git.oblomov.eu Git - linux-2.6/blob - 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         remove_sb_from_cache(root, block_group);
 295         block_group->cached = 1;
 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                 }
2388
2389                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2390                                          mark_free);
2391                 BUG_ON(ret);
2392         }
2393         btrfs_free_path(path);
2394         return ret;
2395 }
2396
2397 /*
2398  * remove an extent from the root, returns 0 on success
2399  */
2400 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2401                                         struct btrfs_root *root,
2402                                         u64 bytenr, u64 num_bytes, u64 parent,
2403                                         u64 root_objectid, u64 ref_generation,
2404                                         u64 owner_objectid, int pin,
2405                                         int refs_to_drop)
2406 {
2407         WARN_ON(num_bytes < root->sectorsize);
2408
2409         /*
2410          * if metadata always pin
2411          * if data pin when any transaction has committed this
2412          */
2413         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID ||
2414             ref_generation != trans->transid)
2415                 pin = 1;
2416
2417         if (ref_generation != trans->transid)
2418                 pin = 1;
2419
2420         return __free_extent(trans, root, bytenr, num_bytes, parent,
2421                             root_objectid, ref_generation,
2422                             owner_objectid, pin, pin == 0, refs_to_drop);
2423 }
2424
2425 /*
2426  * when we free an extent, it is possible (and likely) that we free the last
2427  * delayed ref for that extent as well.  This searches the delayed ref tree for
2428  * a given extent, and if there are no other delayed refs to be processed, it
2429  * removes it from the tree.
2430  */
2431 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
2432                                       struct btrfs_root *root, u64 bytenr)
2433 {
2434         struct btrfs_delayed_ref_head *head;
2435         struct btrfs_delayed_ref_root *delayed_refs;
2436         struct btrfs_delayed_ref_node *ref;
2437         struct rb_node *node;
2438         int ret;
2439
2440         delayed_refs = &trans->transaction->delayed_refs;
2441         spin_lock(&delayed_refs->lock);
2442         head = btrfs_find_delayed_ref_head(trans, bytenr);
2443         if (!head)
2444                 goto out;
2445
2446         node = rb_prev(&head->node.rb_node);
2447         if (!node)
2448                 goto out;
2449
2450         ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2451
2452         /* there are still entries for this ref, we can't drop it */
2453         if (ref->bytenr == bytenr)
2454                 goto out;
2455
2456         /*
2457          * waiting for the lock here would deadlock.  If someone else has it
2458          * locked they are already in the process of dropping it anyway
2459          */
2460         if (!mutex_trylock(&head->mutex))
2461                 goto out;
2462
2463         /*
2464          * at this point we have a head with no other entries.  Go
2465          * ahead and process it.
2466          */
2467         head->node.in_tree = 0;
2468         rb_erase(&head->node.rb_node, &delayed_refs->root);
2469
2470         delayed_refs->num_entries--;
2471
2472         /*
2473          * we don't take a ref on the node because we're removing it from the
2474          * tree, so we just steal the ref the tree was holding.
2475          */
2476         delayed_refs->num_heads--;
2477         if (list_empty(&head->cluster))
2478                 delayed_refs->num_heads_ready--;
2479
2480         list_del_init(&head->cluster);
2481         spin_unlock(&delayed_refs->lock);
2482
2483         ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
2484                                   &head->node, head->must_insert_reserved);
2485         BUG_ON(ret);
2486         btrfs_put_delayed_ref(&head->node);
2487         return 0;
2488 out:
2489         spin_unlock(&delayed_refs->lock);
2490         return 0;
2491 }
2492
2493 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2494                       struct btrfs_root *root,
2495                       u64 bytenr, u64 num_bytes, u64 parent,
2496                       u64 root_objectid, u64 ref_generation,
2497                       u64 owner_objectid, int pin)
2498 {
2499         int ret;
2500
2501         /*
2502          * tree log blocks never actually go into the extent allocation
2503          * tree, just update pinning info and exit early.
2504          *
2505          * data extents referenced by the tree log do need to have
2506          * their reference counts bumped.
2507          */
2508         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID &&
2509             owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2510                 mutex_lock(&root->fs_info->pinned_mutex);
2511
2512                 /* unlocks the pinned mutex */
2513                 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2514                 update_reserved_extents(root, bytenr, num_bytes, 0);
2515                 ret = 0;
2516         } else {
2517                 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent,
2518                                        root_objectid, ref_generation,
2519                                        owner_objectid,
2520                                        BTRFS_DROP_DELAYED_REF, 1);
2521                 BUG_ON(ret);
2522                 ret = check_ref_cleanup(trans, root, bytenr);
2523                 BUG_ON(ret);
2524         }
2525         return ret;
2526 }
2527
2528 static u64 stripe_align(struct btrfs_root *root, u64 val)
2529 {
2530         u64 mask = ((u64)root->stripesize - 1);
2531         u64 ret = (val + mask) & ~mask;
2532         return ret;
2533 }
2534
2535 /*
2536  * walks the btree of allocated extents and find a hole of a given size.
2537  * The key ins is changed to record the hole:
2538  * ins->objectid == block start
2539  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2540  * ins->offset == number of blocks
2541  * Any available blocks before search_start are skipped.
