                    empty_size, hint_byte, search_end,
4869                                    ins, 0);
4870         if (ret)
4871                 return ret;
4872
4873         if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4874                 if (parent == 0)
4875                         parent = ins->objectid;
4876                 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4877         } else
4878                 BUG_ON(parent > 0);
4879
4880         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
4881                 struct btrfs_delayed_extent_op *extent_op;
4882                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
4883                 BUG_ON(!extent_op);
4884                 if (key)
4885                         memcpy(&extent_op->key, key, sizeof(extent_op->key));
4886                 else
4887                         memset(&extent_op->key, 0, sizeof(extent_op->key));
4888                 extent_op->flags_to_set = flags;
4889                 extent_op->update_key = 1;
4890                 extent_op->update_flags = 1;
4891                 extent_op->is_data = 0;
4892
4893                 ret = btrfs_add_delayed_tree_ref(trans, ins->objectid,
4894                                         ins->offset, parent, root_objectid,
4895                                         level, BTRFS_ADD_DELAYED_EXTENT,
4896                                         extent_op);
4897                 BUG_ON(ret);
4898         }
4899         return ret;
4900 }
4901
4902 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
4903                                             struct btrfs_root *root,
4904                                             u64 bytenr, u32 blocksize,
4905                                             int level)
4906 {
4907         struct extent_buffer *buf;
4908
4909         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
4910         if (!buf)
4911                 return ERR_PTR(-ENOMEM);
4912         btrfs_set_header_generation(buf, trans->transid);
4913         btrfs_set_buffer_lockdep_class(buf, level);
4914         btrfs_tree_lock(buf);
4915         clean_tree_block(trans, root, buf);
4916
4917         btrfs_set_lock_blocking(buf);
4918         btrfs_set_buffer_uptodate(buf);
4919
4920         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
4921                 set_extent_dirty(&root->dirty_log_pages, buf->start,
4922                          buf->start + buf->len - 1, GFP_NOFS);
4923         } else {
4924                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4925                          buf->start + buf->len - 1, GFP_NOFS);
4926         }
4927         trans->blocks_used++;
4928         /* this returns a buffer locked for blocking */
4929         return buf;
4930 }
4931
4932 /*
4933  * helper function to allocate a block for a given tree
4934  * returns the tree buffer or NULL.
4935  */
4936 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
4937                                         struct btrfs_root *root, u32 blocksize,
4938                                         u64 parent, u64 root_objectid,
4939                                         struct btrfs_disk_key *key, int level,
4940                                         u64 hint, u64 empty_size)
4941 {
4942         struct btrfs_key ins;
4943         int ret;
4944         struct extent_buffer *buf;
4945
4946         ret = alloc_tree_block(trans, root, blocksize, parent, root_objectid,
4947                                key, level, empty_size, hint, (u64)-1, &ins);
4948         if (ret) {
4949                 BUG_ON(ret > 0);
4950                 return ERR_PTR(ret);
4951         }
4952
4953         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
4954                                     blocksize, level);
4955         return buf;
4956 }
4957
4958 struct walk_control {
4959         u64 refs[BTRFS_MAX_LEVEL];
4960         u64 flags[BTRFS_MAX_LEVEL];
4961         struct btrfs_key update_progress;
4962         int stage;
4963         int level;
4964         int shared_level;
4965         int update_ref;
4966         int keep_locks;
4967         int reada_slot;
4968         int reada_count;
4969 };
4970
4971 #define DROP_REFERENCE  1
4972 #define UPDATE_BACKREF  2
4973
4974 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4975                                      struct btrfs_root *root,
4976                                      struct walk_control *wc,
4977                                      struct btrfs_path *path)
4978 {
4979         u64 bytenr;
4980         u64 generation;
4981         u64 refs;
4982         u64 flags;
4983         u64 last = 0;
4984         u32 nritems;
4985         u32 blocksize;
4986         struct btrfs_key key;
4987         struct extent_buffer *eb;
4988         int ret;
4989         int slot;
4990         int nread = 0;
4991
4992         if (path->slots[wc->level] < wc->reada_slot) {
4993                 wc->reada_count = wc->reada_count * 2 / 3;
4994                 wc->reada_count = max(wc->reada_count, 2);
4995         } else {
4996                 wc->reada_count = wc->reada_count * 3 / 2;
4997                 wc->reada_count = min_t(int, wc->reada_count,
4998                                         BTRFS_NODEPTRS_PER_BLOCK(root));
4999         }
5000
5001         eb = path->nodes[wc->level];
5002         nritems = btrfs_header_nritems(eb);
5003         blocksize = btrfs_level_size(root, wc->level - 1);
5004
5005         for (slot = path->slots[wc->level]; slot < nritems; slot++) {
5006                 if (nread >= wc->reada_count)
5007                         break;
5008
5009                 cond_resched();
5010                 bytenr = btrfs_node_blockptr(eb, slot);
5011                 generation = btrfs_node_ptr_generation(eb, slot);
5012
5013                 if (slot == path->slots[wc->level])
5014                         goto reada;
5015
5016                 if (wc->stage == UPDATE_BACKREF &&
5017                     generation <= root->root_key.offset)
5018                         continue;
5019
5020                 /* We don't lock the tree block, it's OK to be racy here */
5021                 ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5022                                                &refs, &flags);
5023                 BUG_ON(ret);
5024                 BUG_ON(refs == 0);
5025
5026                 if (wc->stage == DROP_REFERENCE) {
5027                         if (refs == 1)
5028                                 goto reada;
5029
5030                         if (wc->level == 1 &&
5031                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5032                                 continue;
5033                         if (!wc->update_ref ||
5034                             generation <= root->root_key.offset)
5035                                 continue;
5036                         btrfs_node_key_to_cpu(eb, &key, slot);
5037                         ret = btrfs_comp_cpu_keys(&key,
5038                                                   &wc->update_progress);
5039                         if (ret < 0)
5040                                 continue;
5041                 } else {
5042                         if (wc->level == 1 &&
5043                             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5044                                 continue;
5045                 }
5046 reada:
5047                 ret = readahead_tree_block(root, bytenr, blocksize,
5048                                            generation);
5049                 if (ret)
5050                         break;
5051                 last = bytenr + blocksize;
5052                 nread++;
5053         }
5054         wc->reada_slot = slot;
5055 }
5056
5057 /*
5058  * hepler to process tree block while walking down the tree.
5059  *
5060  * when wc->stage == UPDATE_BACKREF, this function updates
5061  * back refs for pointers in the block.
5062  *
5063  * NOTE: return value 1 means we should stop walking down.
5064  */
5065 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
5066                                    struct btrfs_root *root,
5067                                    struct btrfs_path *path,
5068                                    struct walk_control *wc, int lookup_info)
5069 {
5070         int level = wc->level;
5071         struct extent_buffer *eb = path->nodes[level];
5072         u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5073         int ret;
5074
5075         if (wc->stage == UPDATE_BACKREF &&
5076             btrfs_header_owner(eb) != root->root_key.objectid)
5077                 return 1;
5078
5079         /*
5080          * when reference count of tree block is 1, it won't increase
5081          * again. once full backref flag is set, we never clear it.
5082          */
5083         if (lookup_info &&
5084             ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
5085              (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
5086                 BUG_ON(!path->locks[level]);
5087                 ret = btrfs_lookup_extent_info(trans, root,
5088                                                eb->start, eb->len,
5089                                                &wc->refs[level],
5090                                                &wc->flags[level]);
5091                 BUG_ON(ret);
5092                 BUG_ON(wc->refs[level] == 0);
5093         }
5094
5095         if (wc->stage == DROP_REFERENCE) {
5096                 if (wc->refs[level] > 1)
5097                         return 1;
5098
5099                 if (path->locks[level] && !wc->keep_locks) {
5100                         btrfs_tree_unlock(eb);
5101                         path->locks[level] = 0;
5102                 }
5103                 return 0;
5104         }
5105
5106         /* wc->stage == UPDATE_BACKREF */
5107         if (!(wc->flags[level] & flag)) {
5108                 BUG_ON(!path->locks[level]);
5109                 ret = btrfs_inc_ref(trans, root, eb, 1);
5110                 BUG_ON(ret);
5111                 ret = btrfs_dec_ref(trans, root, eb, 0);
5112                 BUG_ON(ret);
5113                 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
5114                                                   eb->len, flag, 0);
5115                 BUG_ON(ret);
5116                 wc->flags[level] |= flag;
5117         }
5118
5119         /*
5120          * the block is shared by multiple trees, so it's not good to
5121          * keep the tree lock
5122          */
5123         if (path->locks[level] && level > 0) {
5124                 btrfs_tree_unlock(eb);
5125                 path->locks[level] = 0;
5126         }
5127         return 0;
5128 }
5129
5130 /*
5131  * hepler to process tree block pointer.
5132  *
5133  * when wc->stage == DROP_REFERENCE, this function checks
5134  * reference count of the block pointed to. if the block
5135  * is shared and we need update back refs for the subtree
5136  * rooted at the block, this function changes wc->stage to
5137  * UPDATE_BACKREF. if the block is shared and there is no
5138  * need to update back, this function drops the reference
5139  * to the block.
5140  *
5141  * NOTE: return value 1 means we should stop walking down.
5142  */
5143 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
5144                                  struct btrfs_root *root,
5145                                  struct btrfs_path *path,
5146                                  struct walk_control *wc, int *lookup_info)
5147 {
5148         u64 bytenr;
5149         u64 generation;
5150         u64 parent;
5151         u32 blocksize;
5152         struct btrfs_key key;
5153         struct extent_buffer *next;
5154         int level = wc->level;
5155         int reada = 0;
5156         int ret = 0;
5157
5158         generation = btrfs_node_ptr_generation(path->nodes[level],
5159                                                path->slots[level]);
5160         /*
5161          * if the lower level block was created before the snapshot
5162          * was created, we know there is no need to update back refs
5163          * for the subtree
5164          */
5165         if (wc->stage == UPDATE_BACKREF &&
5166             generation <= root->root_key.offset) {
5167                 *lookup_info = 1;
5168                 return 1;
5169         }
5170
5171         bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
5172         blocksize = btrfs_level_size(root, level - 1);
5173
5174         next = btrfs_find_tree_block(root, bytenr, blocksize);
5175         if (!next) {
5176                 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
5177                 reada = 1;
5178         }
5179         btrfs_tree_lock(next);
5180         btrfs_set_lock_blocking(next);
5181
5182         ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
5183                                        &wc->refs[level - 1],
5184                                        &wc->flags[level - 1]);
5185         BUG_ON(ret);
5186         BUG_ON(wc->refs[level - 1] == 0);
5187         *lookup_info = 0;
5188
5189         if (wc->stage == DROP_REFERENCE) {
5190                 if (wc->refs[level - 1] > 1) {
5191                         if (level == 1 &&
5192                             (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5193                                 goto skip;
5194
5195                         if (!wc->update_ref ||
5196                             generation <= root->root_key.offset)
5197                                 goto skip;
5198
5199                         btrfs_node_key_to_cpu(path->nodes[level], &key,
5200                                               path->slots[level]);
5201                         ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
5202                         if (ret < 0)
5203                                 goto skip;
5204
5205                         wc->stage = UPDATE_BACKREF;
5206                         wc->shared_level = level - 1;
5207                 }
5208         } else {
5209                 if (level == 1 &&
5210                     (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
5211                         goto skip;
5212         }
5213
5214         if (!btrfs_buffer_uptodate(next, generation)) {
5215                 btrfs_tree_unlock(next);
5216                 free_extent_buffer(next);
5217                 next = NULL;
5218                 *lookup_info = 1;
5219         }
5220
5221         if (!next) {
5222                 if (reada && level == 1)
5223                         reada_walk_down(trans, root, wc, path);
5224                 next = read_tree_block(root, bytenr, blocksize, generation);
5225                 btrfs_tree_lock(next);
5226                 btrfs_set_lock_blocking(next);
5227         }
5228
5229         level--;
5230         BUG_ON(level != btrfs_header_level(next));
5231         path->nodes[level] = next;
5232         path->slots[level] = 0;
5233         path->locks[level] = 1;
5234         wc->level = level;
5235         if (wc->level == 1)
5236                 wc->reada_slot = 0;
5237         return 0;
5238 skip:
5239         wc->refs[level - 1] = 0;
5240         wc->flags[level - 1] = 0;
5241         if (wc->stage == DROP_REFERENCE) {
5242                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5243                         parent = path->nodes[level]->start;
5244                 } else {
5245                         BUG_ON(root->root_key.objectid !=
5246                                btrfs_header_owner(path->nodes[level]));
5247                         parent = 0;
5248                 }
5249
5250                 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
5251                                         root->root_key.objectid, level - 1, 0);
5252                 BUG_ON(ret);
5253         }
5254         btrfs_tree_unlock(next);
5255         free_extent_buffer(next);
5256         *lookup_info = 1;
5257         return 1;
5258 }
5259
5260 /*
5261  * hepler to process tree block while walking up the tree.
