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 "compat.h"
  25 #include "hash.h"
  26 #include "crc32c.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 "ref-cache.h"
  34
  35 #define PENDING_EXTENT_INSERT 0
  36 #define PENDING_EXTENT_DELETE 1
  37 #define PENDING_BACKREF_UPDATE 2
  38
  39 struct pending_extent_op {
  40         int type;
  41         u64 bytenr;
  42         u64 num_bytes;
  43         u64 parent;
  44         u64 orig_parent;
  45         u64 generation;
  46         u64 orig_generation;
  47         int level;
  48         struct list_head list;
  49         int del;
  50 };
  51
  52 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
  53                                          struct btrfs_root *root, u64 parent,
  54                                          u64 root_objectid, u64 ref_generation,
  55                                          u64 owner, struct btrfs_key *ins,
  56                                          int ref_mod);
  57 static int update_reserved_extents(struct btrfs_root *root,
  58                                    u64 bytenr, u64 num, int reserve);
  59 static int update_block_group(struct btrfs_trans_handle *trans,
  60                               struct btrfs_root *root,
  61                               u64 bytenr, u64 num_bytes, int alloc,
  62                               int mark_free);
  63 static noinline int __btrfs_free_extent(struct btrfs_trans_handle *trans,
  64                                         struct btrfs_root *root,
  65                                         u64 bytenr, u64 num_bytes, u64 parent,
  66                                         u64 root_objectid, u64 ref_generation,
  67                                         u64 owner_objectid, int pin,
  68                                         int ref_to_drop);
  69
  70 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
  71                           struct btrfs_root *extent_root, u64 alloc_bytes,
  72                           u64 flags, int force);
  73
  74 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  75 {
  76         return (cache->flags & bits) == bits;
  77 }
  78
  79 /*
  80  * this adds the block group to the fs_info rb tree for the block group
  81  * cache
  82  */
  83 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  84                                 struct btrfs_block_group_cache *block_group)
  85 {
  86         struct rb_node **p;
  87         struct rb_node *parent = NULL;
  88         struct btrfs_block_group_cache *cache;
  89
  90         spin_lock(&info->block_group_cache_lock);
  91         p = &info->block_group_cache_tree.rb_node;
  92
  93         while (*p) {
  94                 parent = *p;
  95                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  96                                  cache_node);
  97                 if (block_group->key.objectid < cache->key.objectid) {
  98                         p = &(*p)->rb_left;
  99                 } else if (block_group->key.objectid > cache->key.objectid) {
 100                         p = &(*p)->rb_right;
 101                 } else {
 102                         spin_unlock(&info->block_group_cache_lock);
 103                         return -EEXIST;
 104                 }
 105         }
 106
 107         rb_link_node(&block_group->cache_node, parent, p);
 108         rb_insert_color(&block_group->cache_node,
 109                         &info->block_group_cache_tree);
 110         spin_unlock(&info->block_group_cache_lock);
 111
 112         return 0;
 113 }
 114
 115 /*
 116  * This will return the block group at or after bytenr if contains is 0, else
 117  * it will return the block group that contains the bytenr
 118  */
 119 static struct btrfs_block_group_cache *
 120 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 121                               int contains)
 122 {
 123         struct btrfs_block_group_cache *cache, *ret = NULL;
 124         struct rb_node *n;
 125         u64 end, start;
 126
 127         spin_lock(&info->block_group_cache_lock);
 128         n = info->block_group_cache_tree.rb_node;
 129
 130         while (n) {
 131                 cache = rb_entry(n, struct btrfs_block_group_cache,
 132                                  cache_node);
 133                 end = cache->key.objectid + cache->key.offset - 1;
 134                 start = cache->key.objectid;
 135
 136                 if (bytenr < start) {
 137                         if (!contains && (!ret || start < ret->key.objectid))
 138                                 ret = cache;
 139                         n = n->rb_left;
 140                 } else if (bytenr > start) {
 141                         if (contains && bytenr <= end) {
 142                                 ret = cache;
 143                                 break;
 144                         }
 145                         n = n->rb_right;
 146                 } else {
 147                         ret = cache;
 148                         break;
 149                 }
 150         }
 151         if (ret)
 152                 atomic_inc(&ret->count);
 153         spin_unlock(&info->block_group_cache_lock);
 154
 155         return ret;
 156 }
 157
 158 /*
 159  * this is only called by cache_block_group, since we could have freed extents
 160  * we need to check the pinned_extents for any extents that can't be used yet
 161  * since their free space will be released as soon as the transaction commits.
 162  */
 163 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 164                               struct btrfs_fs_info *info, u64 start, u64 end)
 165 {
 166         u64 extent_start, extent_end, size;
 167         int ret;
 168
 169         mutex_lock(&info->pinned_mutex);
 170         while (start < end) {
 171                 ret = find_first_extent_bit(&info->pinned_extents, start,
 172                                             &extent_start, &extent_end,
 173                                             EXTENT_DIRTY);
 174                 if (ret)
 175                         break;
 176
 177                 if (extent_start == start) {
 178                         start = extent_end + 1;
 179                 } else if (extent_start > start && extent_start < end) {
 180                         size = extent_start - start;
 181                         ret = btrfs_add_free_space(block_group, start,
 182                                                    size);
 183                         BUG_ON(ret);
 184                         start = extent_end + 1;
 185                 } else {
 186                         break;
 187                 }
 188         }
 189
 190         if (start < end) {
 191                 size = end - start;
 192                 ret = btrfs_add_free_space(block_group, start, size);
 193                 BUG_ON(ret);
 194         }
 195         mutex_unlock(&info->pinned_mutex);
 196
 197         return 0;
 198 }
 199
 200 static int remove_sb_from_cache(struct btrfs_root *root,
 201                                 struct btrfs_block_group_cache *cache)
 202 {
 203         u64 bytenr;
 204         u64 *logical;
 205         int stripe_len;
 206         int i, nr, ret;
 207
 208         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 209                 bytenr = btrfs_sb_offset(i);
 210                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 211                                        cache->key.objectid, bytenr, 0,
 212                                        &logical, &nr, &stripe_len);
 213                 BUG_ON(ret);
 214                 while (nr--) {
 215                         btrfs_remove_free_space(cache, logical[nr],
 216                                                 stripe_len);
 217                 }
 218                 kfree(logical);
 219         }
 220         return 0;
 221 }
 222
 223 static int cache_block_group(struct btrfs_root *root,
 224                              struct btrfs_block_group_cache *block_group)
 225 {
 226         struct btrfs_path *path;
 227         int ret = 0;
 228         struct btrfs_key key;
 229         struct extent_buffer *leaf;
 230         int slot;
 231         u64 last;
 232
 233         if (!block_group)
 234                 return 0;
 235
 236         root = root->fs_info->extent_root;
 237
 238         if (block_group->cached)
 239                 return 0;
 240
 241         path = btrfs_alloc_path();
 242         if (!path)
 243                 return -ENOMEM;
 244
 245         path->reada = 2;
 246         /*
 247          * we get into deadlocks with paths held by callers of this function.
 248          * since the alloc_mutex is protecting things right now, just
 249          * skip the locking here
 250          */
 251         path->skip_locking = 1;
 252         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 253         key.objectid = last;
 254         key.offset = 0;
 255         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 256         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 257         if (ret < 0)
 258                 goto err;
 259
 260         while (1) {
 261                 leaf = path->nodes[0];
 262                 slot = path->slots[0];
 263                 if (slot >= btrfs_header_nritems(leaf)) {
 264                         ret = btrfs_next_leaf(root, path);
 265                         if (ret < 0)
 266                                 goto err;
 267                         if (ret == 0)
 268                                 continue;
 269                         else
 270                                 break;
 271                 }
 272                 btrfs_item_key_to_cpu(leaf, &key, slot);
 273                 if (key.objectid < block_group->key.objectid)
 274                         goto next;
 275
 276                 if (key.objectid >= block_group->key.objectid +
 277                     block_group->key.offset)
 278                         break;
 279
 280                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 281                         add_new_free_space(block_group, root->fs_info, last,
 282                                            key.objectid);
 283
 284                         last = key.objectid + key.offset;
 285                 }
 286 next:
 287                 path->slots[0]++;
 288         }
 289
 290         add_new_free_space(block_group, root->fs_info, last,
 291                            block_group->key.objectid +
 292                            block_group->key.offset);
 293
 294         block_group->cached = 1;
 295         remove_sb_from_cache(root, block_group);
 296         ret = 0;
 297 err:
 298         btrfs_free_path(path);
 299         return ret;
 300 }
 301
 302 /*
 303  * return the block group that starts at or after bytenr
 304  */
 305 static struct btrfs_block_group_cache *
 306 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 307 {
 308         struct btrfs_block_group_cache *cache;
 309
 310         cache = block_group_cache_tree_search(info, bytenr, 0);
 311
 312         return cache;
 313 }
 314
 315 /*
 316  * return the block group that contains teh given bytenr
 317  */
 318 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 319                                                  struct btrfs_fs_info *info,
 320                                                  u64 bytenr)
 321 {
 322         struct btrfs_block_group_cache *cache;
 323
 324         cache = block_group_cache_tree_search(info, bytenr, 1);
 325
 326         return cache;
 327 }
 328
 329 static inline void put_block_group(struct btrfs_block_group_cache *cache)
 330 {
 331         if (atomic_dec_and_test(&cache->count))
 332                 kfree(cache);
 333 }
 334
 335 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 336                                                   u64 flags)
 337 {
 338         struct list_head *head = &info->space_info;
 339         struct btrfs_space_info *found;
 340
 341         rcu_read_lock();
 342         list_for_each_entry_rcu(found, head, list) {
 343                 if (found->flags == flags) {
 344                         rcu_read_unlock();
 345                         return found;
 346                 }
 347         }
 348         rcu_read_unlock();
 349         return NULL;
 350 }
 351
 352 /*
 353  * after adding space to the filesystem, we need to clear the full flags
 354  * on all the space infos.
 355  */
 356 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
 357 {
 358         struct list_head *head = &info->space_info;
 359         struct btrfs_space_info *found;
 360
 361         rcu_read_lock();
 362         list_for_each_entry_rcu(found, head, list)
 363                 found->full = 0;
 364         rcu_read_unlock();
 365 }
 366
 367 static u64 div_factor(u64 num, int factor)
 368 {
 369         if (factor == 10)
 370                 return num;
 371         num *= factor;
 372         do_div(num, 10);
 373         return num;
 374 }
 375
 376 u64 btrfs_find_block_group(struct btrfs_root *root,
 377                            u64 search_start, u64 search_hint, int owner)
 378 {
 379         struct btrfs_block_group_cache *cache;
 380         u64 used;
 381         u64 last = max(search_hint, search_start);
 382         u64 group_start = 0;
 383         int full_search = 0;
 384         int factor = 9;
 385         int wrapped = 0;
 386 again:
 387         while (1) {
 388                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 389                 if (!cache)
 390                         break;
 391
 392                 spin_lock(&cache->lock);
 393                 last = cache->key.objectid + cache->key.offset;
 394                 used = btrfs_block_group_used(&cache->item);
 395
 396                 if ((full_search || !cache->ro) &&
 397                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 398                         if (used + cache->pinned + cache->reserved <
 399                             div_factor(cache->key.offset, factor)) {
 400                                 group_start = cache->key.objectid;
 401                                 spin_unlock(&cache->lock);
 402                                 put_block_group(cache);
 403                                 goto found;
 404                         }
 405                 }
 406                 spin_unlock(&cache->lock);
 407                 put_block_group(cache);
 408                 cond_resched();
 409         }
 410         if (!wrapped) {
 411                 last = search_start;
 412                 wrapped = 1;
 413                 goto again;
 414         }
 415         if (!full_search && factor < 10) {
 416                 last = search_start;
 417                 full_search = 1;
 418                 factor = 10;
 419                 goto again;
 420         }
 421 found:
 422         return group_start;
 423 }
 424
 425 /* simple helper to search for an existing extent at a given offset */
 426 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 427 {
 428         int ret;
 429         struct btrfs_key key;
 430         struct btrfs_path *path;
 431
 432         path = btrfs_alloc_path();
 433         BUG_ON(!path);
 434         key.objectid = start;
 435         key.offset = len;
 436         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 437         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 438                                 0, 0);
 439         btrfs_free_path(path);
 440         return ret;
 441 }
 442
 443 /*
 444  * Back reference rules.  Back refs have three main goals:
 445  *
 446  * 1) differentiate between all holders of references to an extent so that
 447  *    when a reference is dropped we can make sure it was a valid reference
 448  *    before freeing the extent.
 449  *
 450  * 2) Provide enough information to quickly find the holders of an extent
 451  *    if we notice a given block is corrupted or bad.
 452  *
 453  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 454  *    maintenance.  This is actually the same as #2, but with a slightly
 455  *    different use case.
 456  *
 457  * File extents can be referenced by:
 458  *
 459  * - multiple snapshots, subvolumes, or different generations in one subvol
 460  * - different files inside a single subvolume
 461  * - different offsets inside a file (bookend extents in file.c)
 462  *
 463  * The extent ref structure has fields for:
 464  *
 465  * - Objectid of the subvolume root
 466  * - Generation number of the tree holding the reference
 467  * - objectid of the file holding the reference
 468  * - number of references holding by parent node (alway 1 for tree blocks)
 469  *
 470  * Btree leaf may hold multiple references to a file extent. In most cases,
 471  * these references are from same file and the corresponding offsets inside
 472  * the file are close together.
 473  *
 474  * When a file extent is allocated the fields are filled in:
 475  *     (root_key.objectid, trans->transid, inode objectid, 1)
 476  *
 477  * When a leaf is cow'd new references are added for every file extent found
 478  * in the leaf.  It looks similar to the create case, but trans->transid will
 479  * be different when the block is cow'd.
 480  *
 481  *     (root_key.objectid, trans->transid, inode objectid,
 482  *      number of references in the leaf)
 483  *
 484  * When a file extent is removed either during snapshot deletion or
 485  * file truncation, we find the corresponding back reference and check
 486  * the following fields:
 487  *
 488  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 489  *      inode objectid)
 490  *
 491  * Btree extents can be referenced by:
 492  *
 493  * - Different subvolumes
 494  * - Different generations of the same subvolume
 495  *
 496  * When a tree block is created, back references are inserted:
 497  *
 498  * (root->root_key.objectid, trans->transid, level, 1)
 499  *
 500  * When a tree block is cow'd, new back references are added for all the
 501  * blocks it points to. If the tree block isn't in reference counted root,
 502  * the old back references are removed. These new back references are of
 503  * the form (trans->transid will have increased since creation):
 504  *
 505  * (root->root_key.objectid, trans->transid, level, 1)
 506  *
 507  * When a backref is in deleting, the following fields are checked:
 508  *
 509  * if backref was for a tree root:
 510  *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
 511  * else
 512  *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
 513  *
 514  * Back Reference Key composing:
 515  *
 516  * The key objectid corresponds to the first byte in the extent, the key
 517  * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
 518  * byte of parent extent. If a extent is tree root, the key offset is set
 519  * to the key objectid.
 520  */
 521
 522 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
 523                                           struct btrfs_root *root,
 524                                           struct btrfs_path *path,
 525                                           u64 bytenr, u64 parent,
 526                                           u64 ref_root, u64 ref_generation,
 527                                           u64 owner_objectid, int del)
 528 {
 529         struct btrfs_key key;
 530         struct btrfs_extent_ref *ref;
 531         struct extent_buffer *leaf;
 532         u64 ref_objectid;
 533         int ret;
 534
 535         key.objectid = bytenr;
 536         key.type = BTRFS_EXTENT_REF_KEY;
 537         key.offset = parent;
 538
 539         ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
 540         if (ret < 0)
 541                 goto out;
 542         if (ret > 0) {
 543                 ret = -ENOENT;
 544                 goto out;
 545         }
 546
 547         leaf = path->nodes[0];
 548         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 549         ref_objectid = btrfs_ref_objectid(leaf, ref);
 550         if (btrfs_ref_root(leaf, ref) != ref_root ||
 551             btrfs_ref_generation(leaf, ref) != ref_generation ||
 552             (ref_objectid != owner_objectid &&
 553              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 554                 ret = -EIO;
 555                 WARN_ON(1);
 556                 goto out;
 557         }
 558         ret = 0;
 559 out:
 560         return ret;
 561 }
 562
 563 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
 564                                           struct btrfs_root *root,
 565                                           struct btrfs_path *path,
 566                                           u64 bytenr, u64 parent,
 567                                           u64 ref_root, u64 ref_generation,
 568                                           u64 owner_objectid,
 569                                           int refs_to_add)
 570 {
 571         struct btrfs_key key;
 572         struct extent_buffer *leaf;
 573         struct btrfs_extent_ref *ref;
 574         u32 num_refs;
 575         int ret;
 576
 577         key.objectid = bytenr;
 578         key.type = BTRFS_EXTENT_REF_KEY;
 579         key.offset = parent;
 580
 581         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
 582         if (ret == 0) {
 583                 leaf = path->nodes[0];
 584                 ref = btrfs_item_ptr(leaf, path->slots[0],
 585                                      struct btrfs_extent_ref);
 586                 btrfs_set_ref_root(leaf, ref, ref_root);
 587                 btrfs_set_ref_generation(leaf, ref, ref_generation);
 588                 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
 589                 btrfs_set_ref_num_refs(leaf, ref, refs_to_add);
 590         } else if (ret == -EEXIST) {
 591                 u64 existing_owner;
 592
 593                 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
 594                 leaf = path->nodes[0];
 595                 ref = btrfs_item_ptr(leaf, path->slots[0],
 596                                      struct btrfs_extent_ref);
 597                 if (btrfs_ref_root(leaf, ref) != ref_root ||
 598                     btrfs_ref_generation(leaf, ref) != ref_generation) {
 599                         ret = -EIO;
 600                         WARN_ON(1);
 601                         goto out;
 602                 }
 603
 604                 num_refs = btrfs_ref_num_refs(leaf, ref);
 605                 BUG_ON(num_refs == 0);
 606                 btrfs_set_ref_num_refs(leaf, ref, num_refs + refs_to_add);
 607
 608                 existing_owner = btrfs_ref_objectid(leaf, ref);
 609                 if (existing_owner != owner_objectid &&
 610                     existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
 611                         btrfs_set_ref_objectid(leaf, ref,
 612                                         BTRFS_MULTIPLE_OBJECTIDS);
 613                 }
 614                 ret = 0;
 615         } else {
 616                 goto out;
 617         }
 618         btrfs_unlock_up_safe(path, 1);
 619         btrfs_mark_buffer_dirty(path->nodes[0]);
 620 out:
 621         btrfs_release_path(root, path);
 622         return ret;
 623 }
 624
 625 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
 626                                           struct btrfs_root *root,
 627                                           struct btrfs_path *path,
 628                                           int refs_to_drop)
 629 {
 630         struct extent_buffer *leaf;
 631         struct btrfs_extent_ref *ref;
 632         u32 num_refs;
 633         int ret = 0;
 634
 635         leaf = path->nodes[0];
 636         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 637         num_refs = btrfs_ref_num_refs(leaf, ref);
 638         BUG_ON(num_refs < refs_to_drop);
 639         num_refs -= refs_to_drop;
 640         if (num_refs == 0) {
 641                 ret = btrfs_del_item(trans, root, path);
 642         } else {
 643                 btrfs_set_ref_num_refs(leaf, ref, num_refs);
 644                 btrfs_mark_buffer_dirty(leaf);
 645         }
 646         btrfs_release_path(root, path);
 647         return ret;
 648 }
 649
 650 #ifdef BIO_RW_DISCARD
 651 static void btrfs_issue_discard(struct block_device *bdev,
 652                                 u64 start, u64 len)
 653 {
 654         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
 655 }
 656 #endif
 657
 658 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
 659                                 u64 num_bytes)
 660 {
 661 #ifdef BIO_RW_DISCARD
 662         int ret;
 663         u64 map_length = num_bytes;
 664         struct btrfs_multi_bio *multi = NULL;
 665
 666         /* Tell the block device(s) that the sectors can be discarded */
 667         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
 668                               bytenr, &map_length, &multi, 0);
 669         if (!ret) {
 670                 struct btrfs_bio_stripe *stripe = multi->stripes;
 671                 int i;
 672
 673                 if (map_length > num_bytes)
 674                         map_length = num_bytes;
 675
 676                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
 677                         btrfs_issue_discard(stripe->dev->bdev,
 678                                             stripe->physical,
 679                                             map_length);
 680                 }
 681                 kfree(multi);
 682         }
 683
 684         return ret;
 685 #else
 686         return 0;
 687 #endif
 688 }
 689
 690 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
 691                                      struct btrfs_root *root, u64 bytenr,
 692                                      u64 num_bytes,
 693                                      u64 orig_parent, u64 parent,
 694                                      u64 orig_root, u64 ref_root,
 695                                      u64 orig_generation, u64 ref_generation,
 696                                      u64 owner_objectid)
 697 {
 698         int ret;
 699         int pin = owner_objectid < BTRFS_FIRST_FREE_OBJECTID;
 700
 701         ret = btrfs_update_delayed_ref(trans, bytenr, num_bytes,
 702                                        orig_parent, parent, orig_root,
 703                                        ref_root, orig_generation,
 704                                        ref_generation, owner_objectid, pin);
 705         BUG_ON(ret);
 706         return ret;
 707 }
 708
 709 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
 710                             struct btrfs_root *root, u64 bytenr,
 711                             u64 num_bytes, u64 orig_parent, u64 parent,
 712                             u64 ref_root, u64 ref_generation,
 713                             u64 owner_objectid)
 714 {
 715         int ret;
 716         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
 717             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
 718                 return 0;
 719
 720         ret = __btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
 721                                         orig_parent, parent, ref_root,
 722                                         ref_root, ref_generation,
 723                                         ref_generation, owner_objectid);
 724         return ret;
 725 }
 726 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 727                                   struct btrfs_root *root, u64 bytenr,
 728                                   u64 num_bytes,
 729                                   u64 orig_parent, u64 parent,
 730                                   u64 orig_root, u64 ref_root,
 731                                   u64 orig_generation, u64 ref_generation,
 732                                   u64 owner_objectid)
 733 {
 734         int ret;
 735
 736         ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, ref_root,
 737                                     ref_generation, owner_objectid,
 738                                     BTRFS_ADD_DELAYED_REF, 0);
 739         BUG_ON(ret);
 740         return ret;
 741 }
 742
 743 static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans,
 744                           struct btrfs_root *root, u64 bytenr,
 745                           u64 num_bytes, u64 parent, u64 ref_root,
 746                           u64 ref_generation, u64 owner_objectid,
 747                           int refs_to_add)
 748 {
 749         struct btrfs_path *path;
 750         int ret;
 751         struct btrfs_key key;
 752         struct extent_buffer *l;
 753         struct btrfs_extent_item *item;
 754         u32 refs;
 755
 756         path = btrfs_alloc_path();
 757         if (!path)
 758                 return -ENOMEM;
 759
 760         path->reada = 1;
 761         path->leave_spinning = 1;
 762         key.objectid = bytenr;
 763         key.type = BTRFS_EXTENT_ITEM_KEY;
 764         key.offset = num_bytes;
 765
 766         /* first find the extent item and update its reference count */
 767         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
 768                                 path, 0, 1);
 769         if (ret < 0) {
 770                 btrfs_set_path_blocking(path);
 771                 return ret;
 772         }
 773
 774         if (ret > 0) {
 775                 WARN_ON(1);
 776                 btrfs_free_path(path);
 777                 return -EIO;
 778         }
 779         l = path->nodes[0];
 780
 781         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 782         if (key.objectid != bytenr) {
 783                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
 784                 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
 785                        (unsigned long long)bytenr,
 786                        (unsigned long long)key.objectid);
 787                 BUG();
 788         }
 789         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
 790
 791         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 792
 793         refs = btrfs_extent_refs(l, item);
 794         btrfs_set_extent_refs(l, item, refs + refs_to_add);
 795         btrfs_unlock_up_safe(path, 1);
 796
 797         btrfs_mark_buffer_dirty(path->nodes[0]);
 798
 799         btrfs_release_path(root->fs_info->extent_root, path);
 800
 801         path->reada = 1;
 802         path->leave_spinning = 1;
 803
 804         /* now insert the actual backref */
 805         ret = insert_extent_backref(trans, root->fs_info->extent_root,
 806                                     path, bytenr, parent,
 807                                     ref_root, ref_generation,
 808                                     owner_objectid, refs_to_add);
 809         BUG_ON(ret);
 810         btrfs_free_path(path);
 811         return 0;
 812 }
 813
 814 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 815                          struct btrfs_root *root,
 816                          u64 bytenr, u64 num_bytes, u64 parent,
 817                          u64 ref_root, u64 ref_generation,
 818                          u64 owner_objectid)
 819 {
 820         int ret;
 821         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
 822             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
 823                 return 0;
 824
 825         ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, parent,
 826                                      0, ref_root, 0, ref_generation,
 827                                      owner_objectid);
 828         return ret;
 829 }
 830
 831 static int drop_delayed_ref(struct btrfs_trans_handle *trans,
 832                                         struct btrfs_root *root,
 833                                         struct btrfs_delayed_ref_node *node)
 834 {
 835         int ret = 0;
 836         struct btrfs_delayed_ref *ref = btrfs_delayed_node_to_ref(node);
 837
 838         BUG_ON(node->ref_mod == 0);
 839         ret = __btrfs_free_extent(trans, root, node->bytenr, node->num_bytes,
 840                                   node->parent, ref->root, ref->generation,
 841                                   ref->owner_objectid, ref->pin, node->ref_mod);
 842
 843         return ret;
 844 }
 845
 846 /* helper function to actually process a single delayed ref entry */
 847 static noinline int run_one_delayed_ref(struct btrfs_trans_handle *trans,
 848                                         struct btrfs_root *root,
 849                                         struct btrfs_delayed_ref_node *node,
 850                                         int insert_reserved)
 851 {
 852         int ret;
 853         struct btrfs_delayed_ref *ref;
 854
 855         if (node->parent == (u64)-1) {
 856                 struct btrfs_delayed_ref_head *head;
 857                 /*
 858                  * we've hit the end of the chain and we were supposed
 859                  * to insert this extent into the tree.  But, it got
 860                  * deleted before we ever needed to insert it, so all
 861                  * we have to do is clean up the accounting
 862                  */
 863                 if (insert_reserved) {
 864                         update_reserved_extents(root, node->bytenr,
 865                                                 node->num_bytes, 0);
 866                 }
 867                 head = btrfs_delayed_node_to_head(node);
 868                 mutex_unlock(&head->mutex);
 869                 return 0;
 870         }
 871
 872         ref = btrfs_delayed_node_to_ref(node);
 873         if (ref->action == BTRFS_ADD_DELAYED_REF) {
 874                 if (insert_reserved) {
 875                         struct btrfs_key ins;
 876
 877                         ins.objectid = node->bytenr;
 878                         ins.offset = node->num_bytes;
 879                         ins.type = BTRFS_EXTENT_ITEM_KEY;
 880
 881                         /* record the full extent allocation */
 882                         ret = __btrfs_alloc_reserved_extent(trans, root,
 883                                         node->parent, ref->root,
 884                                         ref->generation, ref->owner_objectid,
 885                                         &ins, node->ref_mod);
 886                         update_reserved_extents(root, node->bytenr,
 887                                                 node->num_bytes, 0);
 888                 } else {
 889                         /* just add one backref */
 890                         ret = add_extent_ref(trans, root, node->bytenr,
 891                                      node->num_bytes,
 892                                      node->parent, ref->root, ref->generation,
 893                                      ref->owner_objectid, node->ref_mod);
 894                 }
 895                 BUG_ON(ret);
 896         } else if (ref->action == BTRFS_DROP_DELAYED_REF) {
 897                 WARN_ON(insert_reserved);
 898                 ret = drop_delayed_ref(trans, root, node);
 899         }
 900         return 0;
 901 }
 902
 903 static noinline struct btrfs_delayed_ref_node *
 904 select_delayed_ref(struct btrfs_delayed_ref_head *head)
 905 {
 906         struct rb_node *node;
 907         struct btrfs_delayed_ref_node *ref;
 908         int action = BTRFS_ADD_DELAYED_REF;
 909 again:
 910         /*
 911          * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
 912          * this prevents ref count from going down to zero when
 913          * there still are pending delayed ref.