2542  */
2543 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2544                                      struct btrfs_root *orig_root,
2545                                      u64 num_bytes, u64 empty_size,
2546                                      u64 search_start, u64 search_end,
2547                                      u64 hint_byte, struct btrfs_key *ins,
2548                                      u64 exclude_start, u64 exclude_nr,
2549                                      int data)
2550 {
2551         int ret = 0;
2552         struct btrfs_root *root = orig_root->fs_info->extent_root;
2553         u64 total_needed = num_bytes;
2554         u64 *last_ptr = NULL;
2555         u64 last_wanted = 0;
2556         struct btrfs_block_group_cache *block_group = NULL;
2557         int chunk_alloc_done = 0;
2558         int empty_cluster = 2 * 1024 * 1024;
2559         int allowed_chunk_alloc = 0;
2560         struct list_head *head = NULL, *cur = NULL;
2561         int loop = 0;
2562         int extra_loop = 0;
2563         struct btrfs_space_info *space_info;
2564
2565         WARN_ON(num_bytes < root->sectorsize);
2566         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2567         ins->objectid = 0;
2568         ins->offset = 0;
2569
2570         if (orig_root->ref_cows || empty_size)
2571                 allowed_chunk_alloc = 1;
2572
2573         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2574                 last_ptr = &root->fs_info->last_alloc;
2575                 if (!btrfs_test_opt(root, SSD))
2576                         empty_cluster = 64 * 1024;
2577         }
2578
2579         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2580                 last_ptr = &root->fs_info->last_data_alloc;
2581
2582         if (last_ptr) {
2583                 if (*last_ptr) {
2584                         hint_byte = *last_ptr;
2585                         last_wanted = *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 (last_wanted && search_start != last_wanted) {
2595                 last_wanted = 0;
2596                 empty_size += empty_cluster;
2597         }
2598
2599         total_needed += empty_size;
2600         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2601         if (!block_group)
2602                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2603                                                              search_start);
2604         space_info = __find_space_info(root->fs_info, data);
2605
2606         down_read(&space_info->groups_sem);
2607         while (1) {
2608                 struct btrfs_free_space *free_space;
2609                 /*
2610                  * the only way this happens if our hint points to a block
2611                  * group thats not of the proper type, while looping this
2612                  * should never happen
2613                  */
2614                 if (empty_size)
2615                         extra_loop = 1;
2616
2617                 if (!block_group)
2618                         goto new_group_no_lock;
2619
2620                 if (unlikely(!block_group->cached)) {
2621                         mutex_lock(&block_group->cache_mutex);
2622                         ret = cache_block_group(root, block_group);
2623                         mutex_unlock(&block_group->cache_mutex);
2624                         if (ret)
2625                                 break;
2626                 }
2627
2628                 mutex_lock(&block_group->alloc_mutex);
2629                 if (unlikely(!block_group_bits(block_group, data)))
2630                         goto new_group;
2631
2632                 if (unlikely(block_group->ro))
2633                         goto new_group;
2634
2635                 free_space = btrfs_find_free_space(block_group, search_start,
2636                                                    total_needed);
2637                 if (free_space) {
2638                         u64 start = block_group->key.objectid;
2639                         u64 end = block_group->key.objectid +
2640                                 block_group->key.offset;
2641
2642                         search_start = stripe_align(root, free_space->offset);
2643
2644                         /* move on to the next group */
2645                         if (search_start + num_bytes >= search_end)
2646                                 goto new_group;
2647
2648                         /* move on to the next group */
2649                         if (search_start + num_bytes > end)
2650                                 goto new_group;
2651
2652                         if (last_wanted && search_start != last_wanted) {
2653                                 total_needed += empty_cluster;
2654                                 empty_size += empty_cluster;
2655                                 last_wanted = 0;
2656                                 /*
2657                                  * if search_start is still in this block group
2658                                  * then we just re-search this block group
2659                                  */
2660                                 if (search_start >= start &&
2661                                     search_start < end) {
2662                                         mutex_unlock(&block_group->alloc_mutex);
2663                                         continue;
2664                                 }
2665
2666                                 /* else we go to the next block group */
2667                                 goto new_group;
2668                         }
2669
2670                         if (exclude_nr > 0 &&
2671                             (search_start + num_bytes > exclude_start &&
2672                              search_start < exclude_start + exclude_nr)) {
2673                                 search_start = exclude_start + exclude_nr;
2674                                 /*
2675                                  * if search_start is still in this block group
2676                                  * then we just re-search this block group
2677                                  */
2678                                 if (search_start >= start &&
2679                                     search_start < end) {
2680                                         mutex_unlock(&block_group->alloc_mutex);
2681                                         last_wanted = 0;
2682                                         continue;
2683                                 }
2684
2685                                 /* else we go to the next block group */
2686                                 goto new_group;
2687                         }
2688
2689                         ins->objectid = search_start;
2690                         ins->offset = num_bytes;
2691
2692                         btrfs_remove_free_space_lock(block_group, search_start,
2693                                                      num_bytes);
2694                         /* we are all good, lets return */
2695                         mutex_unlock(&block_group->alloc_mutex);
2696                         break;
2697                 }
2698 new_group:
2699                 mutex_unlock(&block_group->alloc_mutex);
2700                 put_block_group(block_group);
2701                 block_group = NULL;
2702 new_group_no_lock:
2703                 /* don't try to compare new allocations against the
2704                  * last allocation any more
2705                  */
2706                 last_wanted = 0;
2707
2708                 /*
2709                  * Here's how this works.