5262  *
5263  * when wc->stage == DROP_REFERENCE, this function drops
5264  * reference count on the block.
5265  *
5266  * when wc->stage == UPDATE_BACKREF, this function changes
5267  * wc->stage back to DROP_REFERENCE if we changed wc->stage
5268  * to UPDATE_BACKREF previously while processing the block.
5269  *
5270  * NOTE: return value 1 means we should stop walking up.
5271  */
5272 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5273                                  struct btrfs_root *root,
5274                                  struct btrfs_path *path,
5275                                  struct walk_control *wc)
5276 {
5277         int ret = 0;
5278         int level = wc->level;
5279         struct extent_buffer *eb = path->nodes[level];
5280         u64 parent = 0;
5281
5282         if (wc->stage == UPDATE_BACKREF) {
5283                 BUG_ON(wc->shared_level < level);
5284                 if (level < wc->shared_level)
5285                         goto out;
5286
5287                 ret = find_next_key(path, level + 1, &wc->update_progress);
5288                 if (ret > 0)
5289                         wc->update_ref = 0;
5290
5291                 wc->stage = DROP_REFERENCE;
5292                 wc->shared_level = -1;
5293                 path->slots[level] = 0;
5294
5295                 /*
5296                  * check reference count again if the block isn't locked.
5297                  * we should start walking down the tree again if reference
5298                  * count is one.
5299                  */
5300                 if (!path->locks[level]) {
5301                         BUG_ON(level == 0);
5302                         btrfs_tree_lock(eb);
5303                         btrfs_set_lock_blocking(eb);
5304                         path->locks[level] = 1;
5305
5306                         ret = btrfs_lookup_extent_info(trans, root,
5307                                                        eb->start, eb->len,
5308                                                        &wc->refs[level],
5309                                                        &wc->flags[level]);
5310                         BUG_ON(ret);
5311                         BUG_ON(wc->refs[level] == 0);
5312                         if (wc->refs[level] == 1) {
5313                                 btrfs_tree_unlock(eb);
5314                                 path->locks[level] = 0;
5315                                 return 1;
5316                         }
5317                 }
5318         }
5319
5320         /* wc->stage == DROP_REFERENCE */
5321         BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5322
5323         if (wc->refs[level] == 1) {
5324                 if (level == 0) {
5325                         if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5326                                 ret = btrfs_dec_ref(trans, root, eb, 1);
5327                         else
5328                                 ret = btrfs_dec_ref(trans, root, eb, 0);
5329                         BUG_ON(ret);
5330                 }
5331                 /* make block locked assertion in clean_tree_block happy */
5332                 if (!path->locks[level] &&
5333                     btrfs_header_generation(eb) == trans->transid) {
5334                         btrfs_tree_lock(eb);
5335                         btrfs_set_lock_blocking(eb);
5336                         path->locks[level] = 1;
5337                 }
5338                 clean_tree_block(trans, root, eb);
5339         }
5340
5341         if (eb == root->node) {
5342                 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5343                         parent = eb->start;
5344                 else
5345                         BUG_ON(root->root_key.objectid !=
5346                                btrfs_header_owner(eb));
5347         } else {
5348                 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5349                         parent = path->nodes[level + 1]->start;
5350                 else
5351                         BUG_ON(root->root_key.objectid !=
5352                                btrfs_header_owner(path->nodes[level + 1]));
5353         }
5354
5355         ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
5356                                 root->root_key.objectid, level, 0);
5357         BUG_ON(ret);
5358 out:
5359         wc->refs[level] = 0;
5360         wc->flags[level] = 0;
5361         return ret;
5362 }
5363
5364 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5365                                    struct btrfs_root *root,
5366                                    struct btrfs_path *path,
5367                                    struct walk_control *wc)
5368 {
5369         int level = wc->level;
5370         int lookup_info = 1;
5371         int ret;
5372
5373         while (level >= 0) {
5374                 if (path->slots[level] >=
5375                     btrfs_header_nritems(path->nodes[level]))
5376                         break;
5377
5378                 ret = walk_down_proc(trans, root, path, wc, lookup_info);
5379                 if (ret > 0)
5380                         break;
5381
5382                 if (level == 0)
5383                         break;
5384
5385                 ret = do_walk_down(trans, root, path, wc, &lookup_info);
5386                 if (ret > 0) {
5387                         path->slots[level]++;
5388                         continue;
5389                 }
5390                 level = wc->level;
5391         }
5392         return 0;
5393 }
5394
5395 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
5396                                  struct btrfs_root *root,
5397                                  struct btrfs_path *path,
5398                                  struct walk_control *wc, int max_level)
5399 {
5400         int level = wc->level;
5401         int ret;
5402
5403         path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5404         while (level < max_level && path->nodes[level]) {
5405                 wc->level = level;
5406                 if (path->slots[level] + 1 <
5407                     btrfs_header_nritems(path->nodes[level])) {
5408                         path->slots[level]++;
5409                         return 0;
5410                 } else {
5411                         ret = walk_up_proc(trans, root, path, wc);
5412                         if (ret > 0)
5413                                 return 0;
5414
5415                         if (path->locks[level]) {
5416                                 btrfs_tree_unlock(path->nodes[level]);
5417                                 path->locks[level] = 0;
5418                         }
5419                         free_extent_buffer(path->nodes[level]);
5420                         path->nodes[level] = NULL;
5421                         level++;
5422                 }
5423         }
5424         return 1;
5425 }
5426
5427 /*
5428  * drop a subvolume tree.
5429  *
5430  * this function traverses the tree freeing any blocks that only
5431  * referenced by the tree.
5432  *
5433  * when a shared tree block is found. this function decreases its
5434  * reference count by one. if update_ref is true, this function
5435  * also make sure backrefs for the shared block and all lower level
5436  * blocks are properly updated.
5437  */
5438 int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
5439 {
5440         struct btrfs_path *path;
5441         struct btrfs_trans_handle *trans;
5442         struct btrfs_root *tree_root = root->fs_info->tree_root;
5443         struct btrfs_root_item *root_item = &root->root_item;
5444         struct walk_control *wc;
5445         struct btrfs_key key;
5446         int err = 0;
5447         int ret;
5448         int level;
5449
5450         path = btrfs_alloc_path();
5451         BUG_ON(!path);
5452
5453         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5454         BUG_ON(!wc);
5455
5456         trans = btrfs_start_transaction(tree_root, 1);
5457
5458         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5459                 level = btrfs_header_level(root->node);
5460                 path->nodes[level] = btrfs_lock_root_node(root);
5461                 btrfs_set_lock_blocking(path->nodes[level]);
5462                 path->slots[level] = 0;
5463                 path->locks[level] = 1;
5464                 memset(&wc->update_progress, 0,
5465                        sizeof(wc->update_progress));
5466         } else {
5467                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5468                 memcpy(&wc->update_progress, &key,
5469                        sizeof(wc->update_progress));
5470
5471                 level = root_item->drop_level;
5472                 BUG_ON(level == 0);
5473                 path->lowest_level = level;
5474                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5475                 path->lowest_level = 0;
5476                 if (ret < 0) {
5477                         err = ret;
5478                         goto out;
5479                 }
5480                 WARN_ON(ret > 0);
5481
5482                 /*
5483                  * unlock our path, this is safe because only this
5484                  * function is allowed to delete this snapshot
5485                  */
5486                 btrfs_unlock_up_safe(path, 0);
5487
5488                 level = btrfs_header_level(root->node);
5489                 while (1) {
5490                         btrfs_tree_lock(path->nodes[level]);
5491                         btrfs_set_lock_blocking(path->nodes[level]);
5492
5493                         ret = btrfs_lookup_extent_info(trans, root,
5494                                                 path->nodes[level]->start,
5495                                                 path->nodes[level]->len,
5496                                                 &wc->refs[level],
5497                                                 &wc->flags[level]);
5498                         BUG_ON(ret);
5499                         BUG_ON(wc->refs[level] == 0);
5500
5501                         if (level == root_item->drop_level)
5502                                 break;
5503
5504                         btrfs_tree_unlock(path->nodes[level]);
5505                         WARN_ON(wc->refs[level] != 1);
5506                         level--;
5507                 }
5508         }
5509
5510         wc->level = level;
5511         wc->shared_level = -1;
5512         wc->stage = DROP_REFERENCE;
5513         wc->update_ref = update_ref;
5514         wc->keep_locks = 0;
5515         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5516
5517         while (1) {
5518                 ret = walk_down_tree(trans, root, path, wc);
5519                 if (ret < 0) {
5520                         err = ret;
5521                         break;
5522                 }
5523
5524                 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5525                 if (ret < 0) {
5526                         err = ret;
5527                         break;
5528                 }
5529
5530                 if (ret > 0) {
5531                         BUG_ON(wc->stage != DROP_REFERENCE);
5532                         break;
5533                 }
5534
5535                 if (wc->stage == DROP_REFERENCE) {
5536                         level = wc->level;
5537                         btrfs_node_key(path->nodes[level],
5538                                        &root_item->drop_progress,
5539                                        path->slots[level]);
5540                         root_item->drop_level = level;
5541                 }
5542
5543                 BUG_ON(wc->level == 0);
5544                 if (trans->transaction->in_commit ||
5545                     trans->transaction->delayed_refs.flushing) {
5546                         ret = btrfs_update_root(trans, tree_root,
5547                                                 &root->root_key,
5548                                                 root_item);
5549                         BUG_ON(ret);
5550
5551                         btrfs_end_transaction(trans, tree_root);
5552                         trans = btrfs_start_transaction(tree_root, 1);
5553                 } else {
5554                         unsigned long update;
5555                         update = trans->delayed_ref_updates;
5556                         trans->delayed_ref_updates = 0;
5557                         if (update)
5558                                 btrfs_run_delayed_refs(trans, tree_root,
5559                                                        update);
5560                 }
5561         }
5562         btrfs_release_path(root, path);
5563         BUG_ON(err);
5564
5565         ret = btrfs_del_root(trans, tree_root, &root->root_key);
5566         BUG_ON(ret);
5567
5568         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5569                 ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
5570                                            NULL, NULL);
5571                 BUG_ON(ret < 0);
5572                 if (ret > 0) {
5573                         ret = btrfs_del_orphan_item(trans, tree_root,
5574                                                     root->root_key.objectid);
5575                         BUG_ON(ret);
5576                 }
5577         }
5578
5579         if (root->in_radix) {
5580                 btrfs_free_fs_root(tree_root->fs_info, root);
5581         } else {
5582                 free_extent_buffer(root->node);
5583                 free_extent_buffer(root->commit_root);
5584                 kfree(root);
5585         }
5586 out:
5587         btrfs_end_transaction(trans, tree_root);
5588         kfree(wc);
5589         btrfs_free_path(path);
5590         return err;
5591 }
5592
5593 /*
5594  * drop subtree rooted at tree block 'node'.