 914          */
 915         node = rb_prev(&head->node.rb_node);
 916         while (1) {
 917                 if (!node)
 918                         break;
 919                 ref = rb_entry(node, struct btrfs_delayed_ref_node,
 920                                 rb_node);
 921                 if (ref->bytenr != head->node.bytenr)
 922                         break;
 923                 if (btrfs_delayed_node_to_ref(ref)->action == action)
 924                         return ref;
 925                 node = rb_prev(node);
 926         }
 927         if (action == BTRFS_ADD_DELAYED_REF) {
 928                 action = BTRFS_DROP_DELAYED_REF;
 929                 goto again;
 930         }
 931         return NULL;
 932 }
 933
 934 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 935                                        struct btrfs_root *root,
 936                                        struct list_head *cluster)
 937 {
 938         struct btrfs_delayed_ref_root *delayed_refs;
 939         struct btrfs_delayed_ref_node *ref;
 940         struct btrfs_delayed_ref_head *locked_ref = NULL;
 941         int ret;
 942         int count = 0;
 943         int must_insert_reserved = 0;
 944
 945         delayed_refs = &trans->transaction->delayed_refs;
 946         while (1) {
 947                 if (!locked_ref) {
 948                         /* pick a new head ref from the cluster list */
 949                         if (list_empty(cluster))
 950                                 break;
 951
 952                         locked_ref = list_entry(cluster->next,
 953                                      struct btrfs_delayed_ref_head, cluster);
 954
 955                         /* grab the lock that says we are going to process
 956                          * all the refs for this head */
 957                         ret = btrfs_delayed_ref_lock(trans, locked_ref);
 958
 959                         /*
 960                          * we may have dropped the spin lock to get the head
 961                          * mutex lock, and that might have given someone else
 962                          * time to free the head.  If that's true, it has been
 963                          * removed from our list and we can move on.
 964                          */
 965                         if (ret == -EAGAIN) {
 966                                 locked_ref = NULL;
 967                                 count++;
 968                                 continue;
 969                         }
 970                 }
 971
 972                 /*
 973                  * record the must insert reserved flag before we
 974                  * drop the spin lock.
 975                  */
 976                 must_insert_reserved = locked_ref->must_insert_reserved;
 977                 locked_ref->must_insert_reserved = 0;
 978
 979                 /*
 980                  * locked_ref is the head node, so we have to go one
 981                  * node back for any delayed ref updates
 982                  */
 983                 ref = select_delayed_ref(locked_ref);
 984                 if (!ref) {
 985                         /* All delayed refs have been processed, Go ahead
 986                          * and send the head node to run_one_delayed_ref,
 987                          * so that any accounting fixes can happen
 988                          */
 989                         ref = &locked_ref->node;
 990                         list_del_init(&locked_ref->cluster);
 991                         locked_ref = NULL;
 992                 }
 993
 994                 ref->in_tree = 0;
 995                 rb_erase(&ref->rb_node, &delayed_refs->root);
 996                 delayed_refs->num_entries--;
 997                 spin_unlock(&delayed_refs->lock);
 998
 999                 ret = run_one_delayed_ref(trans, root, ref,
1000                                           must_insert_reserved);
1001                 BUG_ON(ret);
1002                 btrfs_put_delayed_ref(ref);
1003
1004                 count++;
1005                 cond_resched();
1006                 spin_lock(&delayed_refs->lock);
1007         }
1008         return count;
1009 }
1010
1011 /*
1012  * this starts processing the delayed reference count updates and
1013  * extent insertions we have queued up so far.  count can be
1014  * 0, which means to process everything in the tree at the start
1015  * of the run (but not newly added entries), or it can be some target
1016  * number you'd like to process.
1017  */
1018 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1019                            struct btrfs_root *root, unsigned long count)
1020 {
1021         struct rb_node *node;
1022         struct btrfs_delayed_ref_root *delayed_refs;
1023         struct btrfs_delayed_ref_node *ref;
1024         struct list_head cluster;
1025         int ret;
1026         int run_all = count == (unsigned long)-1;
1027         int run_most = 0;
1028
1029         if (root == root->fs_info->extent_root)
1030                 root = root->fs_info->tree_root;
1031
1032         delayed_refs = &trans->transaction->delayed_refs;
1033         INIT_LIST_HEAD(&cluster);
1034 again:
1035         spin_lock(&delayed_refs->lock);
1036         if (count == 0) {
1037                 count = delayed_refs->num_entries * 2;
1038                 run_most = 1;
1039         }
1040         while (1) {
1041                 if (!(run_all || run_most) &&
1042                     delayed_refs->num_heads_ready < 64)
1043                         break;
1044
1045                 /*
1046                  * go find something we can process in the rbtree.  We start at
1047                  * the beginning of the tree, and then build a cluster
1048                  * of refs to process starting at the first one we are able to
1049                  * lock
1050                  */
1051                 ret = btrfs_find_ref_cluster(trans, &cluster,
1052                                              delayed_refs->run_delayed_start);
1053                 if (ret)
1054                         break;
1055
1056                 ret = run_clustered_refs(trans, root, &cluster);
1057                 BUG_ON(ret < 0);
1058
1059                 count -= min_t(unsigned long, ret, count);
1060
1061                 if (count == 0)
1062                         break;
1063         }
1064
1065         if (run_all) {
1066                 node = rb_first(&delayed_refs->root);
1067                 if (!node)
1068                         goto out;
1069                 count = (unsigned long)-1;
1070
1071                 while (node) {
1072                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
1073                                        rb_node);
1074                         if (btrfs_delayed_ref_is_head(ref)) {
1075                                 struct btrfs_delayed_ref_head *head;
1076
1077                                 head = btrfs_delayed_node_to_head(ref);
1078                                 atomic_inc(&ref->refs);
1079
1080                                 spin_unlock(&delayed_refs->lock);
1081                                 mutex_lock(&head->mutex);
1082                                 mutex_unlock(&head->mutex);
1083
1084                                 btrfs_put_delayed_ref(ref);
1085                                 cond_resched();
1086                                 goto again;
1087                         }
1088                         node = rb_next(node);
1089                 }
1090                 spin_unlock(&delayed_refs->lock);
1091                 schedule_timeout(1);
1092                 goto again;
1093         }
1094 out:
1095         spin_unlock(&delayed_refs->lock);
1096         return 0;
1097 }
1098
1099 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1100                           struct btrfs_root *root, u64 objectid, u64 bytenr)
1101 {
1102         struct btrfs_root *extent_root = root->fs_info->extent_root;
1103         struct btrfs_path *path;
1104         struct extent_buffer *leaf;
1105         struct btrfs_extent_ref *ref_item;
1106         struct btrfs_key key;
1107         struct btrfs_key found_key;
1108         u64 ref_root;
1109         u64 last_snapshot;
1110         u32 nritems;
1111         int ret;
1112
1113         key.objectid = bytenr;
1114         key.offset = (u64)-1;
1115         key.type = BTRFS_EXTENT_ITEM_KEY;
1116
1117         path = btrfs_alloc_path();
1118         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1119         if (ret < 0)
1120                 goto out;
1121         BUG_ON(ret == 0);
1122
1123         ret = -ENOENT;
1124         if (path->slots[0] == 0)
1125                 goto out;
1126
1127         path->slots[0]--;
1128         leaf = path->nodes[0];
1129         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1130
1131         if (found_key.objectid != bytenr ||
1132             found_key.type != BTRFS_EXTENT_ITEM_KEY)
1133                 goto out;
1134
1135         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1136         while (1) {
1137                 leaf = path->nodes[0];
1138                 nritems = btrfs_header_nritems(leaf);
1139                 if (path->slots[0] >= nritems) {
1140                         ret = btrfs_next_leaf(extent_root, path);
1141                         if (ret < 0)
1142                                 goto out;
1143                         if (ret == 0)
1144                                 continue;
1145                         break;
1146                 }
1147                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1148                 if (found_key.objectid != bytenr)
1149                         break;
1150
1151                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1152                         path->slots[0]++;
1153                         continue;
1154                 }
1155
1156                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1157                                           struct btrfs_extent_ref);
1158                 ref_root = btrfs_ref_root(leaf, ref_item);
1159                 if ((ref_root != root->root_key.objectid &&
1160                      ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1161                      objectid != btrfs_ref_objectid(leaf, ref_item)) {
1162                         ret = 1;
1163                         goto out;
1164                 }
1165                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1166                         ret = 1;
1167                         goto out;
1168                 }
1169
1170                 path->slots[0]++;
1171         }
1172         ret = 0;
1173 out:
1174         btrfs_free_path(path);
1175         return ret;
1176 }
1177
1178 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1179                     struct extent_buffer *buf, u32 nr_extents)
1180 {
1181         struct btrfs_key key;
1182         struct btrfs_file_extent_item *fi;
1183         u64 root_gen;
1184         u32 nritems;
1185         int i;
1186         int level;
1187         int ret = 0;
1188         int shared = 0;
1189
1190         if (!root->ref_cows)
1191                 return 0;
1192
1193         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1194                 shared = 0;
1195                 root_gen = root->root_key.offset;
1196         } else {
1197                 shared = 1;
1198                 root_gen = trans->transid - 1;
1199         }
1200
1201         level = btrfs_header_level(buf);
1202         nritems = btrfs_header_nritems(buf);
1203
1204         if (level == 0) {
1205                 struct btrfs_leaf_ref *ref;
1206                 struct btrfs_extent_info *info;
1207
1208                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1209                 if (!ref) {
1210                         ret = -ENOMEM;
1211                         goto out;
1212                 }
1213
1214                 ref->root_gen = root_gen;
1215                 ref->bytenr = buf->start;
1216                 ref->owner = btrfs_header_owner(buf);
1217                 ref->generation = btrfs_header_generation(buf);
1218                 ref->nritems = nr_extents;
1219                 info = ref->extents;
1220
1221                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1222                         u64 disk_bytenr;
1223                         btrfs_item_key_to_cpu(buf, &key, i);
1224                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1225                                 continue;
1226                         fi = btrfs_item_ptr(buf, i,
1227                                             struct btrfs_file_extent_item);
1228                         if (btrfs_file_extent_type(buf, fi) ==
1229                             BTRFS_FILE_EXTENT_INLINE)
1230                                 continue;
1231                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1232                         if (disk_bytenr == 0)
1233                                 continue;
1234
1235                         info->bytenr = disk_bytenr;
1236                         info->num_bytes =
1237                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1238                         info->objectid = key.objectid;
1239                         info->offset = key.offset;
1240                         info++;
1241                 }
1242
1243                 ret = btrfs_add_leaf_ref(root, ref, shared);
1244                 if (ret == -EEXIST && shared) {
1245                         struct btrfs_leaf_ref *old;
1246                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1247                         BUG_ON(!old);
1248                         btrfs_remove_leaf_ref(root, old);
1249                         btrfs_free_leaf_ref(root, old);
1250                         ret = btrfs_add_leaf_ref(root, ref, shared);
1251                 }
1252                 WARN_ON(ret);
1253                 btrfs_free_leaf_ref(root, ref);
1254         }
1255 out:
1256         return ret;
1257 }
1258
1259 /* when a block goes through cow, we update the reference counts of
1260  * everything that block points to.  The internal pointers of the block
1261  * can be in just about any order, and it is likely to have clusters of
1262  * things that are close together and clusters of things that are not.
1263  *
1264  * To help reduce the seeks that come with updating all of these reference
1265  * counts, sort them by byte number before actual updates are done.
1266  *
1267  * struct refsort is used to match byte number to slot in the btree block.
1268  * we sort based on the byte number and then use the slot to actually
1269  * find the item.
1270  *
1271  * struct refsort is smaller than strcut btrfs_item and smaller than
1272  * struct btrfs_key_ptr.  Since we're currently limited to the page size
1273  * for a btree block, there's no way for a kmalloc of refsorts for a
1274  * single node to be bigger than a page.
1275  */
1276 struct refsort {
1277         u64 bytenr;
1278         u32 slot;
1279 };
1280
1281 /*
1282  * for passing into sort()
1283  */
1284 static int refsort_cmp(const void *a_void, const void *b_void)
1285 {
1286         const struct refsort *a = a_void;
1287         const struct refsort *b = b_void;
1288
1289         if (a->bytenr < b->bytenr)
1290                 return -1;
1291         if (a->bytenr > b->bytenr)
1292                 return 1;
1293         return 0;
1294 }
1295
1296
1297 noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1298                            struct btrfs_root *root,
1299                            struct extent_buffer *orig_buf,
1300                            struct extent_buffer *buf, u32 *nr_extents)
1301 {
1302         u64 bytenr;
1303         u64 ref_root;
1304         u64 orig_root;
1305         u64 ref_generation;
1306         u64 orig_generation;
1307         struct refsort *sorted;
1308         u32 nritems;
1309         u32 nr_file_extents = 0;
1310         struct btrfs_key key;
1311         struct btrfs_file_extent_item *fi;
1312         int i;
1313         int level;
1314         int ret = 0;
1315         int faili = 0;
1316         int refi = 0;
1317         int slot;
1318         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1319                             u64, u64, u64, u64, u64, u64, u64, u64, u64);
1320
1321         ref_root = btrfs_header_owner(buf);
1322         ref_generation = btrfs_header_generation(buf);
1323         orig_root = btrfs_header_owner(orig_buf);
1324         orig_generation = btrfs_header_generation(orig_buf);
1325
1326         nritems = btrfs_header_nritems(buf);
1327         level = btrfs_header_level(buf);
1328
1329         sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS);
1330         BUG_ON(!sorted);
1331
1332         if (root->ref_cows) {
1333                 process_func = __btrfs_inc_extent_ref;
1334         } else {
1335                 if (level == 0 &&
1336                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1337                         goto out;
1338                 if (level != 0 &&
1339                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1340                         goto out;
1341                 process_func = __btrfs_update_extent_ref;
1342         }
1343
1344         /*
1345          * we make two passes through the items.  In the first pass we
1346          * only record the byte number and slot.  Then we sort based on
1347          * byte number and do the actual work based on the sorted results
1348          */
1349         for (i = 0; i < nritems; i++) {
1350                 cond_resched();
1351                 if (level == 0) {
1352                         btrfs_item_key_to_cpu(buf, &key, i);
1353                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1354                                 continue;
1355                         fi = btrfs_item_ptr(buf, i,
1356                                             struct btrfs_file_extent_item);
1357                         if (btrfs_file_extent_type(buf, fi) ==
1358                             BTRFS_FILE_EXTENT_INLINE)
1359                                 continue;
1360                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1361                         if (bytenr == 0)
1362                                 continue;
1363
1364                         nr_file_extents++;
1365                         sorted[refi].bytenr = bytenr;
1366                         sorted[refi].slot = i;
1367                         refi++;
1368                 } else {
1369                         bytenr = btrfs_node_blockptr(buf, i);
1370                         sorted[refi].bytenr = bytenr;
1371                         sorted[refi].slot = i;
1372                         refi++;
1373                 }
1374         }
1375         /*
1376          * if refi == 0, we didn't actually put anything into the sorted
1377          * array and we're done
1378          */
1379         if (refi == 0)
1380                 goto out;
1381
1382         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
1383
1384         for (i = 0; i < refi; i++) {
1385                 cond_resched();
1386                 slot = sorted[i].slot;
1387                 bytenr = sorted[i].bytenr;
1388
1389                 if (level == 0) {
1390                         btrfs_item_key_to_cpu(buf, &key, slot);
1391                         fi = btrfs_item_ptr(buf, slot,
1392                                             struct btrfs_file_extent_item);
1393
1394                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1395                         if (bytenr == 0)
1396                                 continue;
1397
1398                         ret = process_func(trans, root, bytenr,
1399                                    btrfs_file_extent_disk_num_bytes(buf, fi),
1400                                    orig_buf->start, buf->start,
1401                                    orig_root, ref_root,
1402                                    orig_generation, ref_generation,
1403                                    key.objectid);
1404
1405                         if (ret) {
1406                                 faili = slot;
1407                                 WARN_ON(1);
1408                                 goto fail;
1409                         }
1410                 } else {
1411                         ret = process_func(trans, root, bytenr, buf->len,
1412                                            orig_buf->start, buf->start,
1413                                            orig_root, ref_root,
1414                                            orig_generation, ref_generation,
1415                                            level - 1);
1416                         if (ret) {
1417                                 faili = slot;
1418                                 WARN_ON(1);
1419                                 goto fail;
1420                         }
1421                 }
1422         }
1423 out:
1424         kfree(sorted);
1425         if (nr_extents) {
1426                 if (level == 0)
1427                         *nr_extents = nr_file_extents;
1428                 else
1429                         *nr_extents = nritems;
1430         }
1431         return 0;
1432 fail:
1433         kfree(sorted);
1434         WARN_ON(1);
1435         return ret;
1436 }
1437
1438 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1439                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1440                      struct extent_buffer *buf, int start_slot, int nr)
1441
1442 {
1443         u64 bytenr;
1444         u64 ref_root;
1445         u64 orig_root;
1446         u64 ref_generation;
1447         u64 orig_generation;
1448         struct btrfs_key key;
1449         struct btrfs_file_extent_item *fi;
1450         int i;
1451         int ret;
1452         int slot;
1453         int level;
1454
1455         BUG_ON(start_slot < 0);
1456         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1457
1458         ref_root = btrfs_header_owner(buf);
1459         ref_generation = btrfs_header_generation(buf);
1460         orig_root = btrfs_header_owner(orig_buf);
1461         orig_generation = btrfs_header_generation(orig_buf);
1462         level = btrfs_header_level(buf);
1463
1464         if (!root->ref_cows) {
1465                 if (level == 0 &&
1466                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1467                         return 0;
1468                 if (level != 0 &&
1469                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1470                         return 0;
1471         }
1472
1473         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1474                 cond_resched();
1475                 if (level == 0) {
1476                         btrfs_item_key_to_cpu(buf, &key, slot);
1477                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1478                                 continue;
1479                         fi = btrfs_item_ptr(buf, slot,
1480                                             struct btrfs_file_extent_item);
1481                         if (btrfs_file_extent_type(buf, fi) ==
1482                             BTRFS_FILE_EXTENT_INLINE)
1483                                 continue;
1484                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1485                         if (bytenr == 0)
1486                                 continue;
1487                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1488                                     btrfs_file_extent_disk_num_bytes(buf, fi),
1489                                     orig_buf->start, buf->start,
1490                                     orig_root, ref_root, orig_generation,
1491                                     ref_generation, key.objectid);
1492                         if (ret)
1493                                 goto fail;
1494                 } else {
1495                         bytenr = btrfs_node_blockptr(buf, slot);
1496                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1497                                             buf->len, orig_buf->start,
1498                                             buf->start, orig_root, ref_root,
1499                                             orig_generation, ref_generation,
1500                                             level - 1);
1501                         if (ret)
1502                                 goto fail;
1503                 }
1504         }
1505         return 0;
1506 fail:
1507         WARN_ON(1);
1508         return -1;
1509 }
1510
1511 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1512                                  struct btrfs_root *root,
1513                                  struct btrfs_path *path,
1514                                  struct btrfs_block_group_cache *cache)
1515 {
1516         int ret;
1517         struct btrfs_root *extent_root = root->fs_info->extent_root;
1518         unsigned long bi;
1519         struct extent_buffer *leaf;
1520
1521         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1522         if (ret < 0)
1523                 goto fail;
1524         BUG_ON(ret);
1525
1526         leaf = path->nodes[0];
1527         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1528         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1529         btrfs_mark_buffer_dirty(leaf);
1530         btrfs_release_path(extent_root, path);
1531 fail:
1532         if (ret)
1533                 return ret;
1534         return 0;
1535
1536 }
1537
1538 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1539                                    struct btrfs_root *root)
1540 {
1541         struct btrfs_block_group_cache *cache, *entry;
1542         struct rb_node *n;
1543         int err = 0;
1544         int werr = 0;
1545         struct btrfs_path *path;
1546         u64 last = 0;
1547
1548         path = btrfs_alloc_path();
1549         if (!path)
1550                 return -ENOMEM;
1551
1552         while (1) {
1553                 cache = NULL;
1554                 spin_lock(&root->fs_info->block_group_cache_lock);
1555                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1556                      n; n = rb_next(n)) {
1557                         entry = rb_entry(n, struct btrfs_block_group_cache,
1558                                          cache_node);
1559                         if (entry->dirty) {
1560                                 cache = entry;
1561                                 break;
1562                         }
1563                 }
1564                 spin_unlock(&root->fs_info->block_group_cache_lock);
1565
1566                 if (!cache)
1567                         break;
1568
1569                 cache->dirty = 0;
1570                 last += cache->key.offset;
1571
1572                 err = write_one_cache_group(trans, root,
1573                                             path, cache);
1574                 /*
1575                  * if we fail to write the cache group, we want
1576                  * to keep it marked dirty in hopes that a later
1577                  * write will work
1578                  */
1579                 if (err) {
1580                         werr = err;
1581                         continue;
1582                 }
1583         }
1584         btrfs_free_path(path);
1585         return werr;
1586 }
1587
1588 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1589 {
1590         struct btrfs_block_group_cache *block_group;
1591         int readonly = 0;
1592
1593         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1594         if (!block_group || block_group->ro)
1595                 readonly = 1;
1596         if (block_group)
1597                 put_block_group(block_group);
1598         return readonly;
1599 }
1600
1601 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1602                              u64 total_bytes, u64 bytes_used,
1603                              struct btrfs_space_info **space_info)
1604 {
1605         struct btrfs_space_info *found;
1606
1607         found = __find_space_info(info, flags);
1608         if (found) {
1609                 spin_lock(&found->lock);
1610                 found->total_bytes += total_bytes;
1611                 found->bytes_used += bytes_used;
1612                 found->full = 0;
1613                 spin_unlock(&found->lock);
1614                 *space_info = found;
1615                 return 0;
1616         }
1617         found = kzalloc(sizeof(*found), GFP_NOFS);
1618         if (!found)
1619                 return -ENOMEM;
1620
1621         INIT_LIST_HEAD(&found->block_groups);
1622         init_rwsem(&found->groups_sem);
1623         spin_lock_init(&found->lock);
1624         found->flags = flags;
1625         found->total_bytes = total_bytes;
1626         found->bytes_used = bytes_used;
1627         found->bytes_pinned = 0;
1628         found->bytes_reserved = 0;
1629         found->bytes_readonly = 0;
1630         found->bytes_delalloc = 0;
1631         found->full = 0;
1632         found->force_alloc = 0;
1633         *space_info = found;
1634         list_add_rcu(&found->list, &info->space_info);
1635         return 0;
1636 }
1637
1638 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1639 {
1640         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1641                                    BTRFS_BLOCK_GROUP_RAID1 |
1642                                    BTRFS_BLOCK_GROUP_RAID10 |
1643                                    BTRFS_BLOCK_GROUP_DUP);
1644         if (extra_flags) {
1645                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1646                         fs_info->avail_data_alloc_bits |= extra_flags;
1647                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1648                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1649                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1650                         fs_info->avail_system_alloc_bits |= extra_flags;
1651         }
1652 }
1653
1654 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1655 {
1656         spin_lock(&cache->space_info->lock);
1657         spin_lock(&cache->lock);
1658         if (!cache->ro) {
1659                 cache->space_info->bytes_readonly += cache->key.offset -
1660                                         btrfs_block_group_used(&cache->item);
1661                 cache->ro = 1;
1662         }
1663         spin_unlock(&cache->lock);
1664         spin_unlock(&cache->space_info->lock);
1665 }
1666
1667 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1668 {
1669         u64 num_devices = root->fs_info->fs_devices->rw_devices;
1670
1671         if (num_devices == 1)
1672                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1673         if (num_devices < 4)
1674                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1675
1676         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1677             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1678                       BTRFS_BLOCK_GROUP_RAID10))) {
1679                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1680         }
1681
1682         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1683             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1684                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1685         }
1686
1687         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1688             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1689              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1690              (flags & BTRFS_BLOCK_GROUP_DUP)))
1691                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1692         return flags;
1693 }
1694
1695 static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data)
1696 {
1697         struct btrfs_fs_info *info = root->fs_info;
1698         u64 alloc_profile;
1699
1700         if (data) {
1701                 alloc_profile = info->avail_data_alloc_bits &
1702                         info->data_alloc_profile;
1703                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1704         } else if (root == root->fs_info->chunk_root) {
1705                 alloc_profile = info->avail_system_alloc_bits &
1706                         info->system_alloc_profile;
1707                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1708         } else {
1709                 alloc_profile = info->avail_metadata_alloc_bits &
1710                         info->metadata_alloc_profile;
1711                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1712         }
1713
1714         return btrfs_reduce_alloc_profile(root, data);
1715 }
1716
1717 void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode)
1718 {
1719         u64 alloc_target;
1720
1721         alloc_target = btrfs_get_alloc_profile(root, 1);
1722         BTRFS_I(inode)->space_info = __find_space_info(root->fs_info,
1723                                                        alloc_target);
1724 }
1725
1726 /*
1727  * for now this just makes sure we have at least 5% of our metadata space free
1728  * for use.