2710                  * loop == 0: we were searching a block group via a hint
2711                  *              and didn't find anything, so we start at
2712                  *              the head of the block groups and keep searching
2713                  * loop == 1: we're searching through all of the block groups
2714                  *              if we hit the head again we have searched
2715                  *              all of the block groups for this space and we
2716                  *              need to try and allocate, if we cant error out.
2717                  * loop == 2: we allocated more space and are looping through
2718                  *              all of the block groups again.
2719                  */
2720                 if (loop == 0) {
2721                         head = &space_info->block_groups;
2722                         cur = head->next;
2723                         loop++;
2724                 } else if (loop == 1 && cur == head) {
2725                         int keep_going;
2726
2727                         /* at this point we give up on the empty_size
2728                          * allocations and just try to allocate the min
2729                          * space.
2730                          *
2731                          * The extra_loop field was set if an empty_size
2732                          * allocation was attempted above, and if this
2733                          * is try we need to try the loop again without
2734                          * the additional empty_size.
2735                          */
2736                         total_needed -= empty_size;
2737                         empty_size = 0;
2738                         keep_going = extra_loop;
2739                         loop++;
2740
2741                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2742                                 up_read(&space_info->groups_sem);
2743                                 ret = do_chunk_alloc(trans, root, num_bytes +
2744                                                      2 * 1024 * 1024, data, 1);
2745                                 down_read(&space_info->groups_sem);
2746                                 if (ret < 0)
2747                                         goto loop_check;
2748                                 head = &space_info->block_groups;
2749                                 /*
2750                                  * we've allocated a new chunk, keep
2751                                  * trying
2752                                  */
2753                                 keep_going = 1;
2754                                 chunk_alloc_done = 1;
2755                         } else if (!allowed_chunk_alloc) {
2756                                 space_info->force_alloc = 1;
2757                         }
2758 loop_check:
2759                         if (keep_going) {
2760                                 cur = head->next;
2761                                 extra_loop = 0;
2762                         } else {
2763                                 break;
2764                         }
2765                 } else if (cur == head) {
2766                         break;
2767                 }
2768
2769                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2770                                          list);
2771                 atomic_inc(&block_group->count);
2772
2773                 search_start = block_group->key.objectid;
2774                 cur = cur->next;
2775         }
2776
2777         /* we found what we needed */
2778         if (ins->objectid) {
2779                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2780                         trans->block_group = block_group->key.objectid;
2781
2782                 if (last_ptr)
2783                         *last_ptr = ins->objectid + ins->offset;
2784                 ret = 0;
2785         } else if (!ret) {
2786                 printk(KERN_ERR "btrfs searching for %llu bytes, "
2787                        "num_bytes %llu, loop %d, allowed_alloc %d\n",
2788                        (unsigned long long)total_needed,
2789                        (unsigned long long)num_bytes,
2790                        loop, allowed_chunk_alloc);
2791                 ret = -ENOSPC;
2792         }
2793         if (block_group)
2794                 put_block_group(block_group);
2795
2796         up_read(&space_info->groups_sem);
2797         return ret;
2798 }
2799
2800 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2801 {
2802         struct btrfs_block_group_cache *cache;
2803
2804         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
2805                (unsigned long long)(info->total_bytes - info->bytes_used -
2806                                     info->bytes_pinned - info->bytes_reserved),
2807                (info->full) ? "" : "not ");
2808         printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu,"
2809                " may_use=%llu, used=%llu\n", info->total_bytes,
2810                info->bytes_pinned, info->bytes_delalloc, info->bytes_may_use,
2811                info->bytes_used);
2812
2813         down_read(&info->groups_sem);
2814         list_for_each_entry(cache, &info->block_groups, list) {
2815                 spin_lock(&cache->lock);
2816                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
2817                        "%llu pinned %llu reserved\n",
2818                        (unsigned long long)cache->key.objectid,
2819                        (unsigned long long)cache->key.offset,
2820                        (unsigned long long)btrfs_block_group_used(&cache->item),
2821                        (unsigned long long)cache->pinned,
2822                        (unsigned long long)cache->reserved);
2823                 btrfs_dump_free_space(cache, bytes);
2824                 spin_unlock(&cache->lock);
2825         }
2826         up_read(&info->groups_sem);
2827 }
2828
2829 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2830                                   struct btrfs_root *root,
2831                                   u64 num_bytes, u64 min_alloc_size,
2832                                   u64 empty_size, u64 hint_byte,
2833                                   u64 search_end, struct btrfs_key *ins,
2834                                   u64 data)
2835 {
2836         int ret;
2837         u64 search_start = 0;
2838         struct btrfs_fs_info *info = root->fs_info;
2839
2840         data = btrfs_get_alloc_profile(root, data);
2841 again:
2842         /*
2843          * the only place that sets empty_size is btrfs_realloc_node, which
2844          * is not called recursively on allocations
2845          */