5595  *
5596  * NOTE: this function will unlock and release tree block 'node'
5597  */
5598 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5599                         struct btrfs_root *root,
5600                         struct extent_buffer *node,
5601                         struct extent_buffer *parent)
5602 {
5603         struct btrfs_path *path;
5604         struct walk_control *wc;
5605         int level;
5606         int parent_level;
5607         int ret = 0;
5608         int wret;
5609
5610         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5611
5612         path = btrfs_alloc_path();
5613         BUG_ON(!path);
5614
5615         wc = kzalloc(sizeof(*wc), GFP_NOFS);
5616         BUG_ON(!wc);
5617
5618         btrfs_assert_tree_locked(parent);
5619         parent_level = btrfs_header_level(parent);
5620         extent_buffer_get(parent);
5621         path->nodes[parent_level] = parent;
5622         path->slots[parent_level] = btrfs_header_nritems(parent);
5623
5624         btrfs_assert_tree_locked(node);
5625         level = btrfs_header_level(node);
5626         path->nodes[level] = node;
5627         path->slots[level] = 0;
5628         path->locks[level] = 1;
5629
5630         wc->refs[parent_level] = 1;
5631         wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5632         wc->level = level;
5633         wc->shared_level = -1;
5634         wc->stage = DROP_REFERENCE;
5635         wc->update_ref = 0;
5636         wc->keep_locks = 1;
5637         wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
5638
5639         while (1) {
5640                 wret = walk_down_tree(trans, root, path, wc);
5641                 if (wret < 0) {
5642                         ret = wret;
5643                         break;
5644                 }
5645
5646                 wret = walk_up_tree(trans, root, path, wc, parent_level);
5647                 if (wret < 0)
5648                         ret = wret;
5649                 if (wret != 0)
5650                         break;
5651         }
5652
5653         kfree(wc);
5654         btrfs_free_path(path);
5655         return ret;
5656 }
5657
5658 #if 0
5659 static unsigned long calc_ra(unsigned long start, unsigned long last,
5660                              unsigned long nr)
5661 {
5662         return min(last, start + nr - 1);
5663 }
5664
5665 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
5666                                          u64 len)
5667 {
5668         u64 page_start;
5669         u64 page_end;
5670         unsigned long first_index;
5671         unsigned long last_index;
5672         unsigned long i;
5673         struct page *page;
5674         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5675         struct file_ra_state *ra;
5676         struct btrfs_ordered_extent *ordered;
5677         unsigned int total_read = 0;
5678         unsigned int total_dirty = 0;
5679         int ret = 0;
5680
5681         ra = kzalloc(sizeof(*ra), GFP_NOFS);
5682
5683         mutex_lock(&inode->i_mutex);
5684         first_index = start >> PAGE_CACHE_SHIFT;
5685         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
5686
5687         /* make sure the dirty trick played by the caller work */
5688         ret = invalidate_inode_pages2_range(inode->i_mapping,
5689                                             first_index, last_index);
5690         if (ret)
5691                 goto out_unlock;
5692
5693         file_ra_state_init(ra, inode->i_mapping);
5694
5695         for (i = first_index ; i <= last_index; i++) {
5696                 if (total_read % ra->ra_pages == 0) {
5697                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
5698                                        calc_ra(i, last_index, ra->ra_pages));
5699                 }
5700                 total_read++;
5701 again:
5702                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
5703                         BUG_ON(1);
5704                 page = grab_cache_page(inode->i_mapping, i);
5705                 if (!page) {
5706                         ret = -ENOMEM;
5707                         goto out_unlock;
5708                 }
5709                 if (!PageUptodate(page)) {
5710                         btrfs_readpage(NULL, page);
5711                         lock_page(page);
5712                         if (!PageUptodate(page)) {
5713                                 unlock_page(page);
5714                                 page_cache_release(page);
5715                                 ret = -EIO;
5716                                 goto out_unlock;
5717                         }
5718                 }
5719                 wait_on_page_writeback(page);
5720
5721                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
5722                 page_end = page_start + PAGE_CACHE_SIZE - 1;
5723                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5724
5725                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5726                 if (ordered) {
5727                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5728                         unlock_page(page);
5729                         page_cache_release(page);
5730                         btrfs_start_ordered_extent(inode, ordered, 1);
5731                         btrfs_put_ordered_extent(ordered);
5732                         goto again;
5733                 }
5734                 set_page_extent_mapped(page);
5735
5736                 if (i == first_index)
5737                         set_extent_bits(io_tree, page_start, page_end,
5738                                         EXTENT_BOUNDARY, GFP_NOFS);
5739                 btrfs_set_extent_delalloc(inode, page_start, page_end);
5740
5741                 set_page_dirty(page);
5742                 total_dirty++;
5743
5744                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5745                 unlock_page(page);
5746                 page_cache_release(page);
5747         }
5748
5749 out_unlock:
5750         kfree(ra);
5751         mutex_unlock(&inode->i_mutex);
5752         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
5753         return ret;
5754 }
5755
5756 static noinline int relocate_data_extent(struct inode *reloc_inode,
5757                                          struct btrfs_key *extent_key,
5758                                          u64 offset)
5759 {
5760         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
5761         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
5762         struct extent_map *em;
5763         u64 start = extent_key->objectid - offset;
5764         u64 end = start + extent_key->offset - 1;
5765
5766         em = alloc_extent_map(GFP_NOFS);
5767         BUG_ON(!em || IS_ERR(em));
5768
5769         em->start = start;
5770         em->len = extent_key->offset;
5771         em->block_len = extent_key->offset;
5772         em->block_start = extent_key->objectid;
5773         em->bdev = root->fs_info->fs_devices->latest_bdev;
5774         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5775
5776         /* setup extent map to cheat btrfs_readpage */
5777         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5778         while (1) {
5779                 int ret;
5780                 write_lock(&em_tree->lock);
5781                 ret = add_extent_mapping(em_tree, em);
5782                 write_unlock(&em_tree->lock);
5783                 if (ret != -EEXIST) {
5784                         free_extent_map(em);
5785                         break;
5786                 }
5787                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
5788         }
5789         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
5790
5791         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
5792 }
5793
5794 struct btrfs_ref_path {
5795         u64 extent_start;
5796         u64 nodes[BTRFS_MAX_LEVEL];
5797         u64 root_objectid;
5798         u64 root_generation;
5799         u64 owner_objectid;
5800         u32 num_refs;
5801         int lowest_level;
5802         int current_level;
5803         int shared_level;
5804
5805         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
5806         u64 new_nodes[BTRFS_MAX_LEVEL];
5807 };
5808
5809 struct disk_extent {
5810         u64 ram_bytes;
5811         u64 disk_bytenr;
5812         u64 disk_num_bytes;
5813         u64 offset;
5814         u64 num_bytes;
5815         u8 compression;
5816         u8 encryption;
5817         u16 other_encoding;
5818 };
5819
5820 static int is_cowonly_root(u64 root_objectid)
5821 {
5822         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
5823             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5824             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
5825             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
5826             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5827             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
5828                 return 1;
5829         return 0;
5830 }
5831
5832 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
5833                                     struct btrfs_root *extent_root,
5834                                     struct btrfs_ref_path *ref_path,
5835                                     int first_time)
5836 {
5837         struct extent_buffer *leaf;
5838         struct btrfs_path *path;
5839         struct btrfs_extent_ref *ref;
5840         struct btrfs_key key;
5841         struct btrfs_key found_key;
5842         u64 bytenr;
5843         u32 nritems;
5844         int level;
5845         int ret = 1;
5846
5847         path = btrfs_alloc_path();
5848         if (!path)
5849                 return -ENOMEM;
5850
5851         if (first_time) {
5852                 ref_path->lowest_level = -1;
5853                 ref_path->current_level = -1;
5854                 ref_path->shared_level = -1;
5855                 goto walk_up;
5856         }
5857 walk_down:
5858         level = ref_path->current_level - 1;
5859         while (level >= -1) {
5860                 u64 parent;
5861                 if (level < ref_path->lowest_level)
5862                         break;
5863
5864                 if (level >= 0)
5865                         bytenr = ref_path->nodes[level];
5866                 else
5867                         bytenr = ref_path->extent_start;
5868                 BUG_ON(bytenr == 0);
5869
5870                 parent = ref_path->nodes[level + 1];
5871                 ref_path->nodes[level + 1] = 0;
5872                 ref_path->current_level = level;
5873                 BUG_ON(parent == 0);
5874
5875                 key.objectid = bytenr;
5876                 key.offset = parent + 1;
5877                 key.type = BTRFS_EXTENT_REF_KEY;
5878
5879                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5880                 if (ret < 0)
5881                         goto out;
5882                 BUG_ON(ret == 0);
5883
5884                 leaf = path->nodes[0];
5885                 nritems = btrfs_header_nritems(leaf);
5886                 if (path->slots[0] >= nritems) {
5887                         ret = btrfs_next_leaf(extent_root, path);
5888                         if (ret < 0)
5889                                 goto out;
5890                         if (ret > 0)
5891                                 goto next;
5892                         leaf = path->nodes[0];
5893                 }
5894
5895                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5896                 if (found_key.objectid == bytenr &&
5897                     found_key.type == BTRFS_EXTENT_REF_KEY) {
5898                         if (level < ref_path->shared_level)
5899                                 ref_path->shared_level = level;
5900                         goto found;
5901                 }
5902 next:
5903                 level--;
5904                 btrfs_release_path(extent_root, path);
5905                 cond_resched();
5906         }
5907         /* reached lowest level */
5908         ret = 1;
5909         goto out;
5910 walk_up:
5911         level = ref_path->current_level;
5912         while (level < BTRFS_MAX_LEVEL - 1) {
5913                 u64 ref_objectid;
5914
5915                 if (level >= 0)
5916                         bytenr = ref_path->nodes[level];
5917                 else
5918                         bytenr = ref_path->extent_start;
5919
5920                 BUG_ON(bytenr == 0);
5921
5922                 key.objectid = bytenr;
5923                 key.offset = 0;
5924                 key.type = BTRFS_EXTENT_REF_KEY;
5925
5926                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
5927                 if (ret < 0)
5928                         goto out;
5929
5930                 leaf = path->nodes[0];
5931                 nritems = btrfs_header_nritems(leaf);
5932                 if (path->slots[0] >= nritems) {
5933                         ret = btrfs_next_leaf(extent_root, path);
5934                         if (ret < 0)
5935                                 goto out;
5936                         if (ret > 0) {
5937                                 /* the extent was freed by someone */
5938                                 if (ref_path->lowest_level == level)
5939                                         goto out;
5940                                 btrfs_release_path(extent_root, path);
5941                                 goto walk_down;
5942                         }
5943                         leaf = path->nodes[0];
5944                 }
5945
5946                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5947                 if (found_key.objectid != bytenr ||
5948                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
5949                         /* the extent was freed by someone */
5950                         if (ref_path->lowest_level == level) {
5951                                 ret = 1;
5952                                 goto out;
5953                         }
5954                         btrfs_release_path(extent_root, path);
5955                         goto walk_down;
5956                 }
5957 found:
5958                 ref = btrfs_item_ptr(leaf, path->slots[0],
5959                                 struct btrfs_extent_ref);
5960                 ref_objectid = btrfs_ref_objectid(leaf, ref);
5961                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
5962                         if (first_time) {
5963                                 level = (int)ref_objectid;
5964                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
5965                                 ref_path->lowest_level = level;
5966                                 ref_path->current_level = level;
5967                                 ref_path->nodes[level] = bytenr;
5968                         } else {
5969                                 WARN_ON(ref_objectid != level);
5970                         }
5971                 } else {
5972                         WARN_ON(level != -1);
5973                 }
5974                 first_time = 0;
5975
5976                 if (ref_path->lowest_level == level) {
5977                         ref_path->owner_objectid = ref_objectid;
5978                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
5979                 }
5980
5981                 /*
5982                  * the block is tree root or the block isn't in reference
5983                  * counted tree.