1729  */
1730 int btrfs_check_metadata_free_space(struct btrfs_root *root)
1731 {
1732         struct btrfs_fs_info *info = root->fs_info;
1733         struct btrfs_space_info *meta_sinfo;
1734         u64 alloc_target, thresh;
1735         int committed = 0, ret;
1736
1737         /* get the space info for where the metadata will live */
1738         alloc_target = btrfs_get_alloc_profile(root, 0);
1739         meta_sinfo = __find_space_info(info, alloc_target);
1740
1741 again:
1742         spin_lock(&meta_sinfo->lock);
1743         if (!meta_sinfo->full)
1744                 thresh = meta_sinfo->total_bytes * 80;
1745         else
1746                 thresh = meta_sinfo->total_bytes * 95;
1747
1748         do_div(thresh, 100);
1749
1750         if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved +
1751             meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) {
1752                 struct btrfs_trans_handle *trans;
1753                 if (!meta_sinfo->full) {
1754                         meta_sinfo->force_alloc = 1;
1755                         spin_unlock(&meta_sinfo->lock);
1756
1757                         trans = btrfs_start_transaction(root, 1);
1758                         if (!trans)
1759                                 return -ENOMEM;
1760
1761                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1762                                              2 * 1024 * 1024, alloc_target, 0);
1763                         btrfs_end_transaction(trans, root);
1764                         goto again;
1765                 }
1766                 spin_unlock(&meta_sinfo->lock);
1767
1768                 if (!committed) {
1769                         committed = 1;
1770                         trans = btrfs_join_transaction(root, 1);
1771                         if (!trans)
1772                                 return -ENOMEM;
1773                         ret = btrfs_commit_transaction(trans, root);
1774                         if (ret)
1775                                 return ret;
1776                         goto again;
1777                 }
1778                 return -ENOSPC;
1779         }
1780         spin_unlock(&meta_sinfo->lock);
1781
1782         return 0;
1783 }
1784
1785 /*
1786  * This will check the space that the inode allocates from to make sure we have
1787  * enough space for bytes.
1788  */
1789 int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
1790                                 u64 bytes)
1791 {
1792         struct btrfs_space_info *data_sinfo;
1793         int ret = 0, committed = 0;
1794
1795         /* make sure bytes are sectorsize aligned */
1796         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
1797
1798         data_sinfo = BTRFS_I(inode)->space_info;
1799 again:
1800         /* make sure we have enough space to handle the data first */
1801         spin_lock(&data_sinfo->lock);
1802         if (data_sinfo->total_bytes - data_sinfo->bytes_used -
1803             data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved -
1804             data_sinfo->bytes_pinned - data_sinfo->bytes_readonly -
1805             data_sinfo->bytes_may_use < bytes) {
1806                 struct btrfs_trans_handle *trans;
1807
1808                 /*
1809                  * if we don't have enough free bytes in this space then we need
1810                  * to alloc a new chunk.
1811                  */
1812                 if (!data_sinfo->full) {
1813                         u64 alloc_target;
1814
1815                         data_sinfo->force_alloc = 1;
1816                         spin_unlock(&data_sinfo->lock);
1817
1818                         alloc_target = btrfs_get_alloc_profile(root, 1);
1819                         trans = btrfs_start_transaction(root, 1);
1820                         if (!trans)
1821                                 return -ENOMEM;
1822
1823                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1824                                              bytes + 2 * 1024 * 1024,
1825                                              alloc_target, 0);
1826                         btrfs_end_transaction(trans, root);
1827                         if (ret)
1828                                 return ret;
1829                         goto again;
1830                 }
1831                 spin_unlock(&data_sinfo->lock);
1832
1833                 /* commit the current transaction and try again */
1834                 if (!committed) {
1835                         committed = 1;
1836                         trans = btrfs_join_transaction(root, 1);
1837                         if (!trans)
1838                                 return -ENOMEM;
1839                         ret = btrfs_commit_transaction(trans, root);
1840                         if (ret)
1841                                 return ret;
1842                         goto again;
1843                 }
1844
1845                 printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
1846                        ", %llu bytes_used, %llu bytes_reserved, "
1847                        "%llu bytes_pinned, %llu bytes_readonly, %llu may use"
1848                        "%llu total\n", bytes, data_sinfo->bytes_delalloc,
1849                        data_sinfo->bytes_used, data_sinfo->bytes_reserved,
1850                        data_sinfo->bytes_pinned, data_sinfo->bytes_readonly,
1851                        data_sinfo->bytes_may_use, data_sinfo->total_bytes);
1852                 return -ENOSPC;
1853         }
1854         data_sinfo->bytes_may_use += bytes;
1855         BTRFS_I(inode)->reserved_bytes += bytes;
1856         spin_unlock(&data_sinfo->lock);
1857
1858         return btrfs_check_metadata_free_space(root);
1859 }
1860
1861 /*
1862  * if there was an error for whatever reason after calling
1863  * btrfs_check_data_free_space, call this so we can cleanup the counters.
1864  */
1865 void btrfs_free_reserved_data_space(struct btrfs_root *root,
1866                                     struct inode *inode, u64 bytes)
1867 {
1868         struct btrfs_space_info *data_sinfo;
1869
1870         /* make sure bytes are sectorsize aligned */
1871         bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
1872
1873         data_sinfo = BTRFS_I(inode)->space_info;
1874         spin_lock(&data_sinfo->lock);
1875         data_sinfo->bytes_may_use -= bytes;
1876         BTRFS_I(inode)->reserved_bytes -= bytes;
1877         spin_unlock(&data_sinfo->lock);
1878 }
1879
1880 /* called when we are adding a delalloc extent to the inode's io_tree */
1881 void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
1882                                   u64 bytes)
1883 {
1884         struct btrfs_space_info *data_sinfo;
1885
1886         /* get the space info for where this inode will be storing its data */
1887         data_sinfo = BTRFS_I(inode)->space_info;
1888
1889         /* make sure we have enough space to handle the data first */
1890         spin_lock(&data_sinfo->lock);
1891         data_sinfo->bytes_delalloc += bytes;
1892
1893         /*
1894          * we are adding a delalloc extent without calling
1895          * btrfs_check_data_free_space first.  This happens on a weird
1896          * writepage condition, but shouldn't hurt our accounting
1897          */
1898         if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) {
1899                 data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes;
1900                 BTRFS_I(inode)->reserved_bytes = 0;
1901         } else {
1902                 data_sinfo->bytes_may_use -= bytes;
1903                 BTRFS_I(inode)->reserved_bytes -= bytes;
1904         }
1905
1906         spin_unlock(&data_sinfo->lock);
1907 }
1908
1909 /* called when we are clearing an delalloc extent from the inode's io_tree */
1910 void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
1911                               u64 bytes)
1912 {
1913         struct btrfs_space_info *info;
1914
1915         info = BTRFS_I(inode)->space_info;
1916
1917         spin_lock(&info->lock);
1918         info->bytes_delalloc -= bytes;
1919         spin_unlock(&info->lock);
1920 }
1921
1922 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1923                           struct btrfs_root *extent_root, u64 alloc_bytes,
1924                           u64 flags, int force)
1925 {
1926         struct btrfs_space_info *space_info;
1927         u64 thresh;
1928         int ret = 0;
1929
1930         mutex_lock(&extent_root->fs_info->chunk_mutex);
1931
1932         flags = btrfs_reduce_alloc_profile(extent_root, flags);
1933
1934         space_info = __find_space_info(extent_root->fs_info, flags);
1935         if (!space_info) {
1936                 ret = update_space_info(extent_root->fs_info, flags,
1937                                         0, 0, &space_info);
1938                 BUG_ON(ret);
1939         }
1940         BUG_ON(!space_info);
1941
1942         spin_lock(&space_info->lock);
1943         if (space_info->force_alloc) {
1944                 force = 1;
1945                 space_info->force_alloc = 0;
1946         }
1947         if (space_info->full) {
1948                 spin_unlock(&space_info->lock);
1949                 goto out;
1950         }
1951
1952         thresh = space_info->total_bytes - space_info->bytes_readonly;
1953         thresh = div_factor(thresh, 6);
1954         if (!force &&
1955            (space_info->bytes_used + space_info->bytes_pinned +
1956             space_info->bytes_reserved + alloc_bytes) < thresh) {
1957                 spin_unlock(&space_info->lock);
1958                 goto out;
1959         }
1960         spin_unlock(&space_info->lock);
1961
1962         ret = btrfs_alloc_chunk(trans, extent_root, flags);
1963         if (ret)
1964                 space_info->full = 1;
1965 out:
1966         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1967         return ret;
1968 }
1969
1970 static int update_block_group(struct btrfs_trans_handle *trans,
1971                               struct btrfs_root *root,
1972                               u64 bytenr, u64 num_bytes, int alloc,
1973                               int mark_free)
1974 {
1975         struct btrfs_block_group_cache *cache;
1976         struct btrfs_fs_info *info = root->fs_info;
1977         u64 total = num_bytes;
1978         u64 old_val;
1979         u64 byte_in_group;
1980
1981         while (total) {
1982                 cache = btrfs_lookup_block_group(info, bytenr);
1983                 if (!cache)
1984                         return -1;
1985                 byte_in_group = bytenr - cache->key.objectid;
1986                 WARN_ON(byte_in_group > cache->key.offset);
1987
1988                 spin_lock(&cache->space_info->lock);
1989                 spin_lock(&cache->lock);
1990                 cache->dirty = 1;
1991                 old_val = btrfs_block_group_used(&cache->item);
1992                 num_bytes = min(total, cache->key.offset - byte_in_group);
1993                 if (alloc) {
1994                         old_val += num_bytes;
1995                         cache->space_info->bytes_used += num_bytes;
1996                         if (cache->ro)
1997                                 cache->space_info->bytes_readonly -= num_bytes;
1998                         btrfs_set_block_group_used(&cache->item, old_val);
1999                         spin_unlock(&cache->lock);
2000                         spin_unlock(&cache->space_info->lock);
2001                 } else {
2002                         old_val -= num_bytes;
2003                         cache->space_info->bytes_used -= num_bytes;
2004                         if (cache->ro)
2005                                 cache->space_info->bytes_readonly += num_bytes;
2006                         btrfs_set_block_group_used(&cache->item, old_val);
2007                         spin_unlock(&cache->lock);
2008                         spin_unlock(&cache->space_info->lock);
2009                         if (mark_free) {
2010                                 int ret;
2011
2012                                 ret = btrfs_discard_extent(root, bytenr,
2013                                                            num_bytes);
2014                                 WARN_ON(ret);
2015
2016                                 ret = btrfs_add_free_space(cache, bytenr,
2017                                                            num_bytes);
2018                                 WARN_ON(ret);
2019                         }
2020                 }
2021                 put_block_group(cache);
2022                 total -= num_bytes;
2023                 bytenr += num_bytes;
2024         }
2025         return 0;
2026 }
2027
2028 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2029 {
2030         struct btrfs_block_group_cache *cache;
2031         u64 bytenr;
2032
2033         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2034         if (!cache)
2035                 return 0;
2036
2037         bytenr = cache->key.objectid;
2038         put_block_group(cache);
2039
2040         return bytenr;
2041 }
2042
2043 int btrfs_update_pinned_extents(struct btrfs_root *root,
2044                                 u64 bytenr, u64 num, int pin)
2045 {
2046         u64 len;
2047         struct btrfs_block_group_cache *cache;
2048         struct btrfs_fs_info *fs_info = root->fs_info;
2049
2050         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2051         if (pin) {
2052                 set_extent_dirty(&fs_info->pinned_extents,
2053                                 bytenr, bytenr + num - 1, GFP_NOFS);
2054         } else {
2055                 clear_extent_dirty(&fs_info->pinned_extents,
2056                                 bytenr, bytenr + num - 1, GFP_NOFS);
2057         }
2058         mutex_unlock(&root->fs_info->pinned_mutex);
2059
2060         while (num > 0) {
2061                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2062                 BUG_ON(!cache);
2063                 len = min(num, cache->key.offset -
2064                           (bytenr - cache->key.objectid));
2065                 if (pin) {
2066                         spin_lock(&cache->space_info->lock);
2067                         spin_lock(&cache->lock);
2068                         cache->pinned += len;
2069                         cache->space_info->bytes_pinned += len;
2070                         spin_unlock(&cache->lock);
2071                         spin_unlock(&cache->space_info->lock);
2072                         fs_info->total_pinned += len;
2073                 } else {
2074                         spin_lock(&cache->space_info->lock);
2075                         spin_lock(&cache->lock);
2076                         cache->pinned -= len;
2077                         cache->space_info->bytes_pinned -= len;
2078                         spin_unlock(&cache->lock);
2079                         spin_unlock(&cache->space_info->lock);
2080                         fs_info->total_pinned -= len;
2081                         if (cache->cached)
2082                                 btrfs_add_free_space(cache, bytenr, len);
2083                 }
2084                 put_block_group(cache);
2085                 bytenr += len;
2086                 num -= len;
2087         }
2088         return 0;
2089 }
2090
2091 static int update_reserved_extents(struct btrfs_root *root,
2092                                    u64 bytenr, u64 num, int reserve)
2093 {
2094         u64 len;
2095         struct btrfs_block_group_cache *cache;
2096         struct btrfs_fs_info *fs_info = root->fs_info;
2097
2098         while (num > 0) {
2099                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2100                 BUG_ON(!cache);
2101                 len = min(num, cache->key.offset -
2102                           (bytenr - cache->key.objectid));
2103
2104                 spin_lock(&cache->space_info->lock);
2105                 spin_lock(&cache->lock);
2106                 if (reserve) {
2107                         cache->reserved += len;
2108                         cache->space_info->bytes_reserved += len;
2109                 } else {
2110                         cache->reserved -= len;
2111                         cache->space_info->bytes_reserved -= len;
2112                 }
2113                 spin_unlock(&cache->lock);
2114                 spin_unlock(&cache->space_info->lock);
2115                 put_block_group(cache);
2116                 bytenr += len;
2117                 num -= len;
2118         }
2119         return 0;
2120 }
2121
2122 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2123 {
2124         u64 last = 0;
2125         u64 start;
2126         u64 end;
2127         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2128         int ret;
2129
2130         mutex_lock(&root->fs_info->pinned_mutex);
2131         while (1) {
2132                 ret = find_first_extent_bit(pinned_extents, last,
2133                                             &start, &end, EXTENT_DIRTY);
2134                 if (ret)
2135                         break;
2136                 set_extent_dirty(copy, start, end, GFP_NOFS);
2137                 last = end + 1;
2138         }
2139         mutex_unlock(&root->fs_info->pinned_mutex);
2140         return 0;
2141 }
2142
2143 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2144                                struct btrfs_root *root,
2145                                struct extent_io_tree *unpin)
2146 {
2147         u64 start;
2148         u64 end;
2149         int ret;
2150
2151         while (1) {
2152                 mutex_lock(&root->fs_info->pinned_mutex);
2153                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2154                                             EXTENT_DIRTY);
2155                 if (ret)
2156                         break;
2157
2158                 ret = btrfs_discard_extent(root, start, end + 1 - start);
2159
2160                 /* unlocks the pinned mutex */
2161                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2162                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2163
2164                 cond_resched();
2165         }
2166         mutex_unlock(&root->fs_info->pinned_mutex);
2167         return ret;
2168 }
2169
2170 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2171                           struct btrfs_root *root,
2172                           struct btrfs_path *path,
2173                           u64 bytenr, u64 num_bytes, int is_data,
2174                           struct extent_buffer **must_clean)
2175 {
2176         int err = 0;
2177         struct extent_buffer *buf;
2178
2179         if (is_data)
2180                 goto pinit;
2181
2182         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2183         if (!buf)
2184                 goto pinit;
2185
2186         /* we can reuse a block if it hasn't been written
2187          * and it is from this transaction.  We can't
2188          * reuse anything from the tree log root because
2189          * it has tiny sub-transactions.