2846         if (empty_size || root->ref_cows) {
2847                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2848                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2849                                      2 * 1024 * 1024,
2850                                      BTRFS_BLOCK_GROUP_METADATA |
2851                                      (info->metadata_alloc_profile &
2852                                       info->avail_metadata_alloc_bits), 0);
2853                 }
2854                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2855                                      num_bytes + 2 * 1024 * 1024, data, 0);
2856         }
2857
2858         WARN_ON(num_bytes < root->sectorsize);
2859         ret = find_free_extent(trans, root, num_bytes, empty_size,
2860                                search_start, search_end, hint_byte, ins,
2861                                trans->alloc_exclude_start,
2862                                trans->alloc_exclude_nr, data);
2863
2864         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2865                 num_bytes = num_bytes >> 1;
2866                 num_bytes = num_bytes & ~(root->sectorsize - 1);
2867                 num_bytes = max(num_bytes, min_alloc_size);
2868                 do_chunk_alloc(trans, root->fs_info->extent_root,
2869                                num_bytes, data, 1);
2870                 goto again;
2871         }
2872         if (ret) {
2873                 struct btrfs_space_info *sinfo;
2874
2875                 sinfo = __find_space_info(root->fs_info, data);
2876                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
2877                        "wanted %llu\n", (unsigned long long)data,
2878                        (unsigned long long)num_bytes);
2879                 dump_space_info(sinfo, num_bytes);
2880                 BUG();
2881         }
2882
2883         return ret;
2884 }
2885
2886 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2887 {
2888         struct btrfs_block_group_cache *cache;
2889         int ret = 0;
2890
2891         cache = btrfs_lookup_block_group(root->fs_info, start);
2892         if (!cache) {
2893                 printk(KERN_ERR "Unable to find block group for %llu\n",
2894                        (unsigned long long)start);
2895                 return -ENOSPC;
2896         }
2897
2898         ret = btrfs_discard_extent(root, start, len);
2899
2900         btrfs_add_free_space(cache, start, len);
2901         put_block_group(cache);
2902         update_reserved_extents(root, start, len, 0);
2903
2904         return ret;
2905 }
2906
2907 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2908                                   struct btrfs_root *root,
2909                                   u64 num_bytes, u64 min_alloc_size,
2910                                   u64 empty_size, u64 hint_byte,
2911                                   u64 search_end, struct btrfs_key *ins,
2912                                   u64 data)
2913 {
2914         int ret;
2915         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2916                                      empty_size, hint_byte, search_end, ins,
2917                                      data);
2918         update_reserved_extents(root, ins->objectid, ins->offset, 1);
2919         return ret;
2920 }
2921
2922 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2923                                          struct btrfs_root *root, u64 parent,
2924                                          u64 root_objectid, u64 ref_generation,
2925                                          u64 owner, struct btrfs_key *ins,
2926                                          int ref_mod)
2927 {
2928         int ret;
2929         u64 super_used;
2930         u64 root_used;
2931         u64 num_bytes = ins->offset;
2932         u32 sizes[2];
2933         struct btrfs_fs_info *info = root->fs_info;
2934         struct btrfs_root *extent_root = info->extent_root;
2935         struct btrfs_extent_item *extent_item;
2936         struct btrfs_extent_ref *ref;
2937         struct btrfs_path *path;
2938         struct btrfs_key keys[2];
2939
2940         if (parent == 0)
2941                 parent = ins->objectid;
2942
2943         /* block accounting for super block */
2944         spin_lock(&info->delalloc_lock);
2945         super_used = btrfs_super_bytes_used(&info->super_copy);
2946         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2947
2948         /* block accounting for root item */
2949         root_used = btrfs_root_used(&root->root_item);
2950         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2951         spin_unlock(&info->delalloc_lock);
2952
2953         memcpy(&keys[0], ins, sizeof(*ins));
2954         keys[1].objectid = ins->objectid;
2955         keys[1].type = BTRFS_EXTENT_REF_KEY;
2956         keys[1].offset = parent;
2957         sizes[0] = sizeof(*extent_item);
2958         sizes[1] = sizeof(*ref);
2959
2960         path = btrfs_alloc_path();
2961         BUG_ON(!path);
2962
2963         path->leave_spinning = 1;
2964         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2965                                        sizes, 2);
2966         BUG_ON(ret);
2967
2968         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2969                                      struct btrfs_extent_item);
2970         btrfs_set_extent_refs(path->nodes[0], extent_item, ref_mod);
2971         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2972                              struct btrfs_extent_ref);
2973
2974         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2975         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2976         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2977         btrfs_set_ref_num_refs(path->nodes[0], ref, ref_mod);
2978
2979         btrfs_mark_buffer_dirty(path->nodes[0]);
2980
2981         trans->alloc_exclude_start = 0;
2982         trans->alloc_exclude_nr = 0;
2983         btrfs_free_path(path);
2984
2985         if (ret)
2986                 goto out;
2987
2988         ret = update_block_group(trans, root, ins->objectid,
2989                                  ins->offset, 1, 0);
2990         if (ret) {
2991                 printk(KERN_ERR "btrfs update block group failed for %llu "
2992                        "%llu\n", (unsigned long long)ins->objectid,
2993                        (unsigned long long)ins->offset);
2994                 BUG();
2995         }
2996 out:
2997         return ret;
2998 }
2999
3000 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3001                                 struct btrfs_root *root, u64 parent,
3002                                 u64 root_objectid, u64 ref_generation,