5984                  */
5985                 if (found_key.objectid == found_key.offset ||
5986                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
5987                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
5988                         ref_path->root_generation =
5989                                 btrfs_ref_generation(leaf, ref);
5990                         if (level < 0) {
5991                                 /* special reference from the tree log */
5992                                 ref_path->nodes[0] = found_key.offset;
5993                                 ref_path->current_level = 0;
5994                         }
5995                         ret = 0;
5996                         goto out;
5997                 }
5998
5999                 level++;
6000                 BUG_ON(ref_path->nodes[level] != 0);
6001                 ref_path->nodes[level] = found_key.offset;
6002                 ref_path->current_level = level;
6003
6004                 /*
6005                  * the reference was created in the running transaction,
6006                  * no need to continue walking up.
6007                  */
6008                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
6009                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
6010                         ref_path->root_generation =
6011                                 btrfs_ref_generation(leaf, ref);
6012                         ret = 0;
6013                         goto out;
6014                 }
6015
6016                 btrfs_release_path(extent_root, path);
6017                 cond_resched();
6018         }
6019         /* reached max tree level, but no tree root found. */
6020         BUG();
6021 out:
6022         btrfs_free_path(path);
6023         return ret;
6024 }
6025
6026 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
6027                                 struct btrfs_root *extent_root,
6028                                 struct btrfs_ref_path *ref_path,
6029                                 u64 extent_start)
6030 {
6031         memset(ref_path, 0, sizeof(*ref_path));
6032         ref_path->extent_start = extent_start;
6033
6034         return __next_ref_path(trans, extent_root, ref_path, 1);
6035 }
6036
6037 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
6038                                struct btrfs_root *extent_root,
6039                                struct btrfs_ref_path *ref_path)
6040 {
6041         return __next_ref_path(trans, extent_root, ref_path, 0);
6042 }
6043
6044 static noinline int get_new_locations(struct inode *reloc_inode,
6045                                       struct btrfs_key *extent_key,
6046                                       u64 offset, int no_fragment,
6047                                       struct disk_extent **extents,
6048                                       int *nr_extents)
6049 {
6050         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
6051         struct btrfs_path *path;
6052         struct btrfs_file_extent_item *fi;
6053         struct extent_buffer *leaf;
6054         struct disk_extent *exts = *extents;
6055         struct btrfs_key found_key;
6056         u64 cur_pos;
6057         u64 last_byte;
6058         u32 nritems;
6059         int nr = 0;
6060         int max = *nr_extents;
6061         int ret;
6062
6063         WARN_ON(!no_fragment && *extents);
6064         if (!exts) {
6065                 max = 1;
6066                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
6067                 if (!exts)
6068                         return -ENOMEM;
6069         }
6070
6071         path = btrfs_alloc_path();
6072         BUG_ON(!path);
6073
6074         cur_pos = extent_key->objectid - offset;
6075         last_byte = extent_key->objectid + extent_key->offset;
6076         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
6077                                        cur_pos, 0);
6078         if (ret < 0)
6079                 goto out;
6080         if (ret > 0) {
6081                 ret = -ENOENT;
6082                 goto out;
6083         }
6084
6085         while (1) {
6086                 leaf = path->nodes[0];
6087                 nritems = btrfs_header_nritems(leaf);
6088                 if (path->slots[0] >= nritems) {
6089                         ret = btrfs_next_leaf(root, path);
6090                         if (ret < 0)
6091                                 goto out;
6092                         if (ret > 0)
6093                                 break;
6094                         leaf = path->nodes[0];
6095                 }
6096
6097                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6098                 if (found_key.offset != cur_pos ||
6099                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
6100                     found_key.objectid != reloc_inode->i_ino)
6101                         break;
6102
6103                 fi = btrfs_item_ptr(leaf, path->slots[0],
6104                                     struct btrfs_file_extent_item);
6105                 if (btrfs_file_extent_type(leaf, fi) !=
6106                     BTRFS_FILE_EXTENT_REG ||
6107                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6108                         break;
6109
6110                 if (nr == max) {
6111                         struct disk_extent *old = exts;
6112                         max *= 2;
6113                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
6114                         memcpy(exts, old, sizeof(*exts) * nr);
6115                         if (old != *extents)
6116                                 kfree(old);
6117                 }
6118
6119                 exts[nr].disk_bytenr =
6120                         btrfs_file_extent_disk_bytenr(leaf, fi);
6121                 exts[nr].disk_num_bytes =
6122                         btrfs_file_extent_disk_num_bytes(leaf, fi);
6123                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
6124                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6125                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6126                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
6127                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
6128                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
6129                                                                            fi);
6130                 BUG_ON(exts[nr].offset > 0);
6131                 BUG_ON(exts[nr].compression || exts[nr].encryption);
6132                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
6133
6134                 cur_pos += exts[nr].num_bytes;
6135                 nr++;
6136
6137                 if (cur_pos + offset >= last_byte)
6138                         break;
6139
6140                 if (no_fragment) {
6141                         ret = 1;
6142                         goto out;
6143                 }
6144                 path->slots[0]++;
6145         }
6146
6147         BUG_ON(cur_pos + offset > last_byte);
6148         if (cur_pos + offset < last_byte) {
6149                 ret = -ENOENT;
6150                 goto out;
6151         }
6152         ret = 0;
6153 out:
6154         btrfs_free_path(path);
6155         if (ret) {
6156                 if (exts != *extents)
6157                         kfree(exts);
6158         } else {
6159                 *extents = exts;
6160                 *nr_extents = nr;
6161         }
6162         return ret;
6163 }
6164
6165 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
6166                                         struct btrfs_root *root,
6167                                         struct btrfs_path *path,
6168                                         struct btrfs_key *extent_key,
6169                                         struct btrfs_key *leaf_key,
6170                                         struct btrfs_ref_path *ref_path,
6171                                         struct disk_extent *new_extents,
6172                                         int nr_extents)
6173 {
6174         struct extent_buffer *leaf;
6175         struct btrfs_file_extent_item *fi;
6176         struct inode *inode = NULL;
6177         struct btrfs_key key;
6178         u64 lock_start = 0;
6179         u64 lock_end = 0;
6180         u64 num_bytes;
6181         u64 ext_offset;
6182         u64 search_end = (u64)-1;
6183         u32 nritems;
6184         int nr_scaned = 0;
6185         int extent_locked = 0;
6186         int extent_type;
6187         int ret;
6188
6189         memcpy(&key, leaf_key, sizeof(key));
6190         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6191                 if (key.objectid < ref_path->owner_objectid ||
6192                     (key.objectid == ref_path->owner_objectid &&
6193                      key.type < BTRFS_EXTENT_DATA_KEY)) {
6194                         key.objectid = ref_path->owner_objectid;
6195                         key.type = BTRFS_EXTENT_DATA_KEY;
6196                         key.offset = 0;
6197                 }
6198         }
6199
6200         while (1) {
6201                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6202                 if (ret < 0)
6203                         goto out;
6204
6205                 leaf = path->nodes[0];
6206                 nritems = btrfs_header_nritems(leaf);
6207 next:
6208                 if (extent_locked && ret > 0) {
6209                         /*
6210                          * the file extent item was modified by someone
6211                          * before the extent got locked.
6212                          */
6213                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6214                                       lock_end, GFP_NOFS);
6215                         extent_locked = 0;
6216                 }
6217
6218                 if (path->slots[0] >= nritems) {
6219                         if (++nr_scaned > 2)
6220                                 break;
6221
6222                         BUG_ON(extent_locked);
6223                         ret = btrfs_next_leaf(root, path);
6224                         if (ret < 0)
6225                                 goto out;
6226                         if (ret > 0)
6227                                 break;
6228                         leaf = path->nodes[0];
6229                         nritems = btrfs_header_nritems(leaf);
6230                 }
6231
6232                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6233
6234                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
6235                         if ((key.objectid > ref_path->owner_objectid) ||
6236                             (key.objectid == ref_path->owner_objectid &&
6237                              key.type > BTRFS_EXTENT_DATA_KEY) ||
6238                             key.offset >= search_end)
6239                                 break;
6240                 }
6241
6242                 if (inode && key.objectid != inode->i_ino) {
6243                         BUG_ON(extent_locked);
6244                         btrfs_release_path(root, path);
6245                         mutex_unlock(&inode->i_mutex);
6246                         iput(inode);
6247                         inode = NULL;
6248                         continue;
6249                 }
6250
6251                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
6252                         path->slots[0]++;
6253                         ret = 1;
6254                         goto next;
6255                 }
6256                 fi = btrfs_item_ptr(leaf, path->slots[0],
6257                                     struct btrfs_file_extent_item);
6258                 extent_type = btrfs_file_extent_type(leaf, fi);
6259                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
6260                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
6261                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
6262                      extent_key->objectid)) {
6263                         path->slots[0]++;
6264                         ret = 1;
6265                         goto next;
6266                 }
6267
6268                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6269                 ext_offset = btrfs_file_extent_offset(leaf, fi);
6270
6271                 if (search_end == (u64)-1) {
6272                         search_end = key.offset - ext_offset +
6273                                 btrfs_file_extent_ram_bytes(leaf, fi);
6274                 }
6275
6276                 if (!extent_locked) {
6277                         lock_start = key.offset;
6278                         lock_end = lock_start + num_bytes - 1;
6279                 } else {
6280                         if (lock_start > key.offset ||
6281                             lock_end + 1 < key.offset + num_bytes) {
6282                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6283                                               lock_start, lock_end, GFP_NOFS);
6284                                 extent_locked = 0;
6285                         }
6286                 }
6287
6288                 if (!inode) {
6289                         btrfs_release_path(root, path);
6290
6291                         inode = btrfs_iget_locked(root->fs_info->sb,
6292                                                   key.objectid, root);
6293                         if (inode->i_state & I_NEW) {
6294                                 BTRFS_I(inode)->root = root;
6295                                 BTRFS_I(inode)->location.objectid =
6296                                         key.objectid;
6297                                 BTRFS_I(inode)->location.type =
6298                                         BTRFS_INODE_ITEM_KEY;
6299                                 BTRFS_I(inode)->location.offset = 0;
6300                                 btrfs_read_locked_inode(inode);
6301                                 unlock_new_inode(inode);
6302                         }
6303                         /*
6304                          * some code call btrfs_commit_transaction while
6305                          * holding the i_mutex, so we can't use mutex_lock
6306                          * here.