2190          */
2191         if (btrfs_buffer_uptodate(buf, 0) &&
2192             btrfs_try_tree_lock(buf)) {
2193                 u64 header_owner = btrfs_header_owner(buf);
2194                 u64 header_transid = btrfs_header_generation(buf);
2195                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2196                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2197                     header_owner != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2198                     header_transid == trans->transid &&
2199                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2200                         *must_clean = buf;
2201                         return 1;
2202                 }
2203                 btrfs_tree_unlock(buf);
2204         }
2205         free_extent_buffer(buf);
2206 pinit:
2207         btrfs_set_path_blocking(path);
2208         mutex_lock(&root->fs_info->pinned_mutex);
2209         /* unlocks the pinned mutex */
2210         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2211
2212         BUG_ON(err < 0);
2213         return 0;
2214 }
2215
2216 /*
2217  * remove an extent from the root, returns 0 on success
2218  */
2219 static int __free_extent(struct btrfs_trans_handle *trans,
2220                          struct btrfs_root *root,
2221                          u64 bytenr, u64 num_bytes, u64 parent,
2222                          u64 root_objectid, u64 ref_generation,
2223                          u64 owner_objectid, int pin, int mark_free,
2224                          int refs_to_drop)
2225 {
2226         struct btrfs_path *path;
2227         struct btrfs_key key;
2228         struct btrfs_fs_info *info = root->fs_info;
2229         struct btrfs_root *extent_root = info->extent_root;
2230         struct extent_buffer *leaf;
2231         int ret;
2232         int extent_slot = 0;
2233         int found_extent = 0;
2234         int num_to_del = 1;
2235         struct btrfs_extent_item *ei;
2236         u32 refs;
2237
2238         key.objectid = bytenr;
2239         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2240         key.offset = num_bytes;
2241         path = btrfs_alloc_path();
2242         if (!path)
2243                 return -ENOMEM;
2244
2245         path->reada = 1;
2246         path->leave_spinning = 1;
2247         ret = lookup_extent_backref(trans, extent_root, path,
2248                                     bytenr, parent, root_objectid,
2249                                     ref_generation, owner_objectid, 1);
2250         if (ret == 0) {
2251                 struct btrfs_key found_key;
2252                 extent_slot = path->slots[0];
2253                 while (extent_slot > 0) {
2254                         extent_slot--;
2255                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2256                                               extent_slot);
2257                         if (found_key.objectid != bytenr)
2258                                 break;
2259                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2260                             found_key.offset == num_bytes) {
2261                                 found_extent = 1;
2262                                 break;
2263                         }
2264                         if (path->slots[0] - extent_slot > 5)
2265                                 break;
2266                 }
2267                 if (!found_extent) {
2268                         ret = remove_extent_backref(trans, extent_root, path,
2269                                                     refs_to_drop);
2270                         BUG_ON(ret);
2271                         btrfs_release_path(extent_root, path);
2272                         path->leave_spinning = 1;
2273                         ret = btrfs_search_slot(trans, extent_root,
2274                                                 &key, path, -1, 1);
2275                         if (ret) {
2276                                 printk(KERN_ERR "umm, got %d back from search"
2277                                        ", was looking for %llu\n", ret,
2278                                        (unsigned long long)bytenr);
2279                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2280                         }
2281                         BUG_ON(ret);
2282                         extent_slot = path->slots[0];
2283                 }
2284         } else {
2285                 btrfs_print_leaf(extent_root, path->nodes[0]);
2286                 WARN_ON(1);
2287                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2288                        "parent %llu root %llu gen %llu owner %llu\n",
2289                        (unsigned long long)bytenr,
2290                        (unsigned long long)parent,
2291                        (unsigned long long)root_objectid,
2292                        (unsigned long long)ref_generation,
2293                        (unsigned long long)owner_objectid);
2294         }
2295
2296         leaf = path->nodes[0];
2297         ei = btrfs_item_ptr(leaf, extent_slot,
2298                             struct btrfs_extent_item);
2299         refs = btrfs_extent_refs(leaf, ei);
2300
2301         /*
2302          * we're not allowed to delete the extent item if there
2303          * are other delayed ref updates pending
2304          */
2305
2306         BUG_ON(refs < refs_to_drop);
2307         refs -= refs_to_drop;
2308         btrfs_set_extent_refs(leaf, ei, refs);
2309         btrfs_mark_buffer_dirty(leaf);
2310
2311         if (refs == 0 && found_extent &&
2312             path->slots[0] == extent_slot + 1) {
2313                 struct btrfs_extent_ref *ref;
2314                 ref = btrfs_item_ptr(leaf, path->slots[0],
2315                                      struct btrfs_extent_ref);
2316                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != refs_to_drop);
2317                 /* if the back ref and the extent are next to each other
2318                  * they get deleted below in one shot
2319                  */
2320                 path->slots[0] = extent_slot;
2321                 num_to_del = 2;
2322         } else if (found_extent) {
2323                 /* otherwise delete the extent back ref */
2324                 ret = remove_extent_backref(trans, extent_root, path,
2325                                             refs_to_drop);
2326                 BUG_ON(ret);
2327                 /* if refs are 0, we need to setup the path for deletion */
2328                 if (refs == 0) {
2329                         btrfs_release_path(extent_root, path);
2330                         path->leave_spinning = 1;
2331                         ret = btrfs_search_slot(trans, extent_root, &key, path,
2332                                                 -1, 1);
2333                         BUG_ON(ret);
2334                 }
2335         }
2336
2337         if (refs == 0) {
2338                 u64 super_used;
2339                 u64 root_used;
2340                 struct extent_buffer *must_clean = NULL;
2341
2342                 if (pin) {
2343                         ret = pin_down_bytes(trans, root, path,
2344                                 bytenr, num_bytes,
2345                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID,
2346                                 &must_clean);
2347                         if (ret > 0)
2348                                 mark_free = 1;
2349                         BUG_ON(ret < 0);
2350                 }
2351
2352                 /* block accounting for super block */
2353                 spin_lock(&info->delalloc_lock);
2354                 super_used = btrfs_super_bytes_used(&info->super_copy);
2355                 btrfs_set_super_bytes_used(&info->super_copy,
2356                                            super_used - num_bytes);
2357
2358                 /* block accounting for root item */
2359                 root_used = btrfs_root_used(&root->root_item);
2360                 btrfs_set_root_used(&root->root_item,
2361                                            root_used - num_bytes);
2362                 spin_unlock(&info->delalloc_lock);
2363
2364                 /*
2365                  * it is going to be very rare for someone to be waiting
2366                  * on the block we're freeing.  del_items might need to
2367                  * schedule, so rather than get fancy, just force it
2368                  * to blocking here
2369                  */
2370                 if (must_clean)
2371                         btrfs_set_lock_blocking(must_clean);
2372
2373                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2374                                       num_to_del);
2375                 BUG_ON(ret);
2376                 btrfs_release_path(extent_root, path);
2377
2378                 if (must_clean) {
2379                         clean_tree_block(NULL, root, must_clean);
2380                         btrfs_tree_unlock(must_clean);
2381                         free_extent_buffer(must_clean);
2382                 }
2383
2384                 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2385                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2386                         BUG_ON(ret);
2387                 } else {
2388                         invalidate_mapping_pages(info->btree_inode->i_mapping,
2389                              bytenr >> PAGE_CACHE_SHIFT,
2390                              (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
2391                 }
2392
2393                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2394                                          mark_free);
2395                 BUG_ON(ret);
2396         }
2397         btrfs_free_path(path);
2398         return ret;
2399 }
2400
2401 /*
2402  * remove an extent from the root, returns 0 on success
2403  */
2404 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2405                                         struct btrfs_root *root,
2406                                         u64 bytenr, u64 num_bytes, u64 parent,
2407                                         u64 root_objectid, u64 ref_generation,
2408                                         u64 owner_objectid, int pin,
2409                                         int refs_to_drop)
2410 {
2411         WARN_ON(num_bytes < root->sectorsize);
2412
2413         /*
2414          * if metadata always pin
2415          * if data pin when any transaction has committed this
2416          */
2417         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID ||
2418             ref_generation != trans->transid)
2419                 pin = 1;
2420
2421         if (ref_generation != trans->transid)
2422                 pin = 1;
2423
2424         return __free_extent(trans, root, bytenr, num_bytes, parent,
2425                             root_objectid, ref_generation,
2426                             owner_objectid, pin, pin == 0, refs_to_drop);
2427 }
2428
2429 /*
2430  * when we free an extent, it is possible (and likely) that we free the last
2431  * delayed ref for that extent as well.  This searches the delayed ref tree for
2432  * a given extent, and if there are no other delayed refs to be processed, it
2433  * removes it from the tree.
2434  */
2435 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
2436                                       struct btrfs_root *root, u64 bytenr)
2437 {
2438         struct btrfs_delayed_ref_head *head;
2439         struct btrfs_delayed_ref_root *delayed_refs;
2440         struct btrfs_delayed_ref_node *ref;
2441         struct rb_node *node;
2442         int ret;
2443
2444         delayed_refs = &trans->transaction->delayed_refs;
2445         spin_lock(&delayed_refs->lock);
2446         head = btrfs_find_delayed_ref_head(trans, bytenr);
2447         if (!head)
2448                 goto out;
2449
2450         node = rb_prev(&head->node.rb_node);
2451         if (!node)
2452                 goto out;
2453
2454         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2455
2456         /* there are still entries for this ref, we can't drop it */
2457         if (ref->bytenr == bytenr)
2458                 goto out;
2459
2460         /*
2461          * waiting for the lock here would deadlock.  If someone else has it
2462          * locked they are already in the process of dropping it anyway
2463          */
2464         if (!mutex_trylock(&head->mutex))
2465                 goto out;
2466
2467         /*
2468          * at this point we have a head with no other entries.  Go
2469          * ahead and process it.
2470          */
2471         head->node.in_tree = 0;
2472         rb_erase(&head->node.rb_node, &delayed_refs->root);
2473
2474         delayed_refs->num_entries--;
2475
2476         /*
2477          * we don't take a ref on the node because we're removing it from the
2478          * tree, so we just steal the ref the tree was holding.
2479          */
2480         delayed_refs->num_heads--;
2481         if (list_empty(&head->cluster))
2482                 delayed_refs->num_heads_ready--;
2483
2484         list_del_init(&head->cluster);
2485         spin_unlock(&delayed_refs->lock);
2486
2487         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
2488                                   &head->node, head->must_insert_reserved);
2489         BUG_ON(ret);
2490         btrfs_put_delayed_ref(&head->node);
2491         return 0;
2492 out:
2493         spin_unlock(&delayed_refs->lock);
2494         return 0;
2495 }
2496
2497 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2498                       struct btrfs_root *root,
2499                       u64 bytenr, u64 num_bytes, u64 parent,
2500                       u64 root_objectid, u64 ref_generation,
2501                       u64 owner_objectid, int pin)
2502 {
2503         int ret;
2504
2505         /*
2506          * tree log blocks never actually go into the extent allocation
2507          * tree, just update pinning info and exit early.
2508          *
2509          * data extents referenced by the tree log do need to have
2510          * their reference counts bumped.
2511          */
2512         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID &&
2513             owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2514                 mutex_lock(&root->fs_info->pinned_mutex);
2515
2516                 /* unlocks the pinned mutex */
2517                 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2518                 update_reserved_extents(root, bytenr, num_bytes, 0);
2519                 ret = 0;
2520         } else {
2521                 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent,
2522                                        root_objectid, ref_generation,
2523                                        owner_objectid,
2524                                        BTRFS_DROP_DELAYED_REF, 1);
2525                 BUG_ON(ret);
2526                 ret = check_ref_cleanup(trans, root, bytenr);
2527                 BUG_ON(ret);
2528         }
2529         return ret;
2530 }
2531
2532 static u64 stripe_align(struct btrfs_root *root, u64 val)
2533 {
2534         u64 mask = ((u64)root->stripesize - 1);
2535         u64 ret = (val + mask) & ~mask;
2536         return ret;
2537 }
2538
2539 /*
2540  * walks the btree of allocated extents and find a hole of a given size.
2541  * The key ins is changed to record the hole:
2542  * ins->objectid == block start
2543  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2544  * ins->offset == number of blocks
2545  * Any available blocks before search_start are skipped.
2546  */
2547 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2548                                      struct btrfs_root *orig_root,
2549                                      u64 num_bytes, u64 empty_size,
2550                                      u64 search_start, u64 search_end,
2551                                      u64 hint_byte, struct btrfs_key *ins,
2552                                      u64 exclude_start, u64 exclude_nr,
2553                                      int data)
2554 {
2555         int ret = 0;
2556         struct btrfs_root *root = orig_root->fs_info->extent_root;
2557         u64 *last_ptr = NULL;
2558         struct btrfs_block_group_cache *block_group = NULL;
2559         int empty_cluster = 2 * 1024 * 1024;
2560         int allowed_chunk_alloc = 0;
2561         int using_hint = 0;
2562         struct btrfs_space_info *space_info;
2563
2564         WARN_ON(num_bytes < root->sectorsize);
2565         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2566         ins->objectid = 0;
2567         ins->offset = 0;
2568
2569         space_info = __find_space_info(root->fs_info, data);
2570
2571         if (orig_root->ref_cows || empty_size)
2572                 allowed_chunk_alloc = 1;
2573
2574         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2575                 last_ptr = &root->fs_info->last_alloc;
2576                 if (!btrfs_test_opt(root, SSD))
2577                         empty_cluster = 64 * 1024;
2578         }
2579
2580         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2581                 last_ptr = &root->fs_info->last_data_alloc;
2582
2583         if (last_ptr) {
2584                 if (*last_ptr)
2585                         hint_byte = *last_ptr;
2586                 else
2587                         empty_size += empty_cluster;
2588         } else {
2589                 empty_cluster = 0;
2590         }
2591         search_start = max(search_start, first_logical_byte(root, 0));
2592         search_start = max(search_start, hint_byte);
2593
2594         if (search_start == hint_byte) {
2595                 using_hint = 1;
2596                 block_group = btrfs_lookup_block_group(root->fs_info,
2597                                                        search_start);
2598                 if (block_group && block_group_bits(block_group, data)) {
2599                         down_read(&space_info->groups_sem);
2600                         goto have_block_group;
2601                 } else if (block_group) {
2602                         put_block_group(block_group);
2603                 }
2604
2605                 empty_size += empty_cluster;
2606                 using_hint = 0;
2607         }
2608
2609 search:
2610         down_read(&space_info->groups_sem);
2611         list_for_each_entry(block_group, &space_info->block_groups, list) {
2612                 u64 offset;
2613
2614                 atomic_inc(&block_group->count);
2615                 search_start = block_group->key.objectid;
2616
2617 have_block_group:
2618                 if (unlikely(!block_group->cached)) {
2619                         mutex_lock(&block_group->cache_mutex);
2620                         ret = cache_block_group(root, block_group);
2621                         mutex_unlock(&block_group->cache_mutex);
2622                         if (ret) {
2623                                 put_block_group(block_group);
2624                                 break;
2625                         }
2626                 }
2627
2628                 if (unlikely(block_group->ro))
2629                         goto loop;
2630
2631                 offset = btrfs_find_space_for_alloc(block_group, search_start,
2632                                                     num_bytes, empty_size);
2633                 if (!offset)
2634                         goto loop;
2635
2636                 search_start = stripe_align(root, offset);
2637
2638                 /* move on to the next group */
2639                 if (search_start + num_bytes >= search_end) {
2640                         btrfs_add_free_space(block_group, offset, num_bytes);
2641                         goto loop;
2642                 }
2643
2644                 /* move on to the next group */
2645                 if (search_start + num_bytes >
2646                     block_group->key.objectid + block_group->key.offset) {
2647                         btrfs_add_free_space(block_group, offset, num_bytes);
2648                         goto loop;
2649                 }
2650
2651                 if (using_hint && search_start > hint_byte) {
2652                         btrfs_add_free_space(block_group, offset, num_bytes);
2653                         goto loop;
2654                 }
2655
2656                 if (exclude_nr > 0 &&
2657                     (search_start + num_bytes > exclude_start &&
2658                      search_start < exclude_start + exclude_nr)) {
2659                         search_start = exclude_start + exclude_nr;
2660
2661                         btrfs_add_free_space(block_group, offset, num_bytes);
2662                         /*
2663                          * if search_start is still in this block group
2664                          * then we just re-search this block group
2665                          */
2666                         if (search_start >= block_group->key.objectid &&
2667                             search_start < (block_group->key.objectid +
2668                                             block_group->key.offset))
2669                                 goto have_block_group;
2670                         goto loop;
2671                 }
2672
2673                 ins->objectid = search_start;
2674                 ins->offset = num_bytes;
2675
2676                 if (offset < search_start)
2677                         btrfs_add_free_space(block_group, offset,
2678                                              search_start - offset);
2679                 BUG_ON(offset > search_start);
2680
2681                 /* we are all good, lets return */
2682                 break;
2683 loop:
2684                 put_block_group(block_group);
2685                 if (using_hint) {
2686                         empty_size += empty_cluster;
2687                         using_hint = 0;
2688                         up_read(&space_info->groups_sem);
2689                         goto search;
2690                 }
2691         }
2692         up_read(&space_info->groups_sem);
2693
2694         if (!ins->objectid && (empty_size || allowed_chunk_alloc)) {
2695                 int try_again = empty_size;
2696
2697                 empty_size = 0;
2698
2699                 if (allowed_chunk_alloc) {
2700                         ret = do_chunk_alloc(trans, root, num_bytes +
2701                                              2 * 1024 * 1024, data, 1);
2702                         if (!ret)
2703                                 try_again = 1;
2704                         allowed_chunk_alloc = 0;
2705                 } else {
2706                         space_info->force_alloc = 1;
2707                 }
2708
2709                 if (try_again)
2710                         goto search;
2711                 ret = -ENOSPC;
2712         } else if (!ins->objectid) {
2713                 ret = -ENOSPC;
2714         }
2715
2716         /* we found what we needed */
2717         if (ins->objectid) {
2718                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2719                         trans->block_group = block_group->key.objectid;
2720
2721                 if (last_ptr)
2722                         *last_ptr = ins->objectid + ins->offset;
2723                 put_block_group(block_group);
2724                 ret = 0;
2725         }
2726
2727         return ret;
2728 }
2729
2730 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2731 {
2732         struct btrfs_block_group_cache *cache;
2733
2734         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
2735                (unsigned long long)(info->total_bytes - info->bytes_used -
2736                                     info->bytes_pinned - info->bytes_reserved),
2737                (info->full) ? "" : "not ");
2738         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
2739                " may_use=%llu, used=%llu\n", info->total_bytes,
2740                info->bytes_pinned, info->bytes_delalloc, info->bytes_may_use,
2741                info->bytes_used);
2742
2743         down_read(&info->groups_sem);
2744         list_for_each_entry(cache, &info->block_groups, list) {
2745                 spin_lock(&cache->lock);
2746                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
2747                        "%llu pinned %llu reserved\n",
2748                        (unsigned long long)cache->key.objectid,
2749                        (unsigned long long)cache->key.offset,
2750                        (unsigned long long)btrfs_block_group_used(&cache->item),
2751                        (unsigned long long)cache->pinned,
2752                        (unsigned long long)cache->reserved);
2753                 btrfs_dump_free_space(cache, bytes);
2754                 spin_unlock(&cache->lock);
2755         }
2756         up_read(&info->groups_sem);
2757 }
2758
2759 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2760                                   struct btrfs_root *root,
2761                                   u64 num_bytes, u64 min_alloc_size,
2762                                   u64 empty_size, u64 hint_byte,
2763                                   u64 search_end, struct btrfs_key *ins,
2764                                   u64 data)
2765 {
2766         int ret;
2767         u64 search_start = 0;
2768         struct btrfs_fs_info *info = root->fs_info;
2769
2770         data = btrfs_get_alloc_profile(root, data);
2771 again:
2772         /*
2773          * the only place that sets empty_size is btrfs_realloc_node, which
2774          * is not called recursively on allocations
2775          */
2776         if (empty_size || root->ref_cows) {
2777                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2778                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2779                                      2 * 1024 * 1024,
2780                                      BTRFS_BLOCK_GROUP_METADATA |
2781                                      (info->metadata_alloc_profile &
2782                                       info->avail_metadata_alloc_bits), 0);
2783                 }
2784                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2785                                      num_bytes + 2 * 1024 * 1024, data, 0);
2786         }
2787
2788         WARN_ON(num_bytes < root->sectorsize);
2789         ret = find_free_extent(trans, root, num_bytes, empty_size,
2790                                search_start, search_end, hint_byte, ins,
2791                                trans->alloc_exclude_start,
2792                                trans->alloc_exclude_nr, data);
2793
2794         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2795                 num_bytes = num_bytes >> 1;
2796                 num_bytes = num_bytes & ~(root->sectorsize - 1);
2797                 num_bytes = max(num_bytes, min_alloc_size);
2798                 do_chunk_alloc(trans, root->fs_info->extent_root,
2799                                num_bytes, data, 1);
2800                 goto again;
2801         }
2802         if (ret) {
2803                 struct btrfs_space_info *sinfo;
2804
2805                 sinfo = __find_space_info(root->fs_info, data);
2806                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
2807                        "wanted %llu\n", (unsigned long long)data,
2808                        (unsigned long long)num_bytes);
2809                 dump_space_info(sinfo, num_bytes);
2810                 BUG();
2811         }
2812
2813         return ret;
2814 }
2815
2816 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2817 {
2818         struct btrfs_block_group_cache *cache;
2819         int ret = 0;
2820
2821         cache = btrfs_lookup_block_group(root->fs_info, start);
2822         if (!cache) {
2823                 printk(KERN_ERR "Unable to find block group for %llu\n",
2824                        (unsigned long long)start);
2825                 return -ENOSPC;
2826         }
2827
2828         ret = btrfs_discard_extent(root, start, len);
2829
2830         btrfs_add_free_space(cache, start, len);
2831         put_block_group(cache);
2832         update_reserved_extents(root, start, len, 0);
2833
2834         return ret;
2835 }
2836
2837 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2838                                   struct btrfs_root *root,
2839                                   u64 num_bytes, u64 min_alloc_size,
2840                                   u64 empty_size, u64 hint_byte,
2841                                   u64 search_end, struct btrfs_key *ins,
2842                                   u64 data)
2843 {
2844         int ret;
2845         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2846                                      empty_size, hint_byte, search_end, ins,
2847                                      data);
2848         update_reserved_extents(root, ins->objectid, ins->offset, 1);
2849         return ret;
2850 }
2851
2852 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2853                                          struct btrfs_root *root, u64 parent,
2854                                          u64 root_objectid, u64 ref_generation,
2855                                          u64 owner, struct btrfs_key *ins,
2856                                          int ref_mod)
2857 {
2858         int ret;
2859         u64 super_used;
2860         u64 root_used;
2861         u64 num_bytes = ins->offset;
2862         u32 sizes[2];
2863         struct btrfs_fs_info *info = root->fs_info;
2864         struct btrfs_root *extent_root = info->extent_root;
2865         struct btrfs_extent_item *extent_item;
2866         struct btrfs_extent_ref *ref;
2867         struct btrfs_path *path;
2868         struct btrfs_key keys[2];
2869
2870         if (parent == 0)
2871                 parent = ins->objectid;
2872
2873         /* block accounting for super block */
2874         spin_lock(&info->delalloc_lock);
2875         super_used = btrfs_super_bytes_used(&info->super_copy);
2876         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2877
2878         /* block accounting for root item */
2879         root_used = btrfs_root_used(&root->root_item);
2880         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2881         spin_unlock(&info->delalloc_lock);
2882
2883         memcpy(&keys[0], ins, sizeof(*ins));
2884         keys[1].objectid = ins->objectid;
2885         keys[1].type = BTRFS_EXTENT_REF_KEY;
2886         keys[1].offset = parent;
2887         sizes[0] = sizeof(*extent_item);
2888         sizes[1] = sizeof(*ref);
2889
2890         path = btrfs_alloc_path();
2891         BUG_ON(!path);
2892
2893         path->leave_spinning = 1;
2894         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2895                                        sizes, 2);
2896         BUG_ON(ret);
2897
2898         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2899                                      struct btrfs_extent_item);
2900         btrfs_set_extent_refs(path->nodes[0], extent_item, ref_mod);
2901         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2902                              struct btrfs_extent_ref);
2903
2904         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2905         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2906         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2907         btrfs_set_ref_num_refs(path->nodes[0], ref, ref_mod);
2908
2909         btrfs_mark_buffer_dirty(path->nodes[0]);
2910
2911         trans->alloc_exclude_start = 0;
2912         trans->alloc_exclude_nr = 0;
2913         btrfs_free_path(path);
2914
2915         if (ret)
2916                 goto out;
2917
2918         ret = update_block_group(trans, root, ins->objectid,
2919                                  ins->offset, 1, 0);
2920         if (ret) {
2921                 printk(KERN_ERR "btrfs update block group failed for %llu "
2922                        "%llu\n", (unsigned long long)ins->objectid,
2923                        (unsigned long long)ins->offset);
2924                 BUG();
2925         }
2926 out:
2927         return ret;
2928 }
2929
2930 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2931                                 struct btrfs_root *root, u64 parent,
2932                                 u64 root_objectid, u64 ref_generation,
2933                                 u64 owner, struct btrfs_key *ins)
2934 {
2935         int ret;
2936
2937         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2938                 return 0;
2939
2940         ret = btrfs_add_delayed_ref(trans, ins->objectid,
2941                                     ins->offset, parent, root_objectid,
2942                                     ref_generation, owner,
2943                                     BTRFS_ADD_DELAYED_EXTENT, 0);
2944         BUG_ON(ret);
2945         return ret;
2946 }
2947
2948 /*
2949  * this is used by the tree logging recovery code.  It records that
2950  * an extent has been allocated and makes sure to clear the free
2951  * space cache bits as well
2952  */
2953 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2954                                 struct btrfs_root *root, u64 parent,
2955                                 u64 root_objectid, u64 ref_generation,
2956                                 u64 owner, struct btrfs_key *ins)
2957 {
2958         int ret;
2959         struct btrfs_block_group_cache *block_group;
2960
2961         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2962         mutex_lock(&block_group->cache_mutex);
2963         cache_block_group(root, block_group);
2964         mutex_unlock(&block_group->cache_mutex);
2965
2966         ret = btrfs_remove_free_space(block_group, ins->objectid,
2967                                       ins->offset);
2968         BUG_ON(ret);
2969         put_block_group(block_group);
2970         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2971                                             ref_generation, owner, ins, 1);
2972         return ret;
2973 }
2974
2975 /*
2976  * finds a free extent and does all the dirty work required for allocation
2977  * returns the key for the extent through ins, and a tree buffer for
2978  * the first block of the extent through buf.