3003                                 u64 owner, struct btrfs_key *ins)
3004 {
3005         int ret;
3006
3007         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3008                 return 0;
3009
3010         ret = btrfs_add_delayed_ref(trans, ins->objectid,
3011                                     ins->offset, parent, root_objectid,
3012                                     ref_generation, owner,
3013                                     BTRFS_ADD_DELAYED_EXTENT, 0);
3014         BUG_ON(ret);
3015         return ret;
3016 }
3017
3018 /*
3019  * this is used by the tree logging recovery code.  It records that
3020  * an extent has been allocated and makes sure to clear the free
3021  * space cache bits as well
3022  */
3023 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3024                                 struct btrfs_root *root, u64 parent,
3025                                 u64 root_objectid, u64 ref_generation,
3026                                 u64 owner, struct btrfs_key *ins)
3027 {
3028         int ret;
3029         struct btrfs_block_group_cache *block_group;
3030
3031         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3032         mutex_lock(&block_group->cache_mutex);
3033         cache_block_group(root, block_group);
3034         mutex_unlock(&block_group->cache_mutex);
3035
3036         ret = btrfs_remove_free_space(block_group, ins->objectid,
3037                                       ins->offset);
3038         BUG_ON(ret);
3039         put_block_group(block_group);
3040         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3041                                             ref_generation, owner, ins, 1);
3042         return ret;
3043 }
3044
3045 /*
3046  * finds a free extent and does all the dirty work required for allocation
3047  * returns the key for the extent through ins, and a tree buffer for
3048  * the first block of the extent through buf.
3049  *
3050  * returns 0 if everything worked, non-zero otherwise.
3051  */
3052 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3053                        struct btrfs_root *root,
3054                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3055                        u64 root_objectid, u64 ref_generation,
3056                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3057                        u64 search_end, struct btrfs_key *ins, u64 data)
3058 {
3059         int ret;
3060         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3061                                      min_alloc_size, empty_size, hint_byte,
3062                                      search_end, ins, data);
3063         BUG_ON(ret);
3064         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3065                 ret = btrfs_add_delayed_ref(trans, ins->objectid,
3066                                             ins->offset, parent, root_objectid,
3067                                             ref_generation, owner_objectid,
3068                                             BTRFS_ADD_DELAYED_EXTENT, 0);
3069                 BUG_ON(ret);
3070         }
3071         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3072         return ret;
3073 }
3074
3075 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3076                                             struct btrfs_root *root,
3077                                             u64 bytenr, u32 blocksize,
3078                                             int level)
3079 {
3080         struct extent_buffer *buf;
3081
3082         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3083         if (!buf)
3084                 return ERR_PTR(-ENOMEM);
3085         btrfs_set_header_generation(buf, trans->transid);
3086         btrfs_set_buffer_lockdep_class(buf, level);
3087         btrfs_tree_lock(buf);
3088         clean_tree_block(trans, root, buf);
3089
3090         btrfs_set_lock_blocking(buf);
3091         btrfs_set_buffer_uptodate(buf);
3092
3093         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3094                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3095                          buf->start + buf->len - 1, GFP_NOFS);
3096         } else {
3097                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3098                          buf->start + buf->len - 1, GFP_NOFS);
3099         }
3100         trans->blocks_used++;
3101         /* this returns a buffer locked for blocking */
3102         return buf;
3103 }
3104
3105 /*
3106  * helper function to allocate a block for a given tree
3107  * returns the tree buffer or NULL.
3108  */
3109 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3110                                              struct btrfs_root *root,
3111                                              u32 blocksize, u64 parent,
3112                                              u64 root_objectid,
3113                                              u64 ref_generation,
3114                                              int level,
3115                                              u64 hint,
3116                                              u64 empty_size)
3117 {
3118         struct btrfs_key ins;
3119         int ret;
3120         struct extent_buffer *buf;
3121
3122         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3123                                  root_objectid, ref_generation, level,
3124                                  empty_size, hint, (u64)-1, &ins, 0);
3125         if (ret) {
3126                 BUG_ON(ret > 0);
3127                 return ERR_PTR(ret);
3128         }
3129
3130         buf = btrfs_init_new_buffer(trans, root, ins.objectid,
3131                                     blocksize, level);
3132         return buf;
3133 }
3134
3135 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3136                         struct btrfs_root *root, struct extent_buffer *leaf)
3137 {
3138         u64 leaf_owner;
3139         u64 leaf_generation;
3140         struct refsort *sorted;
3141         struct btrfs_key key;
3142         struct btrfs_file_extent_item *fi;
3143         int i;
3144         int nritems;
3145         int ret;
3146         int refi = 0;
3147         int slot;
3148
3149         BUG_ON(!btrfs_is_leaf(leaf));
3150         nritems = btrfs_header_nritems(leaf);
3151         leaf_owner = btrfs_header_owner(leaf);
3152         leaf_generation = btrfs_header_generation(leaf);
3153
3154         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3155         /* we do this loop twice.  The first time we build a list
3156          * of the extents we have a reference on, then we sort the list
3157          * by bytenr.  The second time around we actually do the
3158          * extent freeing.