6307                          */
6308                         if (is_bad_inode(inode) ||
6309                             !mutex_trylock(&inode->i_mutex)) {
6310                                 iput(inode);
6311                                 inode = NULL;
6312                                 key.offset = (u64)-1;
6313                                 goto skip;
6314                         }
6315                 }
6316
6317                 if (!extent_locked) {
6318                         struct btrfs_ordered_extent *ordered;
6319
6320                         btrfs_release_path(root, path);
6321
6322                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6323                                     lock_end, GFP_NOFS);
6324                         ordered = btrfs_lookup_first_ordered_extent(inode,
6325                                                                     lock_end);
6326                         if (ordered &&
6327                             ordered->file_offset <= lock_end &&
6328                             ordered->file_offset + ordered->len > lock_start) {
6329                                 unlock_extent(&BTRFS_I(inode)->io_tree,
6330                                               lock_start, lock_end, GFP_NOFS);
6331                                 btrfs_start_ordered_extent(inode, ordered, 1);
6332                                 btrfs_put_ordered_extent(ordered);
6333                                 key.offset += num_bytes;
6334                                 goto skip;
6335                         }
6336                         if (ordered)
6337                                 btrfs_put_ordered_extent(ordered);
6338
6339                         extent_locked = 1;
6340                         continue;
6341                 }
6342
6343                 if (nr_extents == 1) {
6344                         /* update extent pointer in place */
6345                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
6346                                                 new_extents[0].disk_bytenr);
6347                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6348                                                 new_extents[0].disk_num_bytes);
6349                         btrfs_mark_buffer_dirty(leaf);
6350
6351                         btrfs_drop_extent_cache(inode, key.offset,
6352                                                 key.offset + num_bytes - 1, 0);
6353
6354                         ret = btrfs_inc_extent_ref(trans, root,
6355                                                 new_extents[0].disk_bytenr,
6356                                                 new_extents[0].disk_num_bytes,
6357                                                 leaf->start,
6358                                                 root->root_key.objectid,
6359                                                 trans->transid,
6360                                                 key.objectid);
6361                         BUG_ON(ret);
6362
6363                         ret = btrfs_free_extent(trans, root,
6364                                                 extent_key->objectid,
6365                                                 extent_key->offset,
6366                                                 leaf->start,
6367                                                 btrfs_header_owner(leaf),
6368                                                 btrfs_header_generation(leaf),
6369                                                 key.objectid, 0);
6370                         BUG_ON(ret);
6371
6372                         btrfs_release_path(root, path);
6373                         key.offset += num_bytes;
6374                 } else {
6375                         BUG_ON(1);
6376 #if 0
6377                         u64 alloc_hint;
6378                         u64 extent_len;
6379                         int i;
6380                         /*
6381                          * drop old extent pointer at first, then insert the
6382                          * new pointers one bye one
6383                          */
6384                         btrfs_release_path(root, path);
6385                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
6386                                                  key.offset + num_bytes,
6387                                                  key.offset, &alloc_hint);
6388                         BUG_ON(ret);
6389
6390                         for (i = 0; i < nr_extents; i++) {
6391                                 if (ext_offset >= new_extents[i].num_bytes) {
6392                                         ext_offset -= new_extents[i].num_bytes;
6393                                         continue;
6394                                 }
6395                                 extent_len = min(new_extents[i].num_bytes -
6396                                                  ext_offset, num_bytes);
6397
6398                                 ret = btrfs_insert_empty_item(trans, root,
6399                                                               path, &key,
6400                                                               sizeof(*fi));
6401                                 BUG_ON(ret);
6402
6403                                 leaf = path->nodes[0];
6404                                 fi = btrfs_item_ptr(leaf, path->slots[0],
6405                                                 struct btrfs_file_extent_item);
6406                                 btrfs_set_file_extent_generation(leaf, fi,
6407                                                         trans->transid);
6408                                 btrfs_set_file_extent_type(leaf, fi,
6409                                                         BTRFS_FILE_EXTENT_REG);
6410                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6411                                                 new_extents[i].disk_bytenr);
6412                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6413                                                 new_extents[i].disk_num_bytes);
6414                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
6415                                                 new_extents[i].ram_bytes);
6416
6417                                 btrfs_set_file_extent_compression(leaf, fi,
6418                                                 new_extents[i].compression);
6419                                 btrfs_set_file_extent_encryption(leaf, fi,
6420                                                 new_extents[i].encryption);
6421                                 btrfs_set_file_extent_other_encoding(leaf, fi,
6422                                                 new_extents[i].other_encoding);
6423
6424                                 btrfs_set_file_extent_num_bytes(leaf, fi,
6425                                                         extent_len);
6426                                 ext_offset += new_extents[i].offset;
6427                                 btrfs_set_file_extent_offset(leaf, fi,
6428                                                         ext_offset);
6429                                 btrfs_mark_buffer_dirty(leaf);
6430
6431                                 btrfs_drop_extent_cache(inode, key.offset,
6432                                                 key.offset + extent_len - 1, 0);
6433
6434                                 ret = btrfs_inc_extent_ref(trans, root,
6435                                                 new_extents[i].disk_bytenr,
6436                                                 new_extents[i].disk_num_bytes,
6437                                                 leaf->start,
6438                                                 root->root_key.objectid,
6439                                                 trans->transid, key.objectid);
6440                                 BUG_ON(ret);
6441                                 btrfs_release_path(root, path);
6442
6443                                 inode_add_bytes(inode, extent_len);
6444
6445                                 ext_offset = 0;
6446                                 num_bytes -= extent_len;
6447                                 key.offset += extent_len;
6448
6449                                 if (num_bytes == 0)
6450                                         break;
6451                         }
6452                         BUG_ON(i >= nr_extents);
6453 #endif
6454                 }
6455
6456                 if (extent_locked) {
6457                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6458                                       lock_end, GFP_NOFS);
6459                         extent_locked = 0;
6460                 }
6461 skip:
6462                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
6463                     key.offset >= search_end)
6464                         break;
6465
6466                 cond_resched();
6467         }
6468         ret = 0;
6469 out:
6470         btrfs_release_path(root, path);
6471         if (inode) {
6472                 mutex_unlock(&inode->i_mutex);
6473                 if (extent_locked) {
6474                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
6475                                       lock_end, GFP_NOFS);
6476                 }
6477                 iput(inode);
6478         }
6479         return ret;
6480 }
6481
6482 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
6483                                struct btrfs_root *root,
6484                                struct extent_buffer *buf, u64 orig_start)
6485 {
6486         int level;
6487         int ret;
6488
6489         BUG_ON(btrfs_header_generation(buf) != trans->transid);
6490         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
6491
6492         level = btrfs_header_level(buf);
6493         if (level == 0) {
6494                 struct btrfs_leaf_ref *ref;
6495                 struct btrfs_leaf_ref *orig_ref;
6496
6497                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
6498                 if (!orig_ref)
6499                         return -ENOENT;
6500
6501                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
6502                 if (!ref) {
6503                         btrfs_free_leaf_ref(root, orig_ref);
6504                         return -ENOMEM;
6505                 }
6506
6507                 ref->nritems = orig_ref->nritems;
6508                 memcpy(ref->extents, orig_ref->extents,
6509                         sizeof(ref->extents[0]) * ref->nritems);
6510
6511                 btrfs_free_leaf_ref(root, orig_ref);
6512
6513                 ref->root_gen = trans->transid;
6514                 ref->bytenr = buf->start;
6515                 ref->owner = btrfs_header_owner(buf);
6516                 ref->generation = btrfs_header_generation(buf);
6517
6518                 ret = btrfs_add_leaf_ref(root, ref, 0);
6519                 WARN_ON(ret);
6520                 btrfs_free_leaf_ref(root, ref);
6521         }
6522         return 0;
6523 }
6524
6525 static noinline int invalidate_extent_cache(struct btrfs_root *root,
6526                                         struct extent_buffer *leaf,
6527                                         struct btrfs_block_group_cache *group,
6528                                         struct btrfs_root *target_root)
6529 {
6530         struct btrfs_key key;
6531         struct inode *inode = NULL;
6532         struct btrfs_file_extent_item *fi;
6533         u64 num_bytes;
6534         u64 skip_objectid = 0;
6535         u32 nritems;
6536         u32 i;
6537
6538         nritems = btrfs_header_nritems(leaf);
6539         for (i = 0; i < nritems; i++) {
6540                 btrfs_item_key_to_cpu(leaf, &key, i);
6541                 if (key.objectid == skip_objectid ||
6542                     key.type != BTRFS_EXTENT_DATA_KEY)
6543                         continue;
6544                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6545                 if (btrfs_file_extent_type(leaf, fi) ==
6546                     BTRFS_FILE_EXTENT_INLINE)
6547                         continue;
6548                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
6549                         continue;
6550                 if (!inode || inode->i_ino != key.objectid) {
6551                         iput(inode);
6552                         inode = btrfs_ilookup(target_root->fs_info->sb,
6553                                               key.objectid, target_root, 1);
6554                 }
6555                 if (!inode) {
6556                         skip_objectid = key.objectid;
6557                         continue;
6558                 }
6559                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
6560
6561                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6562                             key.offset + num_bytes - 1, GFP_NOFS);
6563                 btrfs_drop_extent_cache(inode, key.offset,
6564                                         key.offset + num_bytes - 1, 1);
6565                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
6566                               key.offset + num_bytes - 1, GFP_NOFS);
6567                 cond_resched();
6568         }
6569         iput(inode);
6570         return 0;
6571 }
6572
6573 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
6574                                         struct btrfs_root *root,
6575                                         struct extent_buffer *leaf,
6576                                         struct btrfs_block_group_cache *group,
6577                                         struct inode *reloc_inode)
6578 {
6579         struct btrfs_key key;
6580         struct btrfs_key extent_key;
6581         struct btrfs_file_extent_item *fi;
6582         struct btrfs_leaf_ref *ref;
6583         struct disk_extent *new_extent;
6584         u64 bytenr;
6585         u64 num_bytes;
6586         u32 nritems;
6587         u32 i;
6588         int ext_index;
6589         int nr_extent;
6590         int ret;
6591
6592         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
6593         BUG_ON(!new_extent);
6594
6595         ref = btrfs_lookup_leaf_ref(root, leaf->start);
6596         BUG_ON(!ref);
6597
6598         ext_index = -1;