2979  *
2980  * returns 0 if everything worked, non-zero otherwise.
2981  */
2982 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2983                        struct btrfs_root *root,
2984                        u64 num_bytes, u64 parent, u64 min_alloc_size,
2985                        u64 root_objectid, u64 ref_generation,
2986                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
2987                        u64 search_end, struct btrfs_key *ins, u64 data)
2988 {
2989         int ret;
2990         ret = __btrfs_reserve_extent(trans, root, num_bytes,
2991                                      min_alloc_size, empty_size, hint_byte,
2992                                      search_end, ins, data);
2993         BUG_ON(ret);
2994         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2995                 ret = btrfs_add_delayed_ref(trans, ins->objectid,
2996                                             ins->offset, parent, root_objectid,
2997                                             ref_generation, owner_objectid,
2998                                             BTRFS_ADD_DELAYED_EXTENT, 0);
2999                 BUG_ON(ret);
3000         }
3001         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3002         return ret;
3003 }
3004
3005 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3006                                             struct btrfs_root *root,
3007                                             u64 bytenr, u32 blocksize,
3008                                             int level)
3009 {
3010         struct extent_buffer *buf;
3011
3012         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3013         if (!buf)
3014                 return ERR_PTR(-ENOMEM);
3015         btrfs_set_header_generation(buf, trans->transid);
3016         btrfs_set_buffer_lockdep_class(buf, level);
3017         btrfs_tree_lock(buf);
3018         clean_tree_block(trans, root, buf);
3019
3020         btrfs_set_lock_blocking(buf);
3021         btrfs_set_buffer_uptodate(buf);
3022
3023         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3024                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3025                          buf->start + buf->len - 1, GFP_NOFS);
3026         } else {
3027                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3028                          buf->start + buf->len - 1, GFP_NOFS);
3029         }
3030         trans->blocks_used++;
3031         /* this returns a buffer locked for blocking */
3032         return buf;
3033 }
3034
3035 /*
3036  * helper function to allocate a block for a given tree
3037  * returns the tree buffer or NULL.
3038  */
3039 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3040                                              struct btrfs_root *root,
3041                                              u32 blocksize, u64 parent,
3042                                              u64 root_objectid,
3043                                              u64 ref_generation,
3044                                              int level,
3045                                              u64 hint,
3046                                              u64 empty_size)
3047 {
3048         struct btrfs_key ins;
3049         int ret;
3050         struct extent_buffer *buf;
3051
3052         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3053                                  root_objectid, ref_generation, level,
3054                                  empty_size, hint, (u64)-1, &ins, 0);
3055         if (ret) {
3056                 BUG_ON(ret > 0);
3057                 return ERR_PTR(ret);
3058         }
3059
3060         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
3061                                     blocksize, level);
3062         return buf;
3063 }
3064
3065 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3066                         struct btrfs_root *root, struct extent_buffer *leaf)
3067 {
3068         u64 leaf_owner;
3069         u64 leaf_generation;
3070         struct refsort *sorted;
3071         struct btrfs_key key;
3072         struct btrfs_file_extent_item *fi;
3073         int i;
3074         int nritems;
3075         int ret;
3076         int refi = 0;
3077         int slot;
3078
3079         BUG_ON(!btrfs_is_leaf(leaf));
3080         nritems = btrfs_header_nritems(leaf);
3081         leaf_owner = btrfs_header_owner(leaf);
3082         leaf_generation = btrfs_header_generation(leaf);
3083
3084         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3085         /* we do this loop twice.  The first time we build a list
3086          * of the extents we have a reference on, then we sort the list
3087          * by bytenr.  The second time around we actually do the
3088          * extent freeing.
3089          */
3090         for (i = 0; i < nritems; i++) {
3091                 u64 disk_bytenr;
3092                 cond_resched();
3093
3094                 btrfs_item_key_to_cpu(leaf, &key, i);
3095
3096                 /* only extents have references, skip everything else */
3097                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3098                         continue;
3099
3100                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3101
3102                 /* inline extents live in the btree, they don't have refs */
3103                 if (btrfs_file_extent_type(leaf, fi) ==
3104                     BTRFS_FILE_EXTENT_INLINE)
3105                         continue;
3106
3107                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3108
3109                 /* holes don't have refs */
3110                 if (disk_bytenr == 0)
3111                         continue;
3112
3113                 sorted[refi].bytenr = disk_bytenr;
3114                 sorted[refi].slot = i;
3115                 refi++;
3116         }
3117
3118         if (refi == 0)
3119                 goto out;
3120
3121         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3122
3123         for (i = 0; i < refi; i++) {
3124                 u64 disk_bytenr;
3125
3126                 disk_bytenr = sorted[i].bytenr;
3127                 slot = sorted[i].slot;
3128
3129                 cond_resched();
3130
3131                 btrfs_item_key_to_cpu(leaf, &key, slot);
3132                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3133                         continue;
3134
3135                 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3136
3137                 ret = btrfs_free_extent(trans, root, disk_bytenr,
3138                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3139                                 leaf->start, leaf_owner, leaf_generation,
3140                                 key.objectid, 0);
3141                 BUG_ON(ret);
3142
3143                 atomic_inc(&root->fs_info->throttle_gen);
3144                 wake_up(&root->fs_info->transaction_throttle);
3145                 cond_resched();
3146         }
3147 out:
3148         kfree(sorted);
3149         return 0;
3150 }
3151
3152 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3153                                         struct btrfs_root *root,
3154                                         struct btrfs_leaf_ref *ref)
3155 {
3156         int i;
3157         int ret;
3158         struct btrfs_extent_info *info;
3159         struct refsort *sorted;
3160
3161         if (ref->nritems == 0)
3162                 return 0;
3163
3164         sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
3165         for (i = 0; i < ref->nritems; i++) {
3166                 sorted[i].bytenr = ref->extents[i].bytenr;
3167                 sorted[i].slot = i;
3168         }
3169         sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
3170
3171         /*
3172          * the items in the ref were sorted when the ref was inserted
3173          * into the ref cache, so this is already in order
3174          */
3175         for (i = 0; i < ref->nritems; i++) {
3176                 info = ref->extents + sorted[i].slot;
3177                 ret = btrfs_free_extent(trans, root, info->bytenr,
3178                                           info->num_bytes, ref->bytenr,
3179                                           ref->owner, ref->generation,
3180                                           info->objectid, 0);
3181
3182                 atomic_inc(&root->fs_info->throttle_gen);
3183                 wake_up(&root->fs_info->transaction_throttle);
3184                 cond_resched();
3185
3186                 BUG_ON(ret);
3187                 info++;
3188         }
3189
3190         kfree(sorted);
3191         return 0;
3192 }
3193
3194 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
3195                                      struct btrfs_root *root, u64 start,
3196                                      u64 len, u32 *refs)
3197 {
3198         int ret;
3199
3200         ret = btrfs_lookup_extent_ref(trans, root, start, len, refs);
3201         BUG_ON(ret);
3202
3203 #if 0 /* some debugging code in case we see problems here */
3204         /* if the refs count is one, it won't get increased again.  But
3205          * if the ref count is > 1, someone may be decreasing it at
3206          * the same time we are.
3207          */
3208         if (*refs != 1) {
3209                 struct extent_buffer *eb = NULL;
3210                 eb = btrfs_find_create_tree_block(root, start, len);
3211                 if (eb)
3212                         btrfs_tree_lock(eb);
3213
3214                 mutex_lock(&root->fs_info->alloc_mutex);
3215                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3216                 BUG_ON(ret);
3217                 mutex_unlock(&root->fs_info->alloc_mutex);
3218
3219                 if (eb) {
3220                         btrfs_tree_unlock(eb);
3221                         free_extent_buffer(eb);
3222                 }
3223                 if (*refs == 1) {
3224                         printk(KERN_ERR "btrfs block %llu went down to one "
3225                                "during drop_snap\n", (unsigned long long)start);
3226                 }
3227
3228         }
3229 #endif
3230
3231         cond_resched();
3232         return ret;
3233 }
3234
3235 /*
3236  * this is used while deleting old snapshots, and it drops the refs
3237  * on a whole subtree starting from a level 1 node.
3238  *
3239  * The idea is to sort all the leaf pointers, and then drop the
3240  * ref on all the leaves in order.  Most of the time the leaves
3241  * will have ref cache entries, so no leaf IOs will be required to
3242  * find the extents they have references on.
3243  *
3244  * For each leaf, any references it has are also dropped in order
3245  *
3246  * This ends up dropping the references in something close to optimal
3247  * order for reading and modifying the extent allocation tree.
3248  */
3249 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3250                                         struct btrfs_root *root,
3251                                         struct btrfs_path *path)
3252 {
3253         u64 bytenr;
3254         u64 root_owner;
3255         u64 root_gen;
3256         struct extent_buffer *eb = path->nodes[1];
3257         struct extent_buffer *leaf;
3258         struct btrfs_leaf_ref *ref;
3259         struct refsort *sorted = NULL;
3260         int nritems = btrfs_header_nritems(eb);
3261         int ret;
3262         int i;
3263         int refi = 0;
3264         int slot = path->slots[1];
3265         u32 blocksize = btrfs_level_size(root, 0);
3266         u32 refs;
3267
3268         if (nritems == 0)
3269                 goto out;
3270
3271         root_owner = btrfs_header_owner(eb);
3272         root_gen = btrfs_header_generation(eb);
3273         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3274
3275         /*
3276          * step one, sort all the leaf pointers so we don't scribble
3277          * randomly into the extent allocation tree
3278          */
3279         for (i = slot; i < nritems; i++) {
3280                 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
3281                 sorted[refi].slot = i;
3282                 refi++;
3283         }
3284
3285         /*
3286          * nritems won't be zero, but if we're picking up drop_snapshot
3287          * after a crash, slot might be > 0, so double check things
3288          * just in case.
3289          */
3290         if (refi == 0)
3291                 goto out;
3292
3293         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3294
3295         /*
3296          * the first loop frees everything the leaves point to
3297          */
3298         for (i = 0; i < refi; i++) {
3299                 u64 ptr_gen;
3300
3301                 bytenr = sorted[i].bytenr;
3302
3303                 /*
3304                  * check the reference count on this leaf.  If it is > 1
3305                  * we just decrement it below and don't update any
3306                  * of the refs the leaf points to.
3307                  */
3308                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3309                                                 blocksize, &refs);
3310                 BUG_ON(ret);
3311                 if (refs != 1)
3312                         continue;
3313
3314                 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
3315
3316                 /*
3317                  * the leaf only had one reference, which means the
3318                  * only thing pointing to this leaf is the snapshot
3319                  * we're deleting.  It isn't possible for the reference
3320                  * count to increase again later
3321                  *
3322                  * The reference cache is checked for the leaf,
3323                  * and if found we'll be able to drop any refs held by
3324                  * the leaf without needing to read it in.
3325                  */
3326                 ref = btrfs_lookup_leaf_ref(root, bytenr);
3327                 if (ref && ref->generation != ptr_gen) {
3328                         btrfs_free_leaf_ref(root, ref);
3329                         ref = NULL;
3330                 }
3331                 if (ref) {
3332                         ret = cache_drop_leaf_ref(trans, root, ref);
3333                         BUG_ON(ret);
3334                         btrfs_remove_leaf_ref(root, ref);
3335                         btrfs_free_leaf_ref(root, ref);
3336                 } else {
3337                         /*
3338                          * the leaf wasn't in the reference cache, so
3339                          * we have to read it.
3340                          */
3341                         leaf = read_tree_block(root, bytenr, blocksize,
3342                                                ptr_gen);
3343                         ret = btrfs_drop_leaf_ref(trans, root, leaf);
3344                         BUG_ON(ret);
3345                         free_extent_buffer(leaf);
3346                 }
3347                 atomic_inc(&root->fs_info->throttle_gen);
3348                 wake_up(&root->fs_info->transaction_throttle);
3349                 cond_resched();
3350         }
3351
3352         /*
3353          * run through the loop again to free the refs on the leaves.
3354          * This is faster than doing it in the loop above because
3355          * the leaves are likely to be clustered together.  We end up
3356          * working in nice chunks on the extent allocation tree.
3357          */
3358         for (i = 0; i < refi; i++) {
3359                 bytenr = sorted[i].bytenr;
3360                 ret = btrfs_free_extent(trans, root, bytenr,
3361                                         blocksize, eb->start,
3362                                         root_owner, root_gen, 0, 1);
3363                 BUG_ON(ret);
3364
3365                 atomic_inc(&root->fs_info->throttle_gen);
3366                 wake_up(&root->fs_info->transaction_throttle);
3367                 cond_resched();
3368         }
3369 out:
3370         kfree(sorted);
3371
3372         /*
3373          * update the path to show we've processed the entire level 1
3374          * node.  This will get saved into the root's drop_snapshot_progress
3375          * field so these drops are not repeated again if this transaction
3376          * commits.
3377          */
3378         path->slots[1] = nritems;
3379         return 0;
3380 }
3381
3382 /*
3383  * helper function for drop_snapshot, this walks down the tree dropping ref
3384  * counts as it goes.
3385  */
3386 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3387                                    struct btrfs_root *root,
3388                                    struct btrfs_path *path, int *level)
3389 {
3390         u64 root_owner;
3391         u64 root_gen;
3392         u64 bytenr;
3393         u64 ptr_gen;
3394         struct extent_buffer *next;
3395         struct extent_buffer *cur;
3396         struct extent_buffer *parent;
3397         u32 blocksize;
3398         int ret;
3399         u32 refs;
3400
3401         WARN_ON(*level < 0);
3402         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3403         ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
3404                                 path->nodes[*level]->len, &refs);
3405         BUG_ON(ret);
3406         if (refs > 1)
3407                 goto out;
3408
3409         /*
3410          * walk down to the last node level and free all the leaves
3411          */
3412         while (*level >= 0) {
3413                 WARN_ON(*level < 0);
3414                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3415                 cur = path->nodes[*level];
3416
3417                 if (btrfs_header_level(cur) != *level)
3418                         WARN_ON(1);
3419
3420                 if (path->slots[*level] >=
3421                     btrfs_header_nritems(cur))
3422                         break;
3423
3424                 /* the new code goes down to level 1 and does all the
3425                  * leaves pointed to that node in bulk.  So, this check
3426                  * for level 0 will always be false.
3427                  *
3428                  * But, the disk format allows the drop_snapshot_progress
3429                  * field in the root to leave things in a state where
3430                  * a leaf will need cleaning up here.  If someone crashes
3431                  * with the old code and then boots with the new code,
3432                  * we might find a leaf here.
3433                  */
3434                 if (*level == 0) {
3435                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3436                         BUG_ON(ret);
3437                         break;
3438                 }
3439
3440                 /*
3441                  * once we get to level one, process the whole node
3442                  * at once, including everything below it.
3443                  */
3444                 if (*level == 1) {
3445                         ret = drop_level_one_refs(trans, root, path);
3446                         BUG_ON(ret);
3447                         break;
3448                 }
3449
3450                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3451                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3452                 blocksize = btrfs_level_size(root, *level - 1);
3453
3454                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3455                                                 blocksize, &refs);
3456                 BUG_ON(ret);
3457
3458                 /*
3459                  * if there is more than one reference, we don't need
3460                  * to read that node to drop any references it has.  We
3461                  * just drop the ref we hold on that node and move on to the
3462                  * next slot in this level.
3463                  */
3464                 if (refs != 1) {
3465                         parent = path->nodes[*level];
3466                         root_owner = btrfs_header_owner(parent);
3467                         root_gen = btrfs_header_generation(parent);
3468                         path->slots[*level]++;
3469
3470                         ret = btrfs_free_extent(trans, root, bytenr,
3471                                                 blocksize, parent->start,
3472                                                 root_owner, root_gen,
3473                                                 *level - 1, 1);
3474                         BUG_ON(ret);
3475
3476                         atomic_inc(&root->fs_info->throttle_gen);
3477                         wake_up(&root->fs_info->transaction_throttle);
3478                         cond_resched();
3479
3480                         continue;
3481                 }
3482
3483                 /*
3484                  * we need to keep freeing things in the next level down.
3485                  * read the block and loop around to process it
3486                  */
3487                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3488                 WARN_ON(*level <= 0);
3489                 if (path->nodes[*level-1])
3490                         free_extent_buffer(path->nodes[*level-1]);
3491                 path->nodes[*level-1] = next;
3492                 *level = btrfs_header_level(next);
3493                 path->slots[*level] = 0;
3494                 cond_resched();
3495         }
3496 out:
3497         WARN_ON(*level < 0);
3498         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3499
3500         if (path->nodes[*level] == root->node) {
3501                 parent = path->nodes[*level];
3502                 bytenr = path->nodes[*level]->start;
3503         } else {
3504                 parent = path->nodes[*level + 1];
3505                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3506         }
3507
3508         blocksize = btrfs_level_size(root, *level);
3509         root_owner = btrfs_header_owner(parent);
3510         root_gen = btrfs_header_generation(parent);
3511
3512         /*
3513          * cleanup and free the reference on the last node
3514          * we processed
3515          */
3516         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3517                                   parent->start, root_owner, root_gen,
3518                                   *level, 1);
3519         free_extent_buffer(path->nodes[*level]);
3520         path->nodes[*level] = NULL;
3521
3522         *level += 1;
3523         BUG_ON(ret);
3524
3525         cond_resched();
3526         return 0;
3527 }
3528
3529 /*
3530  * helper function for drop_subtree, this function is similar to
3531  * walk_down_tree. The main difference is that it checks reference
3532  * counts while tree blocks are locked.
3533  */
3534 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3535                                       struct btrfs_root *root,
3536                                       struct btrfs_path *path, int *level)
3537 {
3538         struct extent_buffer *next;
3539         struct extent_buffer *cur;
3540         struct extent_buffer *parent;
3541         u64 bytenr;
3542         u64 ptr_gen;
3543         u32 blocksize;
3544         u32 refs;
3545         int ret;
3546
3547         cur = path->nodes[*level];
3548         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3549                                       &refs);
3550         BUG_ON(ret);
3551         if (refs > 1)
3552                 goto out;
3553
3554         while (*level >= 0) {
3555                 cur = path->nodes[*level];
3556                 if (*level == 0) {
3557                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3558                         BUG_ON(ret);
3559                         clean_tree_block(trans, root, cur);
3560                         break;
3561                 }
3562                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3563                         clean_tree_block(trans, root, cur);
3564                         break;
3565                 }
3566
3567                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3568                 blocksize = btrfs_level_size(root, *level - 1);
3569                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3570
3571                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3572                 btrfs_tree_lock(next);
3573                 btrfs_set_lock_blocking(next);
3574
3575                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3576                                               &refs);
3577                 BUG_ON(ret);
3578                 if (refs > 1) {
3579                         parent = path->nodes[*level];
3580                         ret = btrfs_free_extent(trans, root, bytenr,
3581                                         blocksize, parent->start,
3582                                         btrfs_header_owner(parent),
3583                                         btrfs_header_generation(parent),
3584                                         *level - 1, 1);
3585                         BUG_ON(ret);
3586                         path->slots[*level]++;
3587                         btrfs_tree_unlock(next);
3588                         free_extent_buffer(next);
3589                         continue;
3590                 }
3591
3592                 *level = btrfs_header_level(next);
3593                 path->nodes[*level] = next;
3594                 path->slots[*level] = 0;
3595                 path->locks[*level] = 1;
3596                 cond_resched();
3597         }
3598 out:
3599         parent = path->nodes[*level + 1];
3600         bytenr = path->nodes[*level]->start;
3601         blocksize = path->nodes[*level]->len;
3602
3603         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3604                         parent->start, btrfs_header_owner(parent),
3605                         btrfs_header_generation(parent), *level, 1);
3606         BUG_ON(ret);
3607
3608         if (path->locks[*level]) {
3609                 btrfs_tree_unlock(path->nodes[*level]);
3610                 path->locks[*level] = 0;
3611         }
3612         free_extent_buffer(path->nodes[*level]);
3613         path->nodes[*level] = NULL;
3614         *level += 1;
3615         cond_resched();
3616         return 0;
3617 }
3618
3619 /*
3620  * helper for dropping snapshots.  This walks back up the tree in the path
3621  * to find the first node higher up where we haven't yet gone through
3622  * all the slots
3623  */
3624 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3625                                  struct btrfs_root *root,
3626                                  struct btrfs_path *path,
3627                                  int *level, int max_level)
3628 {
3629         u64 root_owner;
3630         u64 root_gen;
3631         struct btrfs_root_item *root_item = &root->root_item;
3632         int i;
3633         int slot;
3634         int ret;
3635
3636         for (i = *level; i < max_level && path->nodes[i]; i++) {
3637                 slot = path->slots[i];
3638                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3639                         struct extent_buffer *node;
3640                         struct btrfs_disk_key disk_key;
3641
3642                         /*
3643                          * there is more work to do in this level.