3159          */
3160         for (i = 0; i < nritems; i++) {
3161                 u64 disk_bytenr;
3162                 cond_resched();
3163
3164                 btrfs_item_key_to_cpu(leaf, &key, i);
3165
3166                 /* only extents have references, skip everything else */
3167                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3168                         continue;
3169
3170                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3171
3172                 /* inline extents live in the btree, they don't have refs */
3173                 if (btrfs_file_extent_type(leaf, fi) ==
3174                     BTRFS_FILE_EXTENT_INLINE)
3175                         continue;
3176
3177                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3178
3179                 /* holes don't have refs */
3180                 if (disk_bytenr == 0)
3181                         continue;
3182
3183                 sorted[refi].bytenr = disk_bytenr;
3184                 sorted[refi].slot = i;
3185                 refi++;
3186         }
3187
3188         if (refi == 0)
3189                 goto out;
3190
3191         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3192
3193         for (i = 0; i < refi; i++) {
3194                 u64 disk_bytenr;
3195
3196                 disk_bytenr = sorted[i].bytenr;
3197                 slot = sorted[i].slot;
3198
3199                 cond_resched();
3200
3201                 btrfs_item_key_to_cpu(leaf, &key, slot);
3202                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3203                         continue;
3204
3205                 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3206
3207                 ret = btrfs_free_extent(trans, root, disk_bytenr,
3208                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3209                                 leaf->start, leaf_owner, leaf_generation,
3210                                 key.objectid, 0);
3211                 BUG_ON(ret);
3212
3213                 atomic_inc(&root->fs_info->throttle_gen);
3214                 wake_up(&root->fs_info->transaction_throttle);
3215                 cond_resched();
3216         }
3217 out:
3218         kfree(sorted);
3219         return 0;
3220 }
3221
3222 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3223                                         struct btrfs_root *root,
3224                                         struct btrfs_leaf_ref *ref)
3225 {
3226         int i;
3227         int ret;
3228         struct btrfs_extent_info *info;
3229         struct refsort *sorted;
3230
3231         if (ref->nritems == 0)
3232                 return 0;
3233
3234         sorted = kmalloc(sizeof(*sorted) * ref->nritems, GFP_NOFS);
3235         for (i = 0; i < ref->nritems; i++) {
3236                 sorted[i].bytenr = ref->extents[i].bytenr;
3237                 sorted[i].slot = i;
3238         }
3239         sort(sorted, ref->nritems, sizeof(struct refsort), refsort_cmp, NULL);
3240
3241         /*
3242          * the items in the ref were sorted when the ref was inserted
3243          * into the ref cache, so this is already in order
3244          */
3245         for (i = 0; i < ref->nritems; i++) {
3246                 info = ref->extents + sorted[i].slot;
3247                 ret = btrfs_free_extent(trans, root, info->bytenr,
3248                                           info->num_bytes, ref->bytenr,
3249                                           ref->owner, ref->generation,
3250                                           info->objectid, 0);
3251
3252                 atomic_inc(&root->fs_info->throttle_gen);
3253                 wake_up(&root->fs_info->transaction_throttle);
3254                 cond_resched();
3255
3256                 BUG_ON(ret);
3257                 info++;
3258         }
3259
3260         kfree(sorted);
3261         return 0;
3262 }
3263
3264 static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
3265                                      struct btrfs_root *root, u64 start,
3266                                      u64 len, u32 *refs)
3267 {
3268         int ret;
3269
3270         ret = btrfs_lookup_extent_ref(trans, root, start, len, refs);
3271         BUG_ON(ret);
3272
3273 #if 0 /* some debugging code in case we see problems here */
3274         /* if the refs count is one, it won't get increased again.  But
3275          * if the ref count is > 1, someone may be decreasing it at
3276          * the same time we are.
3277          */
3278         if (*refs != 1) {
3279                 struct extent_buffer *eb = NULL;
3280                 eb = btrfs_find_create_tree_block(root, start, len);
3281                 if (eb)
3282                         btrfs_tree_lock(eb);
3283
3284                 mutex_lock(&root->fs_info->alloc_mutex);
3285                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3286                 BUG_ON(ret);
3287                 mutex_unlock(&root->fs_info->alloc_mutex);
3288
3289                 if (eb) {
3290                         btrfs_tree_unlock(eb);
3291                         free_extent_buffer(eb);
3292                 }
3293                 if (*refs == 1) {
3294                         printk(KERN_ERR "btrfs block %llu went down to one "
3295                                "during drop_snap\n", (unsigned long long)start);
3296                 }
3297
3298         }
3299 #endif
3300
3301         cond_resched();
3302         return ret;
3303 }
3304
3305 /*
3306  * this is used while deleting old snapshots, and it drops the refs
3307  * on a whole subtree starting from a level 1 node.
3308  *
3309  * The idea is to sort all the leaf pointers, and then drop the
3310  * ref on all the leaves in order.  Most of the time the leaves
3311  * will have ref cache entries, so no leaf IOs will be required to
3312  * find the extents they have references on.
3313  *
3314  * For each leaf, any references it has are also dropped in order
3315  *
3316  * This ends up dropping the references in something close to optimal
3317  * order for reading and modifying the extent allocation tree.