6599         nritems = btrfs_header_nritems(leaf);
6600         for (i = 0; i < nritems; i++) {
6601                 btrfs_item_key_to_cpu(leaf, &key, i);
6602                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
6603                         continue;
6604                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
6605                 if (btrfs_file_extent_type(leaf, fi) ==
6606                     BTRFS_FILE_EXTENT_INLINE)
6607                         continue;
6608                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6609                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6610                 if (bytenr == 0)
6611                         continue;
6612
6613                 ext_index++;
6614                 if (bytenr >= group->key.objectid + group->key.offset ||
6615                     bytenr + num_bytes <= group->key.objectid)
6616                         continue;
6617
6618                 extent_key.objectid = bytenr;
6619                 extent_key.offset = num_bytes;
6620                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
6621                 nr_extent = 1;
6622                 ret = get_new_locations(reloc_inode, &extent_key,
6623                                         group->key.objectid, 1,
6624                                         &new_extent, &nr_extent);
6625                 if (ret > 0)
6626                         continue;
6627                 BUG_ON(ret < 0);
6628
6629                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
6630                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
6631                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
6632                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
6633
6634                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
6635                                                 new_extent->disk_bytenr);
6636                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
6637                                                 new_extent->disk_num_bytes);
6638                 btrfs_mark_buffer_dirty(leaf);
6639
6640                 ret = btrfs_inc_extent_ref(trans, root,
6641                                         new_extent->disk_bytenr,
6642                                         new_extent->disk_num_bytes,
6643                                         leaf->start,
6644                                         root->root_key.objectid,
6645                                         trans->transid, key.objectid);
6646                 BUG_ON(ret);
6647
6648                 ret = btrfs_free_extent(trans, root,
6649                                         bytenr, num_bytes, leaf->start,
6650                                         btrfs_header_owner(leaf),
6651                                         btrfs_header_generation(leaf),
6652                                         key.objectid, 0);
6653                 BUG_ON(ret);
6654                 cond_resched();
6655         }
6656         kfree(new_extent);
6657         BUG_ON(ext_index + 1 != ref->nritems);
6658         btrfs_free_leaf_ref(root, ref);
6659         return 0;
6660 }
6661
6662 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
6663                           struct btrfs_root *root)
6664 {
6665         struct btrfs_root *reloc_root;
6666         int ret;
6667
6668         if (root->reloc_root) {
6669                 reloc_root = root->reloc_root;
6670                 root->reloc_root = NULL;
6671                 list_add(&reloc_root->dead_list,
6672                          &root->fs_info->dead_reloc_roots);
6673
6674                 btrfs_set_root_bytenr(&reloc_root->root_item,
6675                                       reloc_root->node->start);
6676                 btrfs_set_root_level(&root->root_item,
6677                                      btrfs_header_level(reloc_root->node));
6678                 memset(&reloc_root->root_item.drop_progress, 0,
6679                         sizeof(struct btrfs_disk_key));
6680                 reloc_root->root_item.drop_level = 0;
6681
6682                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6683                                         &reloc_root->root_key,
6684                                         &reloc_root->root_item);
6685                 BUG_ON(ret);
6686         }
6687         return 0;
6688 }
6689
6690 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
6691 {
6692         struct btrfs_trans_handle *trans;
6693         struct btrfs_root *reloc_root;
6694         struct btrfs_root *prev_root = NULL;
6695         struct list_head dead_roots;
6696         int ret;
6697         unsigned long nr;
6698
6699         INIT_LIST_HEAD(&dead_roots);
6700         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
6701
6702         while (!list_empty(&dead_roots)) {
6703                 reloc_root = list_entry(dead_roots.prev,
6704                                         struct btrfs_root, dead_list);
6705                 list_del_init(&reloc_root->dead_list);
6706
6707                 BUG_ON(reloc_root->commit_root != NULL);
6708                 while (1) {
6709                         trans = btrfs_join_transaction(root, 1);
6710                         BUG_ON(!trans);
6711
6712                         mutex_lock(&root->fs_info->drop_mutex);
6713                         ret = btrfs_drop_snapshot(trans, reloc_root);
6714                         if (ret != -EAGAIN)
6715                                 break;
6716                         mutex_unlock(&root->fs_info->drop_mutex);
6717
6718                         nr = trans->blocks_used;
6719                         ret = btrfs_end_transaction(trans, root);
6720                         BUG_ON(ret);
6721                         btrfs_btree_balance_dirty(root, nr);
6722                 }
6723
6724                 free_extent_buffer(reloc_root->node);
6725
6726                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
6727                                      &reloc_root->root_key);
6728                 BUG_ON(ret);
6729                 mutex_unlock(&root->fs_info->drop_mutex);
6730
6731                 nr = trans->blocks_used;
6732                 ret = btrfs_end_transaction(trans, root);
6733                 BUG_ON(ret);
6734                 btrfs_btree_balance_dirty(root, nr);
6735
6736                 kfree(prev_root);
6737                 prev_root = reloc_root;
6738         }
6739         if (prev_root) {
6740                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
6741                 kfree(prev_root);
6742         }
6743         return 0;
6744 }
6745
6746 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
6747 {
6748         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
6749         return 0;
6750 }
6751
6752 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
6753 {
6754         struct btrfs_root *reloc_root;
6755         struct btrfs_trans_handle *trans;
6756         struct btrfs_key location;
6757         int found;
6758         int ret;
6759
6760         mutex_lock(&root->fs_info->tree_reloc_mutex);
6761         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
6762         BUG_ON(ret);
6763         found = !list_empty(&root->fs_info->dead_reloc_roots);
6764         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6765
6766         if (found) {
6767                 trans = btrfs_start_transaction(root, 1);
6768                 BUG_ON(!trans);
6769                 ret = btrfs_commit_transaction(trans, root);
6770                 BUG_ON(ret);
6771         }
6772
6773         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6774         location.offset = (u64)-1;
6775         location.type = BTRFS_ROOT_ITEM_KEY;
6776
6777         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
6778         BUG_ON(!reloc_root);
6779         btrfs_orphan_cleanup(reloc_root);
6780         return 0;
6781 }
6782
6783 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
6784                                     struct btrfs_root *root)
6785 {
6786         struct btrfs_root *reloc_root;
6787         struct extent_buffer *eb;
6788         struct btrfs_root_item *root_item;
6789         struct btrfs_key root_key;
6790         int ret;
6791
6792         BUG_ON(!root->ref_cows);
6793         if (root->reloc_root)
6794                 return 0;
6795
6796         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
6797         BUG_ON(!root_item);
6798
6799         ret = btrfs_copy_root(trans, root, root->commit_root,
6800                               &eb, BTRFS_TREE_RELOC_OBJECTID);
6801         BUG_ON(ret);
6802
6803         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6804         root_key.offset = root->root_key.objectid;
6805         root_key.type = BTRFS_ROOT_ITEM_KEY;
6806
6807         memcpy(root_item, &root->root_item, sizeof(root_item));
6808         btrfs_set_root_refs(root_item, 0);
6809         btrfs_set_root_bytenr(root_item, eb->start);
6810         btrfs_set_root_level(root_item, btrfs_header_level(eb));
6811         btrfs_set_root_generation(root_item, trans->transid);
6812
6813         btrfs_tree_unlock(eb);
6814         free_extent_buffer(eb);
6815
6816         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
6817                                 &root_key, root_item);
6818         BUG_ON(ret);
6819         kfree(root_item);
6820
6821         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
6822                                                  &root_key);
6823         BUG_ON(!reloc_root);
6824         reloc_root->last_trans = trans->transid;
6825         reloc_root->commit_root = NULL;
6826         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
6827
6828         root->reloc_root = reloc_root;
6829         return 0;
6830 }
6831
6832 /*
6833  * Core function of space balance.
6834  *
6835  * The idea is using reloc trees to relocate tree blocks in reference
6836  * counted roots. There is one reloc tree for each subvol, and all
6837  * reloc trees share same root key objectid. Reloc trees are snapshots
6838  * of the latest committed roots of subvols (root->commit_root).
6839  *
6840  * To relocate a tree block referenced by a subvol, there are two steps.
6841  * COW the block through subvol's reloc tree, then update block pointer
6842  * in the subvol to point to the new block. Since all reloc trees share
6843  * same root key objectid, doing special handing for tree blocks owned
6844  * by them is easy. Once a tree block has been COWed in one reloc tree,
6845  * we can use the resulting new block directly when the same block is
6846  * required to COW again through other reloc trees. By this way, relocated
6847  * tree blocks are shared between reloc trees, so they are also shared
6848  * between subvols.
6849  */
6850 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
6851                                       struct btrfs_root *root,
6852                                       struct btrfs_path *path,
6853                                       struct btrfs_key *first_key,
6854                                       struct btrfs_ref_path *ref_path,
6855                                       struct btrfs_block_group_cache *group,
6856                                       struct inode *reloc_inode)
6857 {
6858         struct btrfs_root *reloc_root;
6859         struct extent_buffer *eb = NULL;
6860         struct btrfs_key *keys;
6861         u64 *nodes;
6862         int level;
6863         int shared_level;
6864         int lowest_level = 0;
6865         int ret;
6866
6867         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
6868                 lowest_level = ref_path->owner_objectid;
6869
6870         if (!root->ref_cows) {
6871                 path->lowest_level = lowest_level;
6872                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
6873                 BUG_ON(ret < 0);
6874                 path->lowest_level = 0;
6875                 btrfs_release_path(root, path);
6876                 return 0;
6877         }
6878
6879         mutex_lock(&root->fs_info->tree_reloc_mutex);
6880         ret = init_reloc_tree(trans, root);
6881         BUG_ON(ret);
6882         reloc_root = root->reloc_root;
6883
6884         shared_level = ref_path->shared_level;
6885         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
6886
6887         keys = ref_path->node_keys;
6888         nodes = ref_path->new_nodes;
6889         memset(&keys[shared_level + 1], 0,
6890                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
6891         memset(&nodes[shared_level + 1], 0,
6892                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
6893
6894         if (nodes[lowest_level] == 0) {
6895                 path->lowest_level = lowest_level;
6896                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6897                                         0, 1);
6898                 BUG_ON(ret);
6899                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
6900                         eb = path->nodes[level];
6901                         if (!eb || eb == reloc_root->node)
6902                                 break;
6903                         nodes[level] = eb->start;
6904                         if (level == 0)
6905                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
6906                         else
6907                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
6908                 }
6909                 if (nodes[0] &&
6910                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6911                         eb = path->nodes[0];
6912                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
6913                                                       group, reloc_inode);
6914                         BUG_ON(ret);
6915                 }
6916                 btrfs_release_path(reloc_root, path);
6917         } else {
6918                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
6919                                        lowest_level);
6920                 BUG_ON(ret);
6921         }
6922
6923         /*
6924          * replace tree blocks in the fs tree with tree blocks in
6925          * the reloc tree.