3644                          * Update the drop_progress marker to reflect
3645                          * the work we've done so far, and then bump
3646                          * the slot number
3647                          */
3648                         node = path->nodes[i];
3649                         path->slots[i]++;
3650                         *level = i;
3651                         WARN_ON(*level == 0);
3652                         btrfs_node_key(node, &disk_key, path->slots[i]);
3653                         memcpy(&root_item->drop_progress,
3654                                &disk_key, sizeof(disk_key));
3655                         root_item->drop_level = i;
3656                         return 0;
3657                 } else {
3658                         struct extent_buffer *parent;
3659
3660                         /*
3661                          * this whole node is done, free our reference
3662                          * on it and go up one level
3663                          */
3664                         if (path->nodes[*level] == root->node)
3665                                 parent = path->nodes[*level];
3666                         else
3667                                 parent = path->nodes[*level + 1];
3668
3669                         root_owner = btrfs_header_owner(parent);
3670                         root_gen = btrfs_header_generation(parent);
3671
3672                         clean_tree_block(trans, root, path->nodes[*level]);
3673                         ret = btrfs_free_extent(trans, root,
3674                                                 path->nodes[*level]->start,
3675                                                 path->nodes[*level]->len,
3676                                                 parent->start, root_owner,
3677                                                 root_gen, *level, 1);
3678                         BUG_ON(ret);
3679                         if (path->locks[*level]) {
3680                                 btrfs_tree_unlock(path->nodes[*level]);
3681                                 path->locks[*level] = 0;
3682                         }
3683                         free_extent_buffer(path->nodes[*level]);
3684                         path->nodes[*level] = NULL;
3685                         *level = i + 1;
3686                 }
3687         }
3688         return 1;
3689 }
3690
3691 /*
3692  * drop the reference count on the tree rooted at 'snap'.  This traverses
3693  * the tree freeing any blocks that have a ref count of zero after being
3694  * decremented.
3695  */
3696 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3697                         *root)
3698 {
3699         int ret = 0;
3700         int wret;
3701         int level;
3702         struct btrfs_path *path;
3703         int i;
3704         int orig_level;
3705         int update_count;
3706         struct btrfs_root_item *root_item = &root->root_item;
3707
3708         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3709         path = btrfs_alloc_path();
3710         BUG_ON(!path);
3711
3712         level = btrfs_header_level(root->node);
3713         orig_level = level;
3714         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3715                 path->nodes[level] = root->node;
3716                 extent_buffer_get(root->node);
3717                 path->slots[level] = 0;
3718         } else {
3719                 struct btrfs_key key;
3720                 struct btrfs_disk_key found_key;
3721                 struct extent_buffer *node;
3722
3723                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3724                 level = root_item->drop_level;
3725                 path->lowest_level = level;
3726                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3727                 if (wret < 0) {
3728                         ret = wret;
3729                         goto out;
3730                 }
3731                 node = path->nodes[level];
3732                 btrfs_node_key(node, &found_key, path->slots[level]);
3733                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3734                                sizeof(found_key)));
3735                 /*
3736                  * unlock our path, this is safe because only this
3737                  * function is allowed to delete this snapshot
3738                  */
3739                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3740                         if (path->nodes[i] && path->locks[i]) {
3741                                 path->locks[i] = 0;
3742                                 btrfs_tree_unlock(path->nodes[i]);
3743                         }
3744                 }
3745         }
3746         while (1) {
3747                 unsigned long update;
3748                 wret = walk_down_tree(trans, root, path, &level);
3749                 if (wret > 0)
3750                         break;
3751                 if (wret < 0)
3752                         ret = wret;
3753
3754                 wret = walk_up_tree(trans, root, path, &level,
3755                                     BTRFS_MAX_LEVEL);
3756                 if (wret > 0)
3757                         break;
3758                 if (wret < 0)
3759                         ret = wret;
3760                 if (trans->transaction->in_commit ||
3761                     trans->transaction->delayed_refs.flushing) {
3762                         ret = -EAGAIN;
3763                         break;
3764                 }
3765                 atomic_inc(&root->fs_info->throttle_gen);
3766                 wake_up(&root->fs_info->transaction_throttle);
3767                 for (update_count = 0; update_count < 16; update_count++) {
3768                         update = trans->delayed_ref_updates;
3769                         trans->delayed_ref_updates = 0;
3770                         if (update)
3771                                 btrfs_run_delayed_refs(trans, root, update);
3772                         else
3773                                 break;
3774                 }
3775         }
3776         for (i = 0; i <= orig_level; i++) {
3777                 if (path->nodes[i]) {
3778                         free_extent_buffer(path->nodes[i]);
3779                         path->nodes[i] = NULL;
3780                 }
3781         }
3782 out:
3783         btrfs_free_path(path);
3784         return ret;
3785 }
3786
3787 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3788                         struct btrfs_root *root,
3789                         struct extent_buffer *node,
3790                         struct extent_buffer *parent)
3791 {
3792         struct btrfs_path *path;
3793         int level;
3794         int parent_level;
3795         int ret = 0;
3796         int wret;
3797
3798         path = btrfs_alloc_path();
3799         BUG_ON(!path);
3800
3801         btrfs_assert_tree_locked(parent);
3802         parent_level = btrfs_header_level(parent);
3803         extent_buffer_get(parent);
3804         path->nodes[parent_level] = parent;
3805         path->slots[parent_level] = btrfs_header_nritems(parent);
3806
3807         btrfs_assert_tree_locked(node);
3808         level = btrfs_header_level(node);
3809         extent_buffer_get(node);
3810         path->nodes[level] = node;
3811         path->slots[level] = 0;
3812
3813         while (1) {
3814                 wret = walk_down_subtree(trans, root, path, &level);
3815                 if (wret < 0)
3816                         ret = wret;
3817                 if (wret != 0)
3818                         break;
3819
3820                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3821                 if (wret < 0)
3822                         ret = wret;
3823                 if (wret != 0)
3824                         break;
3825         }
3826
3827         btrfs_free_path(path);
3828         return ret;
3829 }
3830
3831 static unsigned long calc_ra(unsigned long start, unsigned long last,
3832                              unsigned long nr)
3833 {
3834         return min(last, start + nr - 1);
3835 }
3836
3837 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3838                                          u64 len)
3839 {
3840         u64 page_start;
3841         u64 page_end;
3842         unsigned long first_index;
3843         unsigned long last_index;
3844         unsigned long i;
3845         struct page *page;
3846         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3847         struct file_ra_state *ra;
3848         struct btrfs_ordered_extent *ordered;
3849         unsigned int total_read = 0;
3850         unsigned int total_dirty = 0;
3851         int ret = 0;
3852
3853         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3854
3855         mutex_lock(&inode->i_mutex);
3856         first_index = start >> PAGE_CACHE_SHIFT;
3857         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3858
3859         /* make sure the dirty trick played by the caller work */
3860         ret = invalidate_inode_pages2_range(inode->i_mapping,
3861                                             first_index, last_index);
3862         if (ret)
3863                 goto out_unlock;
3864
3865         file_ra_state_init(ra, inode->i_mapping);
3866
3867         for (i = first_index ; i <= last_index; i++) {
3868                 if (total_read % ra->ra_pages == 0) {
3869                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3870                                        calc_ra(i, last_index, ra->ra_pages));
3871                 }
3872                 total_read++;
3873 again:
3874                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3875                         BUG_ON(1);
3876                 page = grab_cache_page(inode->i_mapping, i);
3877                 if (!page) {
3878                         ret = -ENOMEM;
3879                         goto out_unlock;
3880                 }
3881                 if (!PageUptodate(page)) {
3882                         btrfs_readpage(NULL, page);
3883                         lock_page(page);
3884                         if (!PageUptodate(page)) {
3885                                 unlock_page(page);
3886                                 page_cache_release(page);
3887                                 ret = -EIO;
3888                                 goto out_unlock;
3889                         }
3890                 }
3891                 wait_on_page_writeback(page);
3892
3893                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3894                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3895                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3896
3897                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3898                 if (ordered) {
3899                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3900                         unlock_page(page);
3901                         page_cache_release(page);
3902                         btrfs_start_ordered_extent(inode, ordered, 1);
3903                         btrfs_put_ordered_extent(ordered);
3904                         goto again;
3905                 }
3906                 set_page_extent_mapped(page);
3907
3908                 if (i == first_index)
3909                         set_extent_bits(io_tree, page_start, page_end,
3910                                         EXTENT_BOUNDARY, GFP_NOFS);
3911                 btrfs_set_extent_delalloc(inode, page_start, page_end);
3912
3913                 set_page_dirty(page);
3914                 total_dirty++;
3915
3916                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3917                 unlock_page(page);
3918                 page_cache_release(page);
3919         }
3920
3921 out_unlock:
3922         kfree(ra);
3923         mutex_unlock(&inode->i_mutex);
3924         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3925         return ret;
3926 }
3927
3928 static noinline int relocate_data_extent(struct inode *reloc_inode,
3929                                          struct btrfs_key *extent_key,
3930                                          u64 offset)
3931 {
3932         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3933         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3934         struct extent_map *em;
3935         u64 start = extent_key->objectid - offset;
3936         u64 end = start + extent_key->offset - 1;
3937
3938         em = alloc_extent_map(GFP_NOFS);
3939         BUG_ON(!em || IS_ERR(em));
3940
3941         em->start = start;
3942         em->len = extent_key->offset;
3943         em->block_len = extent_key->offset;
3944         em->block_start = extent_key->objectid;
3945         em->bdev = root->fs_info->fs_devices->latest_bdev;
3946         set_bit(EXTENT_FLAG_PINNED, &em->flags);
3947
3948         /* setup extent map to cheat btrfs_readpage */
3949         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3950         while (1) {
3951                 int ret;
3952                 spin_lock(&em_tree->lock);
3953                 ret = add_extent_mapping(em_tree, em);
3954                 spin_unlock(&em_tree->lock);
3955                 if (ret != -EEXIST) {
3956                         free_extent_map(em);
3957                         break;
3958                 }
3959                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
3960         }
3961         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3962
3963         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
3964 }
3965
3966 struct btrfs_ref_path {
3967         u64 extent_start;
3968         u64 nodes[BTRFS_MAX_LEVEL];
3969         u64 root_objectid;
3970         u64 root_generation;
3971         u64 owner_objectid;
3972         u32 num_refs;
3973         int lowest_level;
3974         int current_level;
3975         int shared_level;
3976
3977         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
3978         u64 new_nodes[BTRFS_MAX_LEVEL];
3979 };
3980
3981 struct disk_extent {
3982         u64 ram_bytes;
3983         u64 disk_bytenr;
3984         u64 disk_num_bytes;
3985         u64 offset;
3986         u64 num_bytes;
3987         u8 compression;
3988         u8 encryption;
3989         u16 other_encoding;
3990 };
3991
3992 static int is_cowonly_root(u64 root_objectid)
3993 {
3994         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3995             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3996             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3997             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3998             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
3999             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4000                 return 1;
4001         return 0;
4002 }
4003
4004 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4005                                     struct btrfs_root *extent_root,
4006                                     struct btrfs_ref_path *ref_path,
4007                                     int first_time)
4008 {
4009         struct extent_buffer *leaf;
4010         struct btrfs_path *path;
4011         struct btrfs_extent_ref *ref;
4012         struct btrfs_key key;
4013         struct btrfs_key found_key;
4014         u64 bytenr;
4015         u32 nritems;
4016         int level;
4017         int ret = 1;
4018
4019         path = btrfs_alloc_path();
4020         if (!path)
4021                 return -ENOMEM;
4022
4023         if (first_time) {
4024                 ref_path->lowest_level = -1;
4025                 ref_path->current_level = -1;
4026                 ref_path->shared_level = -1;
4027                 goto walk_up;
4028         }
4029 walk_down:
4030         level = ref_path->current_level - 1;
4031         while (level >= -1) {
4032                 u64 parent;
4033                 if (level < ref_path->lowest_level)
4034                         break;
4035
4036                 if (level >= 0)
4037                         bytenr = ref_path->nodes[level];
4038                 else
4039                         bytenr = ref_path->extent_start;
4040                 BUG_ON(bytenr == 0);
4041
4042                 parent = ref_path->nodes[level + 1];
4043                 ref_path->nodes[level + 1] = 0;
4044                 ref_path->current_level = level;
4045                 BUG_ON(parent == 0);
4046
4047                 key.objectid = bytenr;
4048                 key.offset = parent + 1;
4049                 key.type = BTRFS_EXTENT_REF_KEY;
4050
4051                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4052                 if (ret < 0)
4053                         goto out;
4054                 BUG_ON(ret == 0);
4055
4056                 leaf = path->nodes[0];
4057                 nritems = btrfs_header_nritems(leaf);
4058                 if (path->slots[0] >= nritems) {
4059                         ret = btrfs_next_leaf(extent_root, path);
4060                         if (ret < 0)
4061                                 goto out;
4062                         if (ret > 0)
4063                                 goto next;
4064                         leaf = path->nodes[0];
4065                 }
4066
4067                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4068                 if (found_key.objectid == bytenr &&
4069                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4070                         if (level < ref_path->shared_level)
4071                                 ref_path->shared_level = level;
4072                         goto found;
4073                 }
4074 next:
4075                 level--;
4076                 btrfs_release_path(extent_root, path);
4077                 cond_resched();
4078         }
4079         /* reached lowest level */
4080         ret = 1;
4081         goto out;
4082 walk_up:
4083         level = ref_path->current_level;
4084         while (level < BTRFS_MAX_LEVEL - 1) {
4085                 u64 ref_objectid;
4086
4087                 if (level >= 0)
4088                         bytenr = ref_path->nodes[level];
4089                 else
4090                         bytenr = ref_path->extent_start;
4091
4092                 BUG_ON(bytenr == 0);
4093
4094                 key.objectid = bytenr;
4095                 key.offset = 0;
4096                 key.type = BTRFS_EXTENT_REF_KEY;
4097
4098                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4099                 if (ret < 0)
4100                         goto out;
4101
4102                 leaf = path->nodes[0];
4103                 nritems = btrfs_header_nritems(leaf);
4104                 if (path->slots[0] >= nritems) {
4105                         ret = btrfs_next_leaf(extent_root, path);
4106                         if (ret < 0)
4107                                 goto out;
4108                         if (ret > 0) {
4109                                 /* the extent was freed by someone */
4110                                 if (ref_path->lowest_level == level)
4111                                         goto out;
4112                                 btrfs_release_path(extent_root, path);
4113                                 goto walk_down;
4114                         }
4115                         leaf = path->nodes[0];
4116                 }
4117
4118                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4119                 if (found_key.objectid != bytenr ||
4120                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4121                         /* the extent was freed by someone */
4122                         if (ref_path->lowest_level == level) {
4123                                 ret = 1;
4124                                 goto out;
4125                         }
4126                         btrfs_release_path(extent_root, path);
4127                         goto walk_down;
4128                 }
4129 found:
4130                 ref = btrfs_item_ptr(leaf, path->slots[0],
4131                                 struct btrfs_extent_ref);
4132                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4133                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4134                         if (first_time) {
4135                                 level = (int)ref_objectid;
4136                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4137                                 ref_path->lowest_level = level;
4138                                 ref_path->current_level = level;
4139                                 ref_path->nodes[level] = bytenr;
4140                         } else {
4141                                 WARN_ON(ref_objectid != level);
4142                         }
4143                 } else {
4144                         WARN_ON(level != -1);
4145                 }
4146                 first_time = 0;
4147
4148                 if (ref_path->lowest_level == level) {
4149                         ref_path->owner_objectid = ref_objectid;
4150                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4151                 }
4152
4153                 /*
4154                  * the block is tree root or the block isn't in reference
4155                  * counted tree.
4156                  */
4157                 if (found_key.objectid == found_key.offset ||
4158                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4159                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4160                         ref_path->root_generation =
4161                                 btrfs_ref_generation(leaf, ref);
4162                         if (level < 0) {
4163                                 /* special reference from the tree log */
4164                                 ref_path->nodes[0] = found_key.offset;
4165                                 ref_path->current_level = 0;
4166                         }
4167                         ret = 0;
4168                         goto out;
4169                 }
4170
4171                 level++;
4172                 BUG_ON(ref_path->nodes[level] != 0);
4173                 ref_path->nodes[level] = found_key.offset;
4174                 ref_path->current_level = level;
4175
4176                 /*
4177                  * the reference was created in the running transaction,
4178                  * no need to continue walking up.
4179                  */
4180                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4181                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4182                         ref_path->root_generation =
4183                                 btrfs_ref_generation(leaf, ref);
4184                         ret = 0;
4185                         goto out;
4186                 }
4187
4188                 btrfs_release_path(extent_root, path);
4189                 cond_resched();
4190         }
4191         /* reached max tree level, but no tree root found. */
4192         BUG();
4193 out:
4194         btrfs_free_path(path);
4195         return ret;
4196 }
4197
4198 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4199                                 struct btrfs_root *extent_root,
4200                                 struct btrfs_ref_path *ref_path,
4201                                 u64 extent_start)
4202 {
4203         memset(ref_path, 0, sizeof(*ref_path));
4204         ref_path->extent_start = extent_start;
4205
4206         return __next_ref_path(trans, extent_root, ref_path, 1);
4207 }
4208
4209 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4210                                struct btrfs_root *extent_root,
4211                                struct btrfs_ref_path *ref_path)
4212 {
4213         return __next_ref_path(trans, extent_root, ref_path, 0);
4214 }
4215
4216 static noinline int get_new_locations(struct inode *reloc_inode,
4217                                       struct btrfs_key *extent_key,
4218                                       u64 offset, int no_fragment,
4219                                       struct disk_extent **extents,
4220                                       int *nr_extents)
4221 {
4222         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4223         struct btrfs_path *path;
4224         struct btrfs_file_extent_item *fi;
4225         struct extent_buffer *leaf;
4226         struct disk_extent *exts = *extents;
4227         struct btrfs_key found_key;
4228         u64 cur_pos;
4229         u64 last_byte;
4230         u32 nritems;
4231         int nr = 0;
4232         int max = *nr_extents;
4233         int ret;
4234
4235         WARN_ON(!no_fragment && *extents);
4236         if (!exts) {
4237                 max = 1;
4238                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4239                 if (!exts)
4240                         return -ENOMEM;
4241         }
4242
4243         path = btrfs_alloc_path();
4244         BUG_ON(!path);
4245
4246         cur_pos = extent_key->objectid - offset;
4247         last_byte = extent_key->objectid + extent_key->offset;
4248         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4249                                        cur_pos, 0);
4250         if (ret < 0)
4251                 goto out;
4252         if (ret > 0) {
4253                 ret = -ENOENT;
4254                 goto out;
4255         }
4256
4257         while (1) {
4258                 leaf = path->nodes[0];
4259                 nritems = btrfs_header_nritems(leaf);
4260                 if (path->slots[0] >= nritems) {
4261                         ret = btrfs_next_leaf(root, path);
4262                         if (ret < 0)
4263                                 goto out;
4264                         if (ret > 0)
4265                                 break;
4266                         leaf = path->nodes[0];
4267                 }
4268
4269                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4270                 if (found_key.offset != cur_pos ||
4271                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4272                     found_key.objectid != reloc_inode->i_ino)
4273                         break;
4274
4275                 fi = btrfs_item_ptr(leaf, path->slots[0],
4276                                     struct btrfs_file_extent_item);
4277                 if (btrfs_file_extent_type(leaf, fi) !=
4278                     BTRFS_FILE_EXTENT_REG ||
4279                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4280                         break;
4281
4282                 if (nr == max) {
4283                         struct disk_extent *old = exts;
4284                         max *= 2;
4285                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4286                         memcpy(exts, old, sizeof(*exts) * nr);
4287                         if (old != *extents)
4288                                 kfree(old);
4289                 }
4290
4291                 exts[nr].disk_bytenr =
4292                         btrfs_file_extent_disk_bytenr(leaf, fi);
4293                 exts[nr].disk_num_bytes =
4294                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4295                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4296                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4297                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4298                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4299                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4300                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4301                                                                            fi);
4302                 BUG_ON(exts[nr].offset > 0);
4303                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4304                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4305
4306                 cur_pos += exts[nr].num_bytes;
4307                 nr++;
4308
4309                 if (cur_pos + offset >= last_byte)
4310                         break;
4311
4312                 if (no_fragment) {
4313                         ret = 1;
4314                         goto out;
4315                 }
4316                 path->slots[0]++;
4317         }
4318
4319         BUG_ON(cur_pos + offset > last_byte);
4320         if (cur_pos + offset < last_byte) {
4321                 ret = -ENOENT;
4322                 goto out;
4323         }
4324         ret = 0;
4325 out:
4326         btrfs_free_path(path);
4327         if (ret) {
4328                 if (exts != *extents)
4329                         kfree(exts);
4330         } else {
4331                 *extents = exts;
4332                 *nr_extents = nr;
4333         }
4334         return ret;
4335 }
4336
4337 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4338                                         struct btrfs_root *root,
4339                                         struct btrfs_path *path,
4340                                         struct btrfs_key *extent_key,
4341                                         struct btrfs_key *leaf_key,
4342                                         struct btrfs_ref_path *ref_path,
4343                                         struct disk_extent *new_extents,
4344                                         int nr_extents)
4345 {
4346         struct extent_buffer *leaf;
4347         struct btrfs_file_extent_item *fi;
4348         struct inode *inode = NULL;
4349         struct btrfs_key key;
4350         u64 lock_start = 0;
4351         u64 lock_end = 0;
4352         u64 num_bytes;
4353         u64 ext_offset;
4354         u64 search_end = (u64)-1;
4355         u32 nritems;
4356         int nr_scaned = 0;
4357         int extent_locked = 0;
4358         int extent_type;
4359         int ret;
4360
4361         memcpy(&key, leaf_key, sizeof(key));
4362         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4363                 if (key.objectid < ref_path->owner_objectid ||
4364                     (key.objectid == ref_path->owner_objectid &&
4365                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4366                         key.objectid = ref_path->owner_objectid;
4367                         key.type = BTRFS_EXTENT_DATA_KEY;
4368                         key.offset = 0;
4369                 }
4370         }
4371
4372         while (1) {
4373                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4374                 if (ret < 0)
4375                         goto out;
4376
4377                 leaf = path->nodes[0];
4378                 nritems = btrfs_header_nritems(leaf);
4379 next:
4380                 if (extent_locked && ret > 0) {
4381                         /*
4382                          * the file extent item was modified by someone
4383                          * before the extent got locked.