3318  */
3319 static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3320                                         struct btrfs_root *root,
3321                                         struct btrfs_path *path)
3322 {
3323         u64 bytenr;
3324         u64 root_owner;
3325         u64 root_gen;
3326         struct extent_buffer *eb = path->nodes[1];
3327         struct extent_buffer *leaf;
3328         struct btrfs_leaf_ref *ref;
3329         struct refsort *sorted = NULL;
3330         int nritems = btrfs_header_nritems(eb);
3331         int ret;
3332         int i;
3333         int refi = 0;
3334         int slot = path->slots[1];
3335         u32 blocksize = btrfs_level_size(root, 0);
3336         u32 refs;
3337
3338         if (nritems == 0)
3339                 goto out;
3340
3341         root_owner = btrfs_header_owner(eb);
3342         root_gen = btrfs_header_generation(eb);
3343         sorted = kmalloc(sizeof(*sorted) * nritems, GFP_NOFS);
3344
3345         /*
3346          * step one, sort all the leaf pointers so we don't scribble
3347          * randomly into the extent allocation tree
3348          */
3349         for (i = slot; i < nritems; i++) {
3350                 sorted[refi].bytenr = btrfs_node_blockptr(eb, i);
3351                 sorted[refi].slot = i;
3352                 refi++;
3353         }
3354
3355         /*
3356          * nritems won't be zero, but if we're picking up drop_snapshot
3357          * after a crash, slot might be > 0, so double check things
3358          * just in case.
3359          */
3360         if (refi == 0)
3361                 goto out;
3362
3363         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
3364
3365         /*
3366          * the first loop frees everything the leaves point to
3367          */
3368         for (i = 0; i < refi; i++) {
3369                 u64 ptr_gen;
3370
3371                 bytenr = sorted[i].bytenr;
3372
3373                 /*
3374                  * check the reference count on this leaf.  If it is > 1
3375                  * we just decrement it below and don't update any
3376                  * of the refs the leaf points to.
3377                  */
3378                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3379                                                 blocksize, &refs);
3380                 BUG_ON(ret);
3381                 if (refs != 1)
3382                         continue;
3383
3384                 ptr_gen = btrfs_node_ptr_generation(eb, sorted[i].slot);
3385
3386                 /*
3387                  * the leaf only had one reference, which means the
3388                  * only thing pointing to this leaf is the snapshot
3389                  * we're deleting.  It isn't possible for the reference
3390                  * count to increase again later
3391                  *
3392                  * The reference cache is checked for the leaf,
3393                  * and if found we'll be able to drop any refs held by
3394                  * the leaf without needing to read it in.
3395                  */
3396                 ref = btrfs_lookup_leaf_ref(root, bytenr);
3397                 if (ref && ref->generation != ptr_gen) {
3398                         btrfs_free_leaf_ref(root, ref);
3399                         ref = NULL;
3400                 }
3401                 if (ref) {
3402                         ret = cache_drop_leaf_ref(trans, root, ref);
3403                         BUG_ON(ret);
3404                         btrfs_remove_leaf_ref(root, ref);
3405                         btrfs_free_leaf_ref(root, ref);
3406                 } else {
3407                         /*
3408                          * the leaf wasn't in the reference cache, so
3409                          * we have to read it.
3410                          */
3411                         leaf = read_tree_block(root, bytenr, blocksize,
3412                                                ptr_gen);
3413                         ret = btrfs_drop_leaf_ref(trans, root, leaf);
3414                         BUG_ON(ret);
3415                         free_extent_buffer(leaf);
3416                 }
3417                 atomic_inc(&root->fs_info->throttle_gen);
3418                 wake_up(&root->fs_info->transaction_throttle);
3419                 cond_resched();
3420         }
3421
3422         /*
3423          * run through the loop again to free the refs on the leaves.
3424          * This is faster than doing it in the loop above because
3425          * the leaves are likely to be clustered together.  We end up
3426          * working in nice chunks on the extent allocation tree.
3427          */
3428         for (i = 0; i < refi; i++) {
3429                 bytenr = sorted[i].bytenr;
3430                 ret = btrfs_free_extent(trans, root, bytenr,
3431                                         blocksize, eb->start,
3432                                         root_owner, root_gen, 0, 1);
3433                 BUG_ON(ret);
3434
3435                 atomic_inc(&root->fs_info->throttle_gen);
3436                 wake_up(&root->fs_info->transaction_throttle);
3437                 cond_resched();
3438         }
3439 out:
3440         kfree(sorted);
3441
3442         /*
3443          * update the path to show we've processed the entire level 1
3444          * node.  This will get saved into the root's drop_snapshot_progress
3445          * field so these drops are not repeated again if this transaction
3446          * commits.
3447          */
3448         path->slots[1] = nritems;
3449         return 0;
3450 }
3451
3452 /*
3453  * helper function for drop_snapshot, this walks down the tree dropping ref
3454  * counts as it goes.