6926          */
6927         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
6928         BUG_ON(ret < 0);
6929
6930         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
6931                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
6932                                         0, 0);
6933                 BUG_ON(ret);
6934                 extent_buffer_get(path->nodes[0]);
6935                 eb = path->nodes[0];
6936                 btrfs_release_path(reloc_root, path);
6937                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
6938                 BUG_ON(ret);
6939                 free_extent_buffer(eb);
6940         }
6941
6942         mutex_unlock(&root->fs_info->tree_reloc_mutex);
6943         path->lowest_level = 0;
6944         return 0;
6945 }
6946
6947 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
6948                                         struct btrfs_root *root,
6949                                         struct btrfs_path *path,
6950                                         struct btrfs_key *first_key,
6951                                         struct btrfs_ref_path *ref_path)
6952 {
6953         int ret;
6954
6955         ret = relocate_one_path(trans, root, path, first_key,
6956                                 ref_path, NULL, NULL);
6957         BUG_ON(ret);
6958
6959         return 0;
6960 }
6961
6962 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
6963                                     struct btrfs_root *extent_root,
6964                                     struct btrfs_path *path,
6965                                     struct btrfs_key *extent_key)
6966 {
6967         int ret;
6968
6969         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
6970         if (ret)
6971                 goto out;
6972         ret = btrfs_del_item(trans, extent_root, path);
6973 out:
6974         btrfs_release_path(extent_root, path);
6975         return ret;
6976 }
6977
6978 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
6979                                                 struct btrfs_ref_path *ref_path)
6980 {
6981         struct btrfs_key root_key;
6982
6983         root_key.objectid = ref_path->root_objectid;
6984         root_key.type = BTRFS_ROOT_ITEM_KEY;
6985         if (is_cowonly_root(ref_path->root_objectid))
6986                 root_key.offset = 0;
6987         else
6988                 root_key.offset = (u64)-1;
6989
6990         return btrfs_read_fs_root_no_name(fs_info, &root_key);
6991 }
6992
6993 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
6994                                         struct btrfs_path *path,
6995                                         struct btrfs_key *extent_key,
6996                                         struct btrfs_block_group_cache *group,
6997                                         struct inode *reloc_inode, int pass)
6998 {
6999         struct btrfs_trans_handle *trans;
7000         struct btrfs_root *found_root;
7001         struct btrfs_ref_path *ref_path = NULL;
7002         struct disk_extent *new_extents = NULL;
7003         int nr_extents = 0;
7004         int loops;
7005         int ret;
7006         int level;
7007         struct btrfs_key first_key;
7008         u64 prev_block = 0;
7009
7010
7011         trans = btrfs_start_transaction(extent_root, 1);
7012         BUG_ON(!trans);
7013
7014         if (extent_key->objectid == 0) {
7015                 ret = del_extent_zero(trans, extent_root, path, extent_key);
7016                 goto out;
7017         }
7018
7019         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
7020         if (!ref_path) {
7021                 ret = -ENOMEM;
7022                 goto out;
7023         }
7024
7025         for (loops = 0; ; loops++) {
7026                 if (loops == 0) {
7027                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
7028                                                    extent_key->objectid);
7029                 } else {
7030                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
7031                 }
7032                 if (ret < 0)
7033                         goto out;
7034                 if (ret > 0)
7035                         break;
7036
7037                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
7038                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
7039                         continue;
7040
7041                 found_root = read_ref_root(extent_root->fs_info, ref_path);
7042                 BUG_ON(!found_root);
7043                 /*
7044                  * for reference counted tree, only process reference paths
7045                  * rooted at the latest committed root.
7046                  */
7047                 if (found_root->ref_cows &&
7048                     ref_path->root_generation != found_root->root_key.offset)
7049                         continue;
7050
7051                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7052                         if (pass == 0) {
7053                                 /*
7054                                  * copy data extents to new locations
7055                                  */
7056                                 u64 group_start = group->key.objectid;
7057                                 ret = relocate_data_extent(reloc_inode,
7058                                                            extent_key,
7059                                                            group_start);
7060                                 if (ret < 0)
7061                                         goto out;
7062                                 break;
7063                         }
7064                         level = 0;
7065                 } else {
7066                         level = ref_path->owner_objectid;
7067                 }
7068
7069                 if (prev_block != ref_path->nodes[level]) {
7070                         struct extent_buffer *eb;
7071                         u64 block_start = ref_path->nodes[level];
7072                         u64 block_size = btrfs_level_size(found_root, level);
7073
7074                         eb = read_tree_block(found_root, block_start,
7075                                              block_size, 0);
7076                         btrfs_tree_lock(eb);
7077                         BUG_ON(level != btrfs_header_level(eb));
7078
7079                         if (level == 0)
7080                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
7081                         else
7082                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
7083
7084                         btrfs_tree_unlock(eb);
7085                         free_extent_buffer(eb);
7086                         prev_block = block_start;
7087                 }
7088
7089                 mutex_lock(&extent_root->fs_info->trans_mutex);
7090                 btrfs_record_root_in_trans(found_root);
7091                 mutex_unlock(&extent_root->fs_info->trans_mutex);
7092                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
7093                         /*
7094                          * try to update data extent references while
7095                          * keeping metadata shared between snapshots.
7096                          */
7097                         if (pass == 1) {
7098                                 ret = relocate_one_path(trans, found_root,
7099                                                 path, &first_key, ref_path,
7100                                                 group, reloc_inode);
7101                                 if (ret < 0)
7102                                         goto out;
7103                                 continue;
7104                         }
7105                         /*
7106                          * use fallback method to process the remaining
7107                          * references.
7108                          */
7109                         if (!new_extents) {
7110                                 u64 group_start = group->key.objectid;
7111                                 new_extents = kmalloc(sizeof(*new_extents),
7112                                                       GFP_NOFS);
7113                                 nr_extents = 1;
7114                                 ret = get_new_locations(reloc_inode,
7115                                                         extent_key,
7116                                                         group_start, 1,
7117                                                         &new_extents,
7118                                                         &nr_extents);
7119                                 if (ret)
7120                                         goto out;
7121                         }
7122                         ret = replace_one_extent(trans, found_root,
7123                                                 path, extent_key,
7124                                                 &first_key, ref_path,
7125                                                 new_extents, nr_extents);
7126                 } else {
7127                         ret = relocate_tree_block(trans, found_root, path,
7128                                                   &first_key, ref_path);
7129                 }
7130                 if (ret < 0)
7131                         goto out;
7132         }
7133         ret = 0;
7134 out:
7135         btrfs_end_transaction(trans, extent_root);
7136         kfree(new_extents);
7137         kfree(ref_path);
7138         return ret;
7139 }
7140 #endif
7141
7142 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
7143 {
7144         u64 num_devices;
7145         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
7146                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
7147
7148         num_devices = root->fs_info->fs_devices->rw_devices;
7149         if (num_devices == 1) {
7150                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7151                 stripped = flags & ~stripped;
7152
7153                 /* turn raid0 into single device chunks */
7154                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
7155                         return stripped;
7156
7157                 /* turn mirroring into duplication */
7158                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
7159                              BTRFS_BLOCK_GROUP_RAID10))
7160                         return stripped | BTRFS_BLOCK_GROUP_DUP;
7161                 return flags;
7162         } else {
7163                 /* they already had raid on here, just return */
7164                 if (flags & stripped)
7165                         return flags;
7166
7167                 stripped |= BTRFS_BLOCK_GROUP_DUP;
7168                 stripped = flags & ~stripped;
7169
7170                 /* switch duplicated blocks with raid1 */
7171                 if (flags & BTRFS_BLOCK_GROUP_DUP)
7172                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
7173
7174                 /* turn single device chunks into raid0 */
7175                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
7176         }
7177         return flags;
7178 }
7179
7180 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
7181                      struct btrfs_block_group_cache *shrink_block_group,
7182                      int force)
7183 {
7184         struct btrfs_trans_handle *trans;
7185         u64 new_alloc_flags;
7186         u64 calc;
7187
7188         spin_lock(&shrink_block_group->lock);
7189         if (btrfs_block_group_used(&shrink_block_group->item) +
7190             shrink_block_group->reserved > 0) {
7191                 spin_unlock(&shrink_block_group->lock);
7192
7193                 trans = btrfs_start_transaction(root, 1);
7194                 spin_lock(&shrink_block_group->lock);
7195
7196                 new_alloc_flags = update_block_group_flags(root,
7197                                                    shrink_block_group->flags);
7198                 if (new_alloc_flags != shrink_block_group->flags) {
7199                         calc =
7200                              btrfs_block_group_used(&shrink_block_group->item);
7201                 } else {
7202                         calc = shrink_block_group->key.offset;
7203                 }
7204                 spin_unlock(&shrink_block_group->lock);
7205
7206                 do_chunk_alloc(trans, root->fs_info->extent_root,
7207                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
7208
7209                 btrfs_end_transaction(trans, root);
7210         } else
7211                 spin_unlock(&shrink_block_group->lock);
7212         return 0;
7213 }
7214
7215
7216 int btrfs_prepare_block_group_relocation(struct btrfs_root *root,
7217                                          struct btrfs_block_group_cache *group)
7218
7219 {
7220         __alloc_chunk_for_shrink(root, group, 1);
7221         set_block_group_readonly(group);
7222         return 0;
7223 }
7224
7225 /*
7226  * checks to see if its even possible to relocate this block group.
7227  *
7228  * @return - -1 if it's not a good idea to relocate this block group, 0 if its
7229  * ok to go ahead and try.
7230  */
7231 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
7232 {
7233         struct btrfs_block_group_cache *block_group;
7234         struct btrfs_space_info *space_info;
7235         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7236         struct btrfs_device *device;
7237         int full = 0;
7238         int ret = 0;
7239
7240         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
7241
7242         /* odd, couldn't find the block group, leave it alone */
7243         if (!block_group)
7244                 return -1;
7245
7246         /* no bytes used, we're good */
7247         if (!btrfs_block_group_used(&block_group->item))
7248                 goto out;
7249
7250         space_info = block_group->space_info;
7251         spin_lock(&space_info->lock);
7252
7253         full = space_info->full;
7254
7255         /*
7256          * if this is the last block group we have in this space, we can't
7257          * relocate it unless we're able to allocate a new chunk below.
7258          *
7259          * Otherwise, we need to make sure we have room in the space to handle
7260          * all of the extents from this block group.  If we can, we're good
7261          */
7262         if ((space_info->total_bytes != block_group->key.offset) &&
7263            (space_info->bytes_used + space_info->bytes_reserved +
7264             space_info->bytes_pinned + space_info->bytes_readonly +
7265             btrfs_block_group_used(&block_group->item) <
7266             space_info->total_bytes)) {
7267                 spin_unlock(&space_info->lock);
7268                 goto out;
7269         }
7270         spin_unlock(&space_info->lock);
7271
7272         /*
7273          * ok we don't have enough space, but maybe we have free space on our
7274          * devices to allocate new chunks for relocation, so loop through our
7275          * alloc devices and guess if we have enough space.  However, if we
7276          * were marked as full, then we know there aren't enough chunks, and we
7277          * can just return.
7278          */
7279         ret = -1;
7280         if (full)
7281                 goto out;
7282
7283         mutex_lock(&root->fs_info->chunk_mutex);
7284         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
7285                 u64 min_free = btrfs_block_group_used(&block_group->item);
7286                 u64 dev_offset, max_avail;
7287
7288                 /*
7289                  * check to make sure we can actually find a chunk with enough
7290                  * space to fit our block group in.