4384                          */
4385                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4386                                       lock_end, GFP_NOFS);
4387                         extent_locked = 0;
4388                 }
4389
4390                 if (path->slots[0] >= nritems) {
4391                         if (++nr_scaned > 2)
4392                                 break;
4393
4394                         BUG_ON(extent_locked);
4395                         ret = btrfs_next_leaf(root, path);
4396                         if (ret < 0)
4397                                 goto out;
4398                         if (ret > 0)
4399                                 break;
4400                         leaf = path->nodes[0];
4401                         nritems = btrfs_header_nritems(leaf);
4402                 }
4403
4404                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4405
4406                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4407                         if ((key.objectid > ref_path->owner_objectid) ||
4408                             (key.objectid == ref_path->owner_objectid &&
4409                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4410                             key.offset >= search_end)
4411                                 break;
4412                 }
4413
4414                 if (inode && key.objectid != inode->i_ino) {
4415                         BUG_ON(extent_locked);
4416                         btrfs_release_path(root, path);
4417                         mutex_unlock(&inode->i_mutex);
4418                         iput(inode);
4419                         inode = NULL;
4420                         continue;
4421                 }
4422
4423                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4424                         path->slots[0]++;
4425                         ret = 1;
4426                         goto next;
4427                 }
4428                 fi = btrfs_item_ptr(leaf, path->slots[0],
4429                                     struct btrfs_file_extent_item);
4430                 extent_type = btrfs_file_extent_type(leaf, fi);
4431                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4432                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4433                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4434                      extent_key->objectid)) {
4435                         path->slots[0]++;
4436                         ret = 1;
4437                         goto next;
4438                 }
4439
4440                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4441                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4442
4443                 if (search_end == (u64)-1) {
4444                         search_end = key.offset - ext_offset +
4445                                 btrfs_file_extent_ram_bytes(leaf, fi);
4446                 }
4447
4448                 if (!extent_locked) {
4449                         lock_start = key.offset;
4450                         lock_end = lock_start + num_bytes - 1;
4451                 } else {
4452                         if (lock_start > key.offset ||
4453                             lock_end + 1 < key.offset + num_bytes) {
4454                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4455                                               lock_start, lock_end, GFP_NOFS);
4456                                 extent_locked = 0;
4457                         }
4458                 }
4459
4460                 if (!inode) {
4461                         btrfs_release_path(root, path);
4462
4463                         inode = btrfs_iget_locked(root->fs_info->sb,
4464                                                   key.objectid, root);
4465                         if (inode->i_state & I_NEW) {
4466                                 BTRFS_I(inode)->root = root;
4467                                 BTRFS_I(inode)->location.objectid =
4468                                         key.objectid;
4469                                 BTRFS_I(inode)->location.type =
4470                                         BTRFS_INODE_ITEM_KEY;
4471                                 BTRFS_I(inode)->location.offset = 0;
4472                                 btrfs_read_locked_inode(inode);
4473                                 unlock_new_inode(inode);
4474                         }
4475                         /*
4476                          * some code call btrfs_commit_transaction while
4477                          * holding the i_mutex, so we can't use mutex_lock
4478                          * here.
4479                          */
4480                         if (is_bad_inode(inode) ||
4481                             !mutex_trylock(&inode->i_mutex)) {
4482                                 iput(inode);
4483                                 inode = NULL;
4484                                 key.offset = (u64)-1;
4485                                 goto skip;
4486                         }
4487                 }
4488
4489                 if (!extent_locked) {
4490                         struct btrfs_ordered_extent *ordered;
4491
4492                         btrfs_release_path(root, path);
4493
4494                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4495                                     lock_end, GFP_NOFS);
4496                         ordered = btrfs_lookup_first_ordered_extent(inode,
4497                                                                     lock_end);
4498                         if (ordered &&
4499                             ordered->file_offset <= lock_end &&
4500                             ordered->file_offset + ordered->len > lock_start) {
4501                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4502                                               lock_start, lock_end, GFP_NOFS);
4503                                 btrfs_start_ordered_extent(inode, ordered, 1);
4504                                 btrfs_put_ordered_extent(ordered);
4505                                 key.offset += num_bytes;
4506                                 goto skip;
4507                         }
4508                         if (ordered)
4509                                 btrfs_put_ordered_extent(ordered);
4510
4511                         extent_locked = 1;
4512                         continue;
4513                 }
4514
4515                 if (nr_extents == 1) {
4516                         /* update extent pointer in place */
4517                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4518                                                 new_extents[0].disk_bytenr);
4519                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4520                                                 new_extents[0].disk_num_bytes);
4521                         btrfs_mark_buffer_dirty(leaf);
4522
4523                         btrfs_drop_extent_cache(inode, key.offset,
4524                                                 key.offset + num_bytes - 1, 0);
4525
4526                         ret = btrfs_inc_extent_ref(trans, root,
4527                                                 new_extents[0].disk_bytenr,
4528                                                 new_extents[0].disk_num_bytes,
4529                                                 leaf->start,
4530                                                 root->root_key.objectid,
4531                                                 trans->transid,
4532                                                 key.objectid);
4533                         BUG_ON(ret);
4534
4535                         ret = btrfs_free_extent(trans, root,
4536                                                 extent_key->objectid,
4537                                                 extent_key->offset,
4538                                                 leaf->start,
4539                                                 btrfs_header_owner(leaf),
4540                                                 btrfs_header_generation(leaf),
4541                                                 key.objectid, 0);
4542                         BUG_ON(ret);
4543
4544                         btrfs_release_path(root, path);
4545                         key.offset += num_bytes;
4546                 } else {
4547                         BUG_ON(1);
4548 #if 0
4549                         u64 alloc_hint;
4550                         u64 extent_len;
4551                         int i;
4552                         /*
4553                          * drop old extent pointer at first, then insert the
4554                          * new pointers one bye one
4555                          */
4556                         btrfs_release_path(root, path);
4557                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4558                                                  key.offset + num_bytes,
4559                                                  key.offset, &alloc_hint);
4560                         BUG_ON(ret);
4561
4562                         for (i = 0; i < nr_extents; i++) {
4563                                 if (ext_offset >= new_extents[i].num_bytes) {
4564                                         ext_offset -= new_extents[i].num_bytes;
4565                                         continue;
4566                                 }
4567                                 extent_len = min(new_extents[i].num_bytes -
4568                                                  ext_offset, num_bytes);
4569
4570                                 ret = btrfs_insert_empty_item(trans, root,
4571                                                               path, &key,
4572                                                               sizeof(*fi));
4573                                 BUG_ON(ret);
4574
4575                                 leaf = path->nodes[0];
4576                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4577                                                 struct btrfs_file_extent_item);
4578                                 btrfs_set_file_extent_generation(leaf, fi,
4579                                                         trans->transid);
4580                                 btrfs_set_file_extent_type(leaf, fi,
4581                                                         BTRFS_FILE_EXTENT_REG);
4582                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4583                                                 new_extents[i].disk_bytenr);
4584                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4585                                                 new_extents[i].disk_num_bytes);
4586                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4587                                                 new_extents[i].ram_bytes);
4588
4589                                 btrfs_set_file_extent_compression(leaf, fi,
4590                                                 new_extents[i].compression);
4591                                 btrfs_set_file_extent_encryption(leaf, fi,
4592                                                 new_extents[i].encryption);
4593                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4594                                                 new_extents[i].other_encoding);
4595
4596                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4597                                                         extent_len);
4598                                 ext_offset += new_extents[i].offset;
4599                                 btrfs_set_file_extent_offset(leaf, fi,
4600                                                         ext_offset);
4601                                 btrfs_mark_buffer_dirty(leaf);
4602
4603                                 btrfs_drop_extent_cache(inode, key.offset,
4604                                                 key.offset + extent_len - 1, 0);
4605
4606                                 ret = btrfs_inc_extent_ref(trans, root,
4607                                                 new_extents[i].disk_bytenr,
4608                                                 new_extents[i].disk_num_bytes,
4609                                                 leaf->start,
4610                                                 root->root_key.objectid,
4611                                                 trans->transid, key.objectid);
4612                                 BUG_ON(ret);
4613                                 btrfs_release_path(root, path);
4614
4615                                 inode_add_bytes(inode, extent_len);
4616
4617                                 ext_offset = 0;
4618                                 num_bytes -= extent_len;
4619                                 key.offset += extent_len;
4620
4621                                 if (num_bytes == 0)
4622                                         break;
4623                         }
4624                         BUG_ON(i >= nr_extents);
4625 #endif
4626                 }
4627
4628                 if (extent_locked) {
4629                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4630                                       lock_end, GFP_NOFS);
4631                         extent_locked = 0;
4632                 }
4633 skip:
4634                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4635                     key.offset >= search_end)
4636                         break;
4637
4638                 cond_resched();
4639         }
4640         ret = 0;
4641 out:
4642         btrfs_release_path(root, path);
4643         if (inode) {
4644                 mutex_unlock(&inode->i_mutex);
4645                 if (extent_locked) {
4646                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4647                                       lock_end, GFP_NOFS);
4648                 }
4649                 iput(inode);
4650         }
4651         return ret;
4652 }
4653
4654 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4655                                struct btrfs_root *root,
4656                                struct extent_buffer *buf, u64 orig_start)
4657 {
4658         int level;
4659         int ret;
4660
4661         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4662         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4663
4664         level = btrfs_header_level(buf);
4665         if (level == 0) {
4666                 struct btrfs_leaf_ref *ref;
4667                 struct btrfs_leaf_ref *orig_ref;
4668
4669                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4670                 if (!orig_ref)
4671                         return -ENOENT;
4672
4673                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4674                 if (!ref) {
4675                         btrfs_free_leaf_ref(root, orig_ref);
4676                         return -ENOMEM;
4677                 }
4678
4679                 ref->nritems = orig_ref->nritems;
4680                 memcpy(ref->extents, orig_ref->extents,
4681                         sizeof(ref->extents[0]) * ref->nritems);
4682
4683                 btrfs_free_leaf_ref(root, orig_ref);
4684
4685                 ref->root_gen = trans->transid;
4686                 ref->bytenr = buf->start;
4687                 ref->owner = btrfs_header_owner(buf);
4688                 ref->generation = btrfs_header_generation(buf);
4689
4690                 ret = btrfs_add_leaf_ref(root, ref, 0);
4691                 WARN_ON(ret);
4692                 btrfs_free_leaf_ref(root, ref);
4693         }
4694         return 0;
4695 }
4696
4697 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4698                                         struct extent_buffer *leaf,
4699                                         struct btrfs_block_group_cache *group,
4700                                         struct btrfs_root *target_root)
4701 {
4702         struct btrfs_key key;
4703         struct inode *inode = NULL;
4704         struct btrfs_file_extent_item *fi;
4705         u64 num_bytes;
4706         u64 skip_objectid = 0;
4707         u32 nritems;
4708         u32 i;
4709
4710         nritems = btrfs_header_nritems(leaf);
4711         for (i = 0; i < nritems; i++) {
4712                 btrfs_item_key_to_cpu(leaf, &key, i);
4713                 if (key.objectid == skip_objectid ||
4714                     key.type != BTRFS_EXTENT_DATA_KEY)
4715                         continue;
4716                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4717                 if (btrfs_file_extent_type(leaf, fi) ==
4718                     BTRFS_FILE_EXTENT_INLINE)
4719                         continue;
4720                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4721                         continue;
4722                 if (!inode || inode->i_ino != key.objectid) {
4723                         iput(inode);
4724                         inode = btrfs_ilookup(target_root->fs_info->sb,
4725                                               key.objectid, target_root, 1);
4726                 }
4727                 if (!inode) {
4728                         skip_objectid = key.objectid;
4729                         continue;
4730                 }
4731                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4732
4733                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4734                             key.offset + num_bytes - 1, GFP_NOFS);
4735                 btrfs_drop_extent_cache(inode, key.offset,
4736                                         key.offset + num_bytes - 1, 1);
4737                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4738                               key.offset + num_bytes - 1, GFP_NOFS);
4739                 cond_resched();
4740         }
4741         iput(inode);
4742         return 0;
4743 }
4744
4745 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4746                                         struct btrfs_root *root,
4747                                         struct extent_buffer *leaf,
4748                                         struct btrfs_block_group_cache *group,
4749                                         struct inode *reloc_inode)
4750 {
4751         struct btrfs_key key;
4752         struct btrfs_key extent_key;
4753         struct btrfs_file_extent_item *fi;
4754         struct btrfs_leaf_ref *ref;
4755         struct disk_extent *new_extent;
4756         u64 bytenr;
4757         u64 num_bytes;
4758         u32 nritems;
4759         u32 i;
4760         int ext_index;
4761         int nr_extent;
4762         int ret;
4763
4764         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4765         BUG_ON(!new_extent);
4766
4767         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4768         BUG_ON(!ref);
4769
4770         ext_index = -1;
4771         nritems = btrfs_header_nritems(leaf);
4772         for (i = 0; i < nritems; i++) {
4773                 btrfs_item_key_to_cpu(leaf, &key, i);
4774                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4775                         continue;
4776                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4777                 if (btrfs_file_extent_type(leaf, fi) ==
4778                     BTRFS_FILE_EXTENT_INLINE)
4779                         continue;
4780                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4781                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4782                 if (bytenr == 0)
4783                         continue;
4784
4785                 ext_index++;
4786                 if (bytenr >= group->key.objectid + group->key.offset ||
4787                     bytenr + num_bytes <= group->key.objectid)
4788                         continue;
4789
4790                 extent_key.objectid = bytenr;
4791                 extent_key.offset = num_bytes;
4792                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4793                 nr_extent = 1;
4794                 ret = get_new_locations(reloc_inode, &extent_key,
4795                                         group->key.objectid, 1,
4796                                         &new_extent, &nr_extent);
4797                 if (ret > 0)
4798                         continue;
4799                 BUG_ON(ret < 0);
4800
4801                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4802                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4803                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4804                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4805
4806                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4807                                                 new_extent->disk_bytenr);
4808                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4809                                                 new_extent->disk_num_bytes);
4810                 btrfs_mark_buffer_dirty(leaf);
4811
4812                 ret = btrfs_inc_extent_ref(trans, root,
4813                                         new_extent->disk_bytenr,
4814                                         new_extent->disk_num_bytes,
4815                                         leaf->start,
4816                                         root->root_key.objectid,
4817                                         trans->transid, key.objectid);
4818                 BUG_ON(ret);
4819
4820                 ret = btrfs_free_extent(trans, root,
4821                                         bytenr, num_bytes, leaf->start,
4822                                         btrfs_header_owner(leaf),
4823                                         btrfs_header_generation(leaf),
4824                                         key.objectid, 0);
4825                 BUG_ON(ret);
4826                 cond_resched();
4827         }
4828         kfree(new_extent);
4829         BUG_ON(ext_index + 1 != ref->nritems);
4830         btrfs_free_leaf_ref(root, ref);
4831         return 0;
4832 }
4833
4834 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4835                           struct btrfs_root *root)
4836 {
4837         struct btrfs_root *reloc_root;
4838         int ret;
4839
4840         if (root->reloc_root) {
4841                 reloc_root = root->reloc_root;
4842                 root->reloc_root = NULL;
4843                 list_add(&reloc_root->dead_list,
4844                          &root->fs_info->dead_reloc_roots);
4845
4846                 btrfs_set_root_bytenr(&reloc_root->root_item,
4847                                       reloc_root->node->start);
4848                 btrfs_set_root_level(&root->root_item,
4849                                      btrfs_header_level(reloc_root->node));
4850                 memset(&reloc_root->root_item.drop_progress, 0,
4851                         sizeof(struct btrfs_disk_key));
4852                 reloc_root->root_item.drop_level = 0;
4853
4854                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4855                                         &reloc_root->root_key,
4856                                         &reloc_root->root_item);
4857                 BUG_ON(ret);
4858         }
4859         return 0;
4860 }
4861
4862 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4863 {
4864         struct btrfs_trans_handle *trans;
4865         struct btrfs_root *reloc_root;
4866         struct btrfs_root *prev_root = NULL;
4867         struct list_head dead_roots;
4868         int ret;
4869         unsigned long nr;
4870
4871         INIT_LIST_HEAD(&dead_roots);
4872         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4873
4874         while (!list_empty(&dead_roots)) {
4875                 reloc_root = list_entry(dead_roots.prev,
4876                                         struct btrfs_root, dead_list);
4877                 list_del_init(&reloc_root->dead_list);
4878
4879                 BUG_ON(reloc_root->commit_root != NULL);
4880                 while (1) {
4881                         trans = btrfs_join_transaction(root, 1);
4882                         BUG_ON(!trans);
4883
4884                         mutex_lock(&root->fs_info->drop_mutex);
4885                         ret = btrfs_drop_snapshot(trans, reloc_root);
4886                         if (ret != -EAGAIN)
4887                                 break;
4888                         mutex_unlock(&root->fs_info->drop_mutex);
4889
4890                         nr = trans->blocks_used;
4891                         ret = btrfs_end_transaction(trans, root);
4892                         BUG_ON(ret);
4893                         btrfs_btree_balance_dirty(root, nr);
4894                 }
4895
4896                 free_extent_buffer(reloc_root->node);
4897
4898                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4899                                      &reloc_root->root_key);
4900                 BUG_ON(ret);
4901                 mutex_unlock(&root->fs_info->drop_mutex);
4902
4903                 nr = trans->blocks_used;
4904                 ret = btrfs_end_transaction(trans, root);
4905                 BUG_ON(ret);
4906                 btrfs_btree_balance_dirty(root, nr);
4907
4908                 kfree(prev_root);
4909                 prev_root = reloc_root;
4910         }
4911         if (prev_root) {
4912                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4913                 kfree(prev_root);
4914         }
4915         return 0;
4916 }
4917
4918 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4919 {
4920         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4921         return 0;
4922 }
4923
4924 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4925 {
4926         struct btrfs_root *reloc_root;
4927         struct btrfs_trans_handle *trans;
4928         struct btrfs_key location;
4929         int found;
4930         int ret;
4931
4932         mutex_lock(&root->fs_info->tree_reloc_mutex);
4933         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4934         BUG_ON(ret);
4935         found = !list_empty(&root->fs_info->dead_reloc_roots);
4936         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4937
4938         if (found) {
4939                 trans = btrfs_start_transaction(root, 1);
4940                 BUG_ON(!trans);
4941                 ret = btrfs_commit_transaction(trans, root);
4942                 BUG_ON(ret);
4943         }
4944
4945         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4946         location.offset = (u64)-1;
4947         location.type = BTRFS_ROOT_ITEM_KEY;
4948
4949         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4950         BUG_ON(!reloc_root);
4951         btrfs_orphan_cleanup(reloc_root);
4952         return 0;
4953 }
4954
4955 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
4956                                     struct btrfs_root *root)
4957 {
4958         struct btrfs_root *reloc_root;
4959         struct extent_buffer *eb;
4960         struct btrfs_root_item *root_item;
4961         struct btrfs_key root_key;
4962         int ret;
4963
4964         BUG_ON(!root->ref_cows);
4965         if (root->reloc_root)
4966                 return 0;
4967
4968         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4969         BUG_ON(!root_item);
4970
4971         ret = btrfs_copy_root(trans, root, root->commit_root,
4972                               &eb, BTRFS_TREE_RELOC_OBJECTID);
4973         BUG_ON(ret);
4974
4975         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4976         root_key.offset = root->root_key.objectid;
4977         root_key.type = BTRFS_ROOT_ITEM_KEY;
4978
4979         memcpy(root_item, &root->root_item, sizeof(root_item));
4980         btrfs_set_root_refs(root_item, 0);
4981         btrfs_set_root_bytenr(root_item, eb->start);
4982         btrfs_set_root_level(root_item, btrfs_header_level(eb));
4983         btrfs_set_root_generation(root_item, trans->transid);
4984
4985         btrfs_tree_unlock(eb);
4986         free_extent_buffer(eb);
4987
4988         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4989                                 &root_key, root_item);
4990         BUG_ON(ret);
4991         kfree(root_item);
4992
4993         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4994                                                  &root_key);
4995         BUG_ON(!reloc_root);
4996         reloc_root->last_trans = trans->transid;
4997         reloc_root->commit_root = NULL;
4998         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4999
5000         root->reloc_root = reloc_root;
5001         return 0;
5002 }
5003
5004 /*
5005  * Core function of space balance.
5006  *
5007  * The idea is using reloc trees to relocate tree blocks in reference
5008  * counted roots. There is one reloc tree for each subvol, and all
5009  * reloc trees share same root key objectid. Reloc trees are snapshots
5010  * of the latest committed roots of subvols (root->commit_root).
5011  *
5012  * To relocate a tree block referenced by a subvol, there are two steps.
5013  * COW the block through subvol's reloc tree, then update block pointer
5014  * in the subvol to point to the new block. Since all reloc trees share
5015  * same root key objectid, doing special handing for tree blocks owned
5016  * by them is easy. Once a tree block has been COWed in one reloc tree,
5017  * we can use the resulting new block directly when the same block is
5018  * required to COW again through other reloc trees. By this way, relocated
5019  * tree blocks are shared between reloc trees, so they are also shared
5020  * between subvols.
5021  */
5022 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5023                                       struct btrfs_root *root,
5024                                       struct btrfs_path *path,
5025                                       struct btrfs_key *first_key,
5026                                       struct btrfs_ref_path *ref_path,
5027                                       struct btrfs_block_group_cache *group,
5028                                       struct inode *reloc_inode)
5029 {
5030         struct btrfs_root *reloc_root;
5031         struct extent_buffer *eb = NULL;
5032         struct btrfs_key *keys;
5033         u64 *nodes;
5034         int level;
5035         int shared_level;
5036         int lowest_level = 0;
5037         int ret;
5038
5039         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5040                 lowest_level = ref_path->owner_objectid;
5041
5042         if (!root->ref_cows) {
5043                 path->lowest_level = lowest_level;
5044                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5045                 BUG_ON(ret < 0);
5046                 path->lowest_level = 0;
5047                 btrfs_release_path(root, path);
5048                 return 0;
5049         }
5050
5051         mutex_lock(&root->fs_info->tree_reloc_mutex);
5052         ret = init_reloc_tree(trans, root);
5053         BUG_ON(ret);
5054         reloc_root = root->reloc_root;
5055
5056         shared_level = ref_path->shared_level;
5057         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5058
5059         keys = ref_path->node_keys;
5060         nodes = ref_path->new_nodes;
5061         memset(&keys[shared_level + 1], 0,
5062                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5063         memset(&nodes[shared_level + 1], 0,
5064                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5065
5066         if (nodes[lowest_level] == 0) {
5067                 path->lowest_level = lowest_level;
5068                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5069                                         0, 1);
5070                 BUG_ON(ret);
5071                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5072                         eb = path->nodes[level];
5073                         if (!eb || eb == reloc_root->node)
5074                                 break;
5075                         nodes[level] = eb->start;
5076                         if (level == 0)
5077                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5078                         else
5079                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5080                 }
5081                 if (nodes[0] &&
5082                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5083                         eb = path->nodes[0];
5084                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5085                                                       group, reloc_inode);
5086                         BUG_ON(ret);
5087                 }
5088                 btrfs_release_path(reloc_root, path);
5089         } else {
5090                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5091                                        lowest_level);
5092                 BUG_ON(ret);
5093         }
5094
5095         /*
5096          * replace tree blocks in the fs tree with tree blocks in
5097          * the reloc tree.
5098          */
5099         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5100         BUG_ON(ret < 0);
5101
5102         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5103                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5104                                         0, 0);
5105                 BUG_ON(ret);
5106                 extent_buffer_get(path->nodes[0]);
5107                 eb = path->nodes[0];
5108                 btrfs_release_path(reloc_root, path);
5109                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5110                 BUG_ON(ret);
5111                 free_extent_buffer(eb);
5112         }
5113
5114         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5115         path->lowest_level = 0;
5116         return 0;
5117 }
5118
5119 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5120                                         struct btrfs_root *root,
5121                                         struct btrfs_path *path,
5122                                         struct btrfs_key *first_key,
5123                                         struct btrfs_ref_path *ref_path)
5124 {
5125         int ret;
5126
5127         ret = relocate_one_path(trans, root, path, first_key,
5128                                 ref_path, NULL, NULL);
5129         BUG_ON(ret);
5130
5131         return 0;
5132 }
5133
5134 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5135                                     struct btrfs_root *extent_root,
5136                                     struct btrfs_path *path,
5137                                     struct btrfs_key *extent_key)
5138 {
5139         int ret;
5140
5141         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5142         if (ret)
5143                 goto out;
5144         ret = btrfs_del_item(trans, extent_root, path);
5145 out:
5146         btrfs_release_path(extent_root, path);
5147         return ret;
5148 }
5149
5150 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5151                                                 struct btrfs_ref_path *ref_path)
5152 {
5153         struct btrfs_key root_key;
5154
5155         root_key.objectid = ref_path->root_objectid;
5156         root_key.type = BTRFS_ROOT_ITEM_KEY;
5157         if (is_cowonly_root(ref_path->root_objectid))
5158                 root_key.offset = 0;
5159         else
5160                 root_key.offset = (u64)-1;
5161
5162         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5163 }
5164
5165 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5166                                         struct btrfs_path *path,
5167                                         struct btrfs_key *extent_key,
5168                                         struct btrfs_block_group_cache *group,
5169                                         struct inode *reloc_inode, int pass)
5170 {
5171         struct btrfs_trans_handle *trans;
5172         struct btrfs_root *found_root;
5173         struct btrfs_ref_path *ref_path = NULL;
5174         struct disk_extent *new_extents = NULL;
5175         int nr_extents = 0;
5176         int loops;
5177         int ret;
5178         int level;
5179         struct btrfs_key first_key;
5180         u64 prev_block = 0;
5181
5182
5183         trans = btrfs_start_transaction(extent_root, 1);
5184         BUG_ON(!trans);
5185
5186         if (extent_key->objectid == 0) {
5187                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5188                 goto out;
5189         }
5190
5191         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5192         if (!ref_path) {
5193                 ret = -ENOMEM;
5194                 goto out;
5195         }
5196
5197         for (loops = 0; ; loops++) {
5198                 if (loops == 0) {
5199                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5200                                                    extent_key->objectid);
5201                 } else {
5202                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5203                 }
5204                 if (ret < 0)
5205                         goto out;
5206                 if (ret > 0)
5207                         break;
5208
5209                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5210                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5211                         continue;
5212
5213                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5214                 BUG_ON(!found_root);
5215                 /*
5216                  * for reference counted tree, only process reference paths
5217                  * rooted at the latest committed root.
5218                  */
5219                 if (found_root->ref_cows &&
5220                     ref_path->root_generation != found_root->root_key.offset)
5221                         continue;
5222
5223                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5224                         if (pass == 0) {
5225                                 /*
5226                                  * copy data extents to new locations
5227                                  */
5228                                 u64 group_start = group->key.objectid;
5229                                 ret = relocate_data_extent(reloc_inode,
5230                                                            extent_key,
5231                                                            group_start);
5232                                 if (ret < 0)
5233                                         goto out;
5234                                 break;
5235                         }
5236                         level = 0;
5237                 } else {
5238                         level = ref_path->owner_objectid;
5239                 }
5240
5241                 if (prev_block != ref_path->nodes[level]) {
5242                         struct extent_buffer *eb;
5243                         u64 block_start = ref_path->nodes[level];
5244                         u64 block_size = btrfs_level_size(found_root, level);
5245
5246                         eb = read_tree_block(found_root, block_start,
5247                                              block_size, 0);
5248                         btrfs_tree_lock(eb);
5249                         BUG_ON(level != btrfs_header_level(eb));
5250
5251                         if (level == 0)
5252                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5253                         else
5254                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5255
5256                         btrfs_tree_unlock(eb);
5257                         free_extent_buffer(eb);
5258                         prev_block = block_start;
5259                 }
5260
5261                 mutex_lock(&extent_root->fs_info->trans_mutex);
5262                 btrfs_record_root_in_trans(found_root);
5263                 mutex_unlock(&extent_root->fs_info->trans_mutex);
5264                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5265                         /*
5266                          * try to update data extent references while
5267                          * keeping metadata shared between snapshots.
5268                          */
5269                         if (pass == 1) {
5270                                 ret = relocate_one_path(trans, found_root,
5271                                                 path, &first_key, ref_path,
5272                                                 group, reloc_inode);
5273                                 if (ret < 0)
5274                                         goto out;
5275                                 continue;
5276                         }
5277                         /*
5278                          * use fallback method to process the remaining
5279                          * references.
5280                          */
5281                         if (!new_extents) {
5282                                 u64 group_start = group->key.objectid;
5283                                 new_extents = kmalloc(sizeof(*new_extents),
5284                                                       GFP_NOFS);
5285                                 nr_extents = 1;
5286                                 ret = get_new_locations(reloc_inode,
5287                                                         extent_key,
5288                                                         group_start, 1,
5289                                                         &new_extents,
5290                                                         &nr_extents);
5291                                 if (ret)
5292                                         goto out;
5293                         }
5294                         ret = replace_one_extent(trans, found_root,
5295                                                 path, extent_key,
5296                                                 &first_key, ref_path,
5297                                                 new_extents, nr_extents);
5298                 } else {
5299                         ret = relocate_tree_block(trans, found_root, path,
5300                                                   &first_key, ref_path);
5301                 }
5302                 if (ret < 0)
5303                         goto out;
5304         }
5305         ret = 0;
5306 out:
5307         btrfs_end_transaction(trans, extent_root);
5308         kfree(new_extents);
5309         kfree(ref_path);
5310         return ret;
5311 }
5312
5313 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5314 {
5315         u64 num_devices;
5316         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5317                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5318
5319         num_devices = root->fs_info->fs_devices->rw_devices;
5320         if (num_devices == 1) {
5321                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5322                 stripped = flags & ~stripped;
5323
5324                 /* turn raid0 into single device chunks */
5325                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5326                         return stripped;
5327
5328                 /* turn mirroring into duplication */
5329                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5330                              BTRFS_BLOCK_GROUP_RAID10))
5331                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5332                 return flags;
5333         } else {
5334                 /* they already had raid on here, just return */
5335                 if (flags & stripped)
5336                         return flags;
5337
5338                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5339                 stripped = flags & ~stripped;
5340
5341                 /* switch duplicated blocks with raid1 */
5342                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5343                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5344
5345                 /* turn single device chunks into raid0 */
5346                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5347         }
5348         return flags;
5349 }
5350
5351 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5352                      struct btrfs_block_group_cache *shrink_block_group,
5353                      int force)
5354 {
5355         struct btrfs_trans_handle *trans;
5356         u64 new_alloc_flags;
5357         u64 calc;
5358
5359         spin_lock(&shrink_block_group->lock);
5360         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5361                 spin_unlock(&shrink_block_group->lock);
5362
5363                 trans = btrfs_start_transaction(root, 1);
5364                 spin_lock(&shrink_block_group->lock);
5365
5366                 new_alloc_flags = update_block_group_flags(root,
5367                                                    shrink_block_group->flags);
5368                 if (new_alloc_flags != shrink_block_group->flags) {
5369                         calc =
5370                              btrfs_block_group_used(&shrink_block_group->item);
5371                 } else {
5372                         calc = shrink_block_group->key.offset;
5373                 }
5374                 spin_unlock(&shrink_block_group->lock);
5375
5376                 do_chunk_alloc(trans, root->fs_info->extent_root,
5377                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5378
5379                 btrfs_end_transaction(trans, root);
5380         } else
5381                 spin_unlock(&shrink_block_group->lock);
5382         return 0;
5383 }
5384
5385 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5386                                  struct btrfs_root *root,
5387                                  u64 objectid, u64 size)
5388 {
5389         struct btrfs_path *path;
5390         struct btrfs_inode_item *item;
5391         struct extent_buffer *leaf;
5392         int ret;
5393
5394         path = btrfs_alloc_path();
5395         if (!path)
5396                 return -ENOMEM;
5397
5398         path->leave_spinning = 1;
5399         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5400         if (ret)
5401                 goto out;
5402
5403         leaf = path->nodes[0];
5404         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5405         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5406         btrfs_set_inode_generation(leaf, item, 1);
5407         btrfs_set_inode_size(leaf, item, size);
5408         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5409         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5410         btrfs_mark_buffer_dirty(leaf);
5411         btrfs_release_path(root, path);
5412 out:
5413         btrfs_free_path(path);
5414         return ret;
5415 }
5416
5417 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5418                                         struct btrfs_block_group_cache *group)
5419 {
5420         struct inode *inode = NULL;
5421         struct btrfs_trans_handle *trans;
5422         struct btrfs_root *root;
5423         struct btrfs_key root_key;
5424         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5425         int err = 0;
5426
5427         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5428         root_key.type = BTRFS_ROOT_ITEM_KEY;
5429         root_key.offset = (u64)-1;
5430         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5431         if (IS_ERR(root))
5432                 return ERR_CAST(root);
5433
5434         trans = btrfs_start_transaction(root, 1);
5435         BUG_ON(!trans);
5436
5437         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5438         if (err)
5439                 goto out;
5440
5441         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5442         BUG_ON(err);
5443
5444         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5445                                        group->key.offset, 0, group->key.offset,
5446                                        0, 0, 0);
5447         BUG_ON(err);
5448
5449         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5450         if (inode->i_state & I_NEW) {
5451                 BTRFS_I(inode)->root = root;
5452                 BTRFS_I(inode)->location.objectid = objectid;
5453                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5454                 BTRFS_I(inode)->location.offset = 0;
5455                 btrfs_read_locked_inode(inode);
5456                 unlock_new_inode(inode);
5457                 BUG_ON(is_bad_inode(inode));
5458         } else {
5459                 BUG_ON(1);
5460         }
5461         BTRFS_I(inode)->index_cnt = group->key.objectid;
5462
5463         err = btrfs_orphan_add(trans, inode);
5464 out:
5465         btrfs_end_transaction(trans, root);
5466         if (err) {
5467                 if (inode)
5468                         iput(inode);
5469                 inode = ERR_PTR(err);
5470         }
5471         return inode;
5472 }
5473
5474 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5475 {
5476
5477         struct btrfs_ordered_sum *sums;
5478         struct btrfs_sector_sum *sector_sum;
5479         struct btrfs_ordered_extent *ordered;
5480         struct btrfs_root *root = BTRFS_I(inode)->root;
5481         struct list_head list;
5482         size_t offset;
5483         int ret;
5484         u64 disk_bytenr;
5485
5486         INIT_LIST_HEAD(&list);
5487
5488         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5489         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5490
5491         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5492         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5493                                        disk_bytenr + len - 1, &list);
5494
5495         while (!list_empty(&list)) {
5496                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5497                 list_del_init(&sums->list);
5498
5499                 sector_sum = sums->sums;
5500                 sums->bytenr = ordered->start;
5501
5502                 offset = 0;
5503                 while (offset < sums->len) {
5504                         sector_sum->bytenr += ordered->start - disk_bytenr;
5505                         sector_sum++;
5506                         offset += root->sectorsize;
5507                 }
5508
5509                 btrfs_add_ordered_sum(inode, ordered, sums);
5510         }
5511         btrfs_put_ordered_extent(ordered);
5512         return 0;
5513 }
5514
5515 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5516 {
5517         struct btrfs_trans_handle *trans;
5518         struct btrfs_path *path;
5519         struct btrfs_fs_info *info = root->fs_info;
5520         struct extent_buffer *leaf;
5521         struct inode *reloc_inode;
5522         struct btrfs_block_group_cache *block_group;
5523         struct btrfs_key key;
5524         u64 skipped;
5525         u64 cur_byte;
5526         u64 total_found;
5527         u32 nritems;
5528         int ret;
5529         int progress;
5530         int pass = 0;
5531
5532         root = root->fs_info->extent_root;
5533
5534         block_group = btrfs_lookup_block_group(info, group_start);
5535         BUG_ON(!block_group);
5536
5537         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5538                (unsigned long long)block_group->key.objectid,
5539                (unsigned long long)block_group->flags);
5540
5541         path = btrfs_alloc_path();
5542         BUG_ON(!path);
5543
5544         reloc_inode = create_reloc_inode(info, block_group);
5545         BUG_ON(IS_ERR(reloc_inode));
5546
5547         __alloc_chunk_for_shrink(root, block_group, 1);
5548         set_block_group_readonly(block_group);
5549
5550         btrfs_start_delalloc_inodes(info->tree_root);
5551         btrfs_wait_ordered_extents(info->tree_root, 0);
5552 again:
5553         skipped = 0;
5554         total_found = 0;
5555         progress = 0;
5556         key.objectid = block_group->key.objectid;
5557         key.offset = 0;
5558         key.type = 0;
5559         cur_byte = key.objectid;
5560
5561         trans = btrfs_start_transaction(info->tree_root, 1);
5562         btrfs_commit_transaction(trans, info->tree_root);
5563
5564         mutex_lock(&root->fs_info->cleaner_mutex);
5565         btrfs_clean_old_snapshots(info->tree_root);
5566         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5567         mutex_unlock(&root->fs_info->cleaner_mutex);
5568
5569         trans = btrfs_start_transaction(info->tree_root, 1);
5570         btrfs_commit_transaction(trans, info->tree_root);
5571
5572         while (1) {
5573                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5574                 if (ret < 0)
5575                         goto out;
5576 next:
5577                 leaf = path->nodes[0];
5578                 nritems = btrfs_header_nritems(leaf);
5579                 if (path->slots[0] >= nritems) {
5580                         ret = btrfs_next_leaf(root, path);
5581                         if (ret < 0)
5582                                 goto out;
5583                         if (ret == 1) {
5584                                 ret = 0;
5585                                 break;
5586                         }
5587                         leaf = path->nodes[0];
5588                         nritems = btrfs_header_nritems(leaf);
5589                 }
5590
5591                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5592
5593                 if (key.objectid >= block_group->key.objectid +
5594                     block_group->key.offset)
5595                         break;
5596
5597                 if (progress && need_resched()) {
5598                         btrfs_release_path(root, path);
5599                         cond_resched();
5600                         progress = 0;
5601                         continue;
5602                 }
5603                 progress = 1;
5604
5605                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5606                     key.objectid + key.offset <= cur_byte) {
5607                         path->slots[0]++;
5608                         goto next;
5609                 }
5610
5611                 total_found++;
5612                 cur_byte = key.objectid + key.offset;
5613                 btrfs_release_path(root, path);
5614
5615                 __alloc_chunk_for_shrink(root, block_group, 0);
5616                 ret = relocate_one_extent(root, path, &key, block_group,
5617                                           reloc_inode, pass);
5618                 BUG_ON(ret < 0);
5619                 if (ret > 0)
5620                         skipped++;
5621
5622                 key.objectid = cur_byte;
5623                 key.type = 0;
5624                 key.offset = 0;
5625         }
5626
5627         btrfs_release_path(root, path);
5628
5629         if (pass == 0) {
5630                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5631                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5632         }
5633
5634         if (total_found > 0) {
5635                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5636                        (unsigned long long)total_found, pass);
5637                 pass++;
5638                 if (total_found == skipped && pass > 2) {
5639                         iput(reloc_inode);
5640                         reloc_inode = create_reloc_inode(info, block_group);
5641                         pass = 0;
5642                 }
5643                 goto again;
5644         }
5645
5646         /* delete reloc_inode */
5647         iput(reloc_inode);
5648
5649         /* unpin extents in this range */
5650         trans = btrfs_start_transaction(info->tree_root, 1);
5651         btrfs_commit_transaction(trans, info->tree_root);
5652
5653         spin_lock(&block_group->lock);
5654         WARN_ON(block_group->pinned > 0);
5655         WARN_ON(block_group->reserved > 0);
5656         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5657         spin_unlock(&block_group->lock);
5658         put_block_group(block_group);
5659         ret = 0;
5660 out:
5661         btrfs_free_path(path);
5662         return ret;
5663 }
5664
5665 static int find_first_block_group(struct btrfs_root *root,
5666                 struct btrfs_path *path, struct btrfs_key *key)
5667 {
5668         int ret = 0;
5669         struct btrfs_key found_key;
5670         struct extent_buffer *leaf;
5671         int slot;
5672
5673         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5674         if (ret < 0)
5675                 goto out;
5676
5677         while (1) {
5678                 slot = path->slots[0];
5679                 leaf = path->nodes[0];
5680                 if (slot >= btrfs_header_nritems(leaf)) {
5681                         ret = btrfs_next_leaf(root, path);
5682                         if (ret == 0)
5683                                 continue;
5684                         if (ret < 0)
5685                                 goto out;
5686                         break;
5687                 }
5688                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5689
5690                 if (found_key.objectid >= key->objectid &&
5691                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5692                         ret = 0;
5693                         goto out;
5694                 }
5695                 path->slots[0]++;
5696         }
5697         ret = -ENOENT;
5698 out:
5699         return ret;
5700 }
5701
5702 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5703 {
5704         struct btrfs_block_group_cache *block_group;
5705         struct btrfs_space_info *space_info;
5706         struct rb_node *n;
5707
5708         spin_lock(&info->block_group_cache_lock);
5709         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5710                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5711                                        cache_node);
5712                 rb_erase(&block_group->cache_node,
5713                          &info->block_group_cache_tree);
5714                 spin_unlock(&info->block_group_cache_lock);
5715
5716                 btrfs_remove_free_space_cache(block_group);
5717                 down_write(&block_group->space_info->groups_sem);
5718                 list_del(&block_group->list);
5719                 up_write(&block_group->space_info->groups_sem);
5720
5721                 WARN_ON(atomic_read(&block_group->count) != 1);
5722                 kfree(block_group);
5723
5724                 spin_lock(&info->block_group_cache_lock);
5725         }
5726         spin_unlock(&info->block_group_cache_lock);
5727
5728         /* now that all the block groups are freed, go through and
5729          * free all the space_info structs.  This is only called during
5730          * the final stages of unmount, and so we know nobody is
5731          * using them.  We call synchronize_rcu() once before we start,
5732          * just to be on the safe side.
5733          */
5734         synchronize_rcu();
5735
5736         while(!list_empty(&info->space_info)) {
5737                 space_info = list_entry(info->space_info.next,
5738                                         struct btrfs_space_info,
5739                                         list);
5740
5741                 list_del(&space_info->list);
5742                 kfree(space_info);
5743         }
5744         return 0;
5745 }
5746
5747 int btrfs_read_block_groups(struct btrfs_root *root)
5748 {
5749         struct btrfs_path *path;
5750         int ret;
5751         struct btrfs_block_group_cache *cache;
5752         struct btrfs_fs_info *info = root->fs_info;
5753         struct btrfs_space_info *space_info;
5754         struct btrfs_key key;
5755         struct btrfs_key found_key;
5756         struct extent_buffer *leaf;
5757
5758         root = info->extent_root;
5759         key.objectid = 0;
5760         key.offset = 0;
5761         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5762         path = btrfs_alloc_path();
5763         if (!path)
5764                 return -ENOMEM;
5765
5766         while (1) {
5767                 ret = find_first_block_group(root, path, &key);
5768                 if (ret > 0) {
5769                         ret = 0;
5770                         goto error;
5771                 }
5772                 if (ret != 0)
5773                         goto error;
5774
5775                 leaf = path->nodes[0];
5776                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5777                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5778                 if (!cache) {
5779                         ret = -ENOMEM;
5780                         break;
5781                 }
5782
5783                 atomic_set(&cache->count, 1);
5784                 spin_lock_init(&cache->lock);
5785                 spin_lock_init(&cache->tree_lock);
5786                 mutex_init(&cache->cache_mutex);
5787                 INIT_LIST_HEAD(&cache->list);
5788                 read_extent_buffer(leaf, &cache->item,
5789                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5790                                    sizeof(cache->item));
5791                 memcpy(&cache->key, &found_key, sizeof(found_key));
5792
5793                 key.objectid = found_key.objectid + found_key.offset;
5794                 btrfs_release_path(root, path);
5795                 cache->flags = btrfs_block_group_flags(&cache->item);
5796
5797                 ret = update_space_info(info, cache->flags, found_key.offset,
5798                                         btrfs_block_group_used(&cache->item),
5799                                         &space_info);
5800                 BUG_ON(ret);
5801                 cache->space_info = space_info;
5802                 down_write(&space_info->groups_sem);
5803                 list_add_tail(&cache->list, &space_info->block_groups);
5804                 up_write(&space_info->groups_sem);
5805
5806                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5807                 BUG_ON(ret);
5808
5809                 set_avail_alloc_bits(root->fs_info, cache->flags);
5810                 if (btrfs_chunk_readonly(root, cache->key.objectid))
5811                         set_block_group_readonly(cache);
5812         }
5813         ret = 0;
5814 error:
5815         btrfs_free_path(path);
5816         return ret;
5817 }
5818
5819 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5820                            struct btrfs_root *root, u64 bytes_used,
5821                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5822                            u64 size)
5823 {
5824         int ret;
5825         struct btrfs_root *extent_root;
5826         struct btrfs_block_group_cache *cache;
5827
5828         extent_root = root->fs_info->extent_root;
5829
5830         root->fs_info->last_trans_log_full_commit = trans->transid;
5831
5832         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5833         if (!cache)
5834                 return -ENOMEM;
5835
5836         cache->key.objectid = chunk_offset;
5837         cache->key.offset = size;
5838         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5839         atomic_set(&cache->count, 1);
5840         spin_lock_init(&cache->lock);
5841         spin_lock_init(&cache->tree_lock);
5842         mutex_init(&cache->cache_mutex);
5843         INIT_LIST_HEAD(&cache->list);
5844
5845         btrfs_set_block_group_used(&cache->item, bytes_used);
5846         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5847         cache->flags = type;
5848         btrfs_set_block_group_flags(&cache->item, type);
5849
5850         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5851                                 &cache->space_info);
5852         BUG_ON(ret);
5853         down_write(&cache->space_info->groups_sem);
5854         list_add_tail(&cache->list, &cache->space_info->block_groups);
5855         up_write(&cache->space_info->groups_sem);
5856
5857         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5858         BUG_ON(ret);
5859
5860         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5861                                 sizeof(cache->item));
5862         BUG_ON(ret);
5863
5864         set_avail_alloc_bits(extent_root->fs_info, type);
5865
5866         return 0;
5867 }
5868
5869 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5870                              struct btrfs_root *root, u64 group_start)
5871 {
5872         struct btrfs_path *path;
5873         struct btrfs_block_group_cache *block_group;
5874         struct btrfs_key key;
5875         int ret;
5876
5877         root = root->fs_info->extent_root;
5878
5879         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5880         BUG_ON(!block_group);
5881         BUG_ON(!block_group->ro);
5882
5883         memcpy(&key, &block_group->key, sizeof(key));
5884
5885         path = btrfs_alloc_path();
5886         BUG_ON(!path);
5887
5888         spin_lock(&root->fs_info->block_group_cache_lock);
5889         rb_erase(&block_group->cache_node,
5890                  &root->fs_info->block_group_cache_tree);
5891         spin_unlock(&root->fs_info->block_group_cache_lock);
5892         btrfs_remove_free_space_cache(block_group);
5893         down_write(&block_group->space_info->groups_sem);
5894         list_del(&block_group->list);
5895         up_write(&block_group->space_info->groups_sem);
5896
5897         spin_lock(&block_group->space_info->lock);
5898         block_group->space_info->total_bytes -= block_group->key.offset;
5899         block_group->space_info->bytes_readonly -= block_group->key.offset;
5900         spin_unlock(&block_group->space_info->lock);
5901         block_group->space_info->full = 0;
5902
5903         put_block_group(block_group);
5904         put_block_group(block_group);
5905
5906         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5907         if (ret > 0)
5908                 ret = -EIO;
5909         if (ret < 0)
5910                 goto out;
5911
5912         ret = btrfs_del_item(trans, root, path);
5913 out:
5914         btrfs_free_path(path);
5915         return ret;
5916 }