7291                  */
7292                 if (device->total_bytes > device->bytes_used + min_free) {
7293                         ret = find_free_dev_extent(NULL, device, min_free,
7294                                                    &dev_offset, &max_avail);
7295                         if (!ret)
7296                                 break;
7297                         ret = -1;
7298                 }
7299         }
7300         mutex_unlock(&root->fs_info->chunk_mutex);
7301 out:
7302         btrfs_put_block_group(block_group);
7303         return ret;
7304 }
7305
7306 static int find_first_block_group(struct btrfs_root *root,
7307                 struct btrfs_path *path, struct btrfs_key *key)
7308 {
7309         int ret = 0;
7310         struct btrfs_key found_key;
7311         struct extent_buffer *leaf;
7312         int slot;
7313
7314         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
7315         if (ret < 0)
7316                 goto out;
7317
7318         while (1) {
7319                 slot = path->slots[0];
7320                 leaf = path->nodes[0];
7321                 if (slot >= btrfs_header_nritems(leaf)) {
7322                         ret = btrfs_next_leaf(root, path);
7323                         if (ret == 0)
7324                                 continue;
7325                         if (ret < 0)
7326                                 goto out;
7327                         break;
7328                 }
7329                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7330
7331                 if (found_key.objectid >= key->objectid &&
7332                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7333                         ret = 0;
7334                         goto out;
7335                 }
7336                 path->slots[0]++;
7337         }
7338         ret = -ENOENT;
7339 out:
7340         return ret;
7341 }
7342
7343 int btrfs_free_block_groups(struct btrfs_fs_info *info)
7344 {
7345         struct btrfs_block_group_cache *block_group;
7346         struct btrfs_space_info *space_info;
7347         struct btrfs_caching_control *caching_ctl;
7348         struct rb_node *n;
7349
7350         down_write(&info->extent_commit_sem);
7351         while (!list_empty(&info->caching_block_groups)) {
7352                 caching_ctl = list_entry(info->caching_block_groups.next,
7353                                          struct btrfs_caching_control, list);
7354                 list_del(&caching_ctl->list);
7355                 put_caching_control(caching_ctl);
7356         }
7357         up_write(&info->extent_commit_sem);
7358
7359         spin_lock(&info->block_group_cache_lock);
7360         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
7361                 block_group = rb_entry(n, struct btrfs_block_group_cache,
7362                                        cache_node);
7363                 rb_erase(&block_group->cache_node,
7364                          &info->block_group_cache_tree);
7365                 spin_unlock(&info->block_group_cache_lock);
7366
7367                 down_write(&block_group->space_info->groups_sem);
7368                 list_del(&block_group->list);
7369                 up_write(&block_group->space_info->groups_sem);
7370
7371                 if (block_group->cached == BTRFS_CACHE_STARTED)
7372                         wait_block_group_cache_done(block_group);
7373
7374                 btrfs_remove_free_space_cache(block_group);
7375
7376                 WARN_ON(atomic_read(&block_group->count) != 1);
7377                 kfree(block_group);
7378
7379                 spin_lock(&info->block_group_cache_lock);
7380         }
7381         spin_unlock(&info->block_group_cache_lock);
7382
7383         /* now that all the block groups are freed, go through and
7384          * free all the space_info structs.  This is only called during
7385          * the final stages of unmount, and so we know nobody is
7386          * using them.  We call synchronize_rcu() once before we start,
7387          * just to be on the safe side.
7388          */
7389         synchronize_rcu();
7390
7391         while(!list_empty(&info->space_info)) {
7392                 space_info = list_entry(info->space_info.next,
7393                                         struct btrfs_space_info,
7394                                         list);
7395
7396                 list_del(&space_info->list);
7397                 kfree(space_info);
7398         }
7399         return 0;
7400 }
7401
7402 int btrfs_read_block_groups(struct btrfs_root *root)
7403 {
7404         struct btrfs_path *path;
7405         int ret;
7406         struct btrfs_block_group_cache *cache;
7407         struct btrfs_fs_info *info = root->fs_info;
7408         struct btrfs_space_info *space_info;
7409         struct btrfs_key key;
7410         struct btrfs_key found_key;
7411         struct extent_buffer *leaf;
7412
7413         root = info->extent_root;
7414         key.objectid = 0;
7415         key.offset = 0;
7416         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
7417         path = btrfs_alloc_path();
7418         if (!path)
7419                 return -ENOMEM;
7420
7421         while (1) {
7422                 ret = find_first_block_group(root, path, &key);
7423                 if (ret > 0) {
7424                         ret = 0;
7425                         goto error;
7426                 }
7427                 if (ret != 0)
7428                         goto error;
7429
7430                 leaf = path->nodes[0];
7431                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7432                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
7433                 if (!cache) {
7434                         ret = -ENOMEM;
7435                         break;
7436                 }
7437
7438                 atomic_set(&cache->count, 1);
7439                 spin_lock_init(&cache->lock);
7440                 spin_lock_init(&cache->tree_lock);
7441                 cache->fs_info = info;
7442                 INIT_LIST_HEAD(&cache->list);
7443                 INIT_LIST_HEAD(&cache->cluster_list);
7444
7445                 /*
7446                  * we only want to have 32k of ram per block group for keeping
7447                  * track of free space, and if we pass 1/2 of that we want to
7448                  * start converting things over to using bitmaps
7449                  */
7450                 cache->extents_thresh = ((1024 * 32) / 2) /
7451                         sizeof(struct btrfs_free_space);
7452
7453                 read_extent_buffer(leaf, &cache->item,
7454                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
7455                                    sizeof(cache->item));
7456                 memcpy(&cache->key, &found_key, sizeof(found_key));
7457
7458                 key.objectid = found_key.objectid + found_key.offset;
7459                 btrfs_release_path(root, path);
7460                 cache->flags = btrfs_block_group_flags(&cache->item);
7461                 cache->sectorsize = root->sectorsize;
7462
7463                 /*
7464                  * check for two cases, either we are full, and therefore
7465                  * don't need to bother with the caching work since we won't
7466                  * find any space, or we are empty, and we can just add all
7467                  * the space in and be done with it.  This saves us _alot_ of
7468                  * time, particularly in the full case.
7469                  */
7470                 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
7471                         exclude_super_stripes(root, cache);
7472                         cache->last_byte_to_unpin = (u64)-1;
7473                         cache->cached = BTRFS_CACHE_FINISHED;
7474                         free_excluded_extents(root, cache);
7475                 } else if (btrfs_block_group_used(&cache->item) == 0) {
7476                         exclude_super_stripes(root, cache);
7477                         cache->last_byte_to_unpin = (u64)-1;
7478                         cache->cached = BTRFS_CACHE_FINISHED;
7479                         add_new_free_space(cache, root->fs_info,
7480                                            found_key.objectid,
7481                                            found_key.objectid +
7482                                            found_key.offset);
7483                         free_excluded_extents(root, cache);
7484                 }
7485
7486                 ret = update_space_info(info, cache->flags, found_key.offset,
7487                                         btrfs_block_group_used(&cache->item),
7488                                         &space_info);
7489                 BUG_ON(ret);
7490                 cache->space_info = space_info;
7491                 spin_lock(&cache->space_info->lock);
7492                 cache->space_info->bytes_super += cache->bytes_super;
7493                 spin_unlock(&cache->space_info->lock);
7494
7495                 down_write(&space_info->groups_sem);
7496                 list_add_tail(&cache->list, &space_info->block_groups);
7497                 up_write(&space_info->groups_sem);
7498
7499                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
7500                 BUG_ON(ret);
7501
7502                 set_avail_alloc_bits(root->fs_info, cache->flags);
7503                 if (btrfs_chunk_readonly(root, cache->key.objectid))
7504                         set_block_group_readonly(cache);
7505         }
7506         ret = 0;
7507 error:
7508         btrfs_free_path(path);
7509         return ret;
7510 }
7511
7512 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
7513                            struct btrfs_root *root, u64 bytes_used,
7514                            u64 type, u64 chunk_objectid, u64 chunk_offset,
7515                            u64 size)
7516 {
7517         int ret;
7518         struct btrfs_root *extent_root;
7519         struct btrfs_block_group_cache *cache;
7520
7521         extent_root = root->fs_info->extent_root;
7522
7523         root->fs_info->last_trans_log_full_commit = trans->transid;
7524
7525         cache = kzalloc(sizeof(*cache), GFP_NOFS);
7526         if (!cache)
7527                 return -ENOMEM;
7528
7529         cache->key.objectid = chunk_offset;
7530         cache->key.offset = size;
7531         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
7532         cache->sectorsize = root->sectorsize;
7533
7534         /*
7535          * we only want to have 32k of ram per block group for keeping track
7536          * of free space, and if we pass 1/2 of that we want to start
7537          * converting things over to using bitmaps
7538          */
7539         cache->extents_thresh = ((1024 * 32) / 2) /
7540                 sizeof(struct btrfs_free_space);
7541         atomic_set(&cache->count, 1);
7542         spin_lock_init(&cache->lock);
7543         spin_lock_init(&cache->tree_lock);
7544         INIT_LIST_HEAD(&cache->list);
7545         INIT_LIST_HEAD(&cache->cluster_list);
7546
7547         btrfs_set_block_group_used(&cache->item, bytes_used);
7548         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
7549         cache->flags = type;
7550         btrfs_set_block_group_flags(&cache->item, type);
7551
7552         cache->last_byte_to_unpin = (u64)-1;
7553         cache->cached = BTRFS_CACHE_FINISHED;
7554         exclude_super_stripes(root, cache);
7555
7556         add_new_free_space(cache, root->fs_info, chunk_offset,
7557                            chunk_offset + size);
7558
7559         free_excluded_extents(root, cache);
7560
7561         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
7562                                 &cache->space_info);
7563         BUG_ON(ret);
7564
7565         spin_lock(&cache->space_info->lock);
7566         cache->space_info->bytes_super += cache->bytes_super;
7567         spin_unlock(&cache->space_info->lock);
7568
7569         down_write(&cache->space_info->groups_sem);
7570         list_add_tail(&cache->list, &cache->space_info->block_groups);
7571         up_write(&cache->space_info->groups_sem);
7572
7573         ret = btrfs_add_block_group_cache(root->fs_info, cache);
7574         BUG_ON(ret);
7575
7576         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
7577                                 sizeof(cache->item));
7578         BUG_ON(ret);
7579
7580         set_avail_alloc_bits(extent_root->fs_info, type);
7581
7582         return 0;
7583 }
7584
7585 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
7586                              struct btrfs_root *root, u64 group_start)
7587 {
7588         struct btrfs_path *path;
7589         struct btrfs_block_group_cache *block_group;
7590         struct btrfs_free_cluster *cluster;
7591         struct btrfs_key key;
7592         int ret;
7593
7594         root = root->fs_info->extent_root;
7595
7596         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
7597         BUG_ON(!block_group);
7598         BUG_ON(!block_group->ro);
7599
7600         memcpy(&key, &block_group->key, sizeof(key));
7601
7602         /* make sure this block group isn't part of an allocation cluster */
7603         cluster = &root->fs_info->data_alloc_cluster;
7604         spin_lock(&cluster->refill_lock);
7605         btrfs_return_cluster_to_free_space(block_group, cluster);
7606         spin_unlock(&cluster->refill_lock);
7607
7608         /*
7609          * make sure this block group isn't part of a metadata
7610          * allocation cluster
7611          */
7612         cluster = &root->fs_info->meta_alloc_cluster;
7613         spin_lock(&cluster->refill_lock);
7614         btrfs_return_cluster_to_free_space(block_group, cluster);
7615         spin_unlock(&cluster->refill_lock);
7616
7617         path = btrfs_alloc_path();
7618         BUG_ON(!path);
7619
7620         spin_lock(&root->fs_info->block_group_cache_lock);
7621         rb_erase(&block_group->cache_node,
7622                  &root->fs_info->block_group_cache_tree);
7623         spin_unlock(&root->fs_info->block_group_cache_lock);
7624
7625         down_write(&block_group->space_info->groups_sem);
7626         /*
7627          * we must use list_del_init so people can check to see if they
7628          * are still on the list after taking the semaphore
7629          */
7630         list_del_init(&block_group->list);
7631         up_write(&block_group->space_info->groups_sem);
7632
7633         if (block_group->cached == BTRFS_CACHE_STARTED)
7634                 wait_block_group_cache_done(block_group);
7635
7636         btrfs_remove_free_space_cache(block_group);
7637
7638         spin_lock(&block_group->space_info->lock);
7639         block_group->space_info->total_bytes -= block_group->key.offset;
7640         block_group->space_info->bytes_readonly -= block_group->key.offset;
7641         spin_unlock(&block_group->space_info->lock);
7642
7643         btrfs_clear_space_info_full(root->fs_info);
7644
7645         btrfs_put_block_group(block_group);
7646         btrfs_put_block_group(block_group);
7647
7648         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7649         if (ret > 0)
7650                 ret = -EIO;
7651         if (ret < 0)
7652                 goto out;
7653
7654         ret = btrfs_del_item(trans, root, path);
7655 out:
7656         btrfs_free_path(path);
7657         return ret;
7658 }