3455  */
3456 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3457                                    struct btrfs_root *root,
3458                                    struct btrfs_path *path, int *level)
3459 {
3460         u64 root_owner;
3461         u64 root_gen;
3462         u64 bytenr;
3463         u64 ptr_gen;
3464         struct extent_buffer *next;
3465         struct extent_buffer *cur;
3466         struct extent_buffer *parent;
3467         u32 blocksize;
3468         int ret;
3469         u32 refs;
3470
3471         WARN_ON(*level < 0);
3472         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3473         ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
3474                                 path->nodes[*level]->len, &refs);
3475         BUG_ON(ret);
3476         if (refs > 1)
3477                 goto out;
3478
3479         /*
3480          * walk down to the last node level and free all the leaves
3481          */
3482         while (*level >= 0) {
3483                 WARN_ON(*level < 0);
3484                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3485                 cur = path->nodes[*level];
3486
3487                 if (btrfs_header_level(cur) != *level)
3488                         WARN_ON(1);
3489
3490                 if (path->slots[*level] >=
3491                     btrfs_header_nritems(cur))
3492                         break;
3493
3494                 /* the new code goes down to level 1 and does all the
3495                  * leaves pointed to that node in bulk.  So, this check
3496                  * for level 0 will always be false.
3497                  *
3498                  * But, the disk format allows the drop_snapshot_progress
3499                  * field in the root to leave things in a state where
3500                  * a leaf will need cleaning up here.  If someone crashes
3501                  * with the old code and then boots with the new code,
3502                  * we might find a leaf here.
3503                  */
3504                 if (*level == 0) {
3505                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3506                         BUG_ON(ret);
3507                         break;
3508                 }
3509
3510                 /*
3511                  * once we get to level one, process the whole node
3512                  * at once, including everything below it.
3513                  */
3514                 if (*level == 1) {
3515                         ret = drop_level_one_refs(trans, root, path);
3516                         BUG_ON(ret);
3517                         break;
3518                 }
3519
3520                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3521                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3522                 blocksize = btrfs_level_size(root, *level - 1);
3523
3524                 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3525                                                 blocksize, &refs);
3526                 BUG_ON(ret);
3527
3528                 /*
3529                  * if there is more than one reference, we don't need
3530                  * to read that node to drop any references it has.  We
3531                  * just drop the ref we hold on that node and move on to the
3532                  * next slot in this level.
3533                  */
3534                 if (refs != 1) {
3535                         parent = path->nodes[*level];
3536                         root_owner = btrfs_header_owner(parent);
3537                         root_gen = btrfs_header_generation(parent);
3538                         path->slots[*level]++;
3539
3540                         ret = btrfs_free_extent(trans, root, bytenr,
3541                                                 blocksize, parent->start,
3542                                                 root_owner, root_gen,
3543                                                 *level - 1, 1);
3544                         BUG_ON(ret);
3545
3546                         atomic_inc(&root->fs_info->throttle_gen);
3547                         wake_up(&root->fs_info->transaction_throttle);
3548                         cond_resched();
3549
3550                         continue;
3551                 }
3552
3553                 /*
3554                  * we need to keep freeing things in the next level down.
3555                  * read the block and loop around to process it
3556                  */
3557                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3558                 WARN_ON(*level <= 0);
3559                 if (path->nodes[*level-1])
3560                         free_extent_buffer(path->nodes[*level-1]);
3561                 path->nodes[*level-1] = next;
3562                 *level = btrfs_header_level(next);
3563                 path->slots[*level] = 0;
3564                 cond_resched();
3565         }
3566 out:
3567         WARN_ON(*level < 0);
3568         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3569
3570         if (path->nodes[*level] == root->node) {
3571                 parent = path->nodes[*level];
3572                 bytenr = path->nodes[*level]->start;
3573         } else {
3574                 parent = path->nodes[*level + 1];
3575                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3576         }
3577
3578         blocksize = btrfs_level_size(root, *level);
3579         root_owner = btrfs_header_owner(parent);
3580         root_gen = btrfs_header_generation(parent);
3581
3582         /*
3583          * cleanup and free the reference on the last node
3584          * we processed
3585          */
3586         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3587                                   parent->start, root_owner, root_gen,
3588                                   *level, 1);
3589         free_extent_buffer(path->nodes[*level]);
3590         path->nodes[*level] = NULL;
3591
3592         *level += 1;
3593         BUG_ON(ret);
3594
3595         cond_resched();
3596         return 0;
3597 }
3598
3599 /*
3600  * helper function for drop_subtree, this function is similar to
3601  * walk_down_tree. The main difference is that it checks reference
3602  * counts while tree blocks are locked.
3603  */
3604 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3605                                       struct btrfs_root *root,
3606                                       struct btrfs_path *path, int *level)
3607 {
3608         struct extent_buffer *next;
3609         struct extent_buffer *cur;
3610         struct extent_buffer *parent;
3611         u64 bytenr;
3612         u64 ptr_gen;
3613         u32 blocksize;
3614         u32 refs;
3615         int ret;
3616
3617         cur = path->nodes[*level];
3618         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3619                                       &refs);
3620         BUG_ON(ret);
3621         if (refs > 1)
3622                 goto out;
3623
3624         while (*level >= 0) {
3625                 cur = path->nodes[*level];
3626                 if (*level == 0) {
3627                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3628                         BUG_ON(ret);
3629                         clean_tree_block(trans, root, cur);
3630                         break;
3631                 }
3632                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3633                         clean_tree_block(trans, root, cur);
3634                         break;
3635                 }
3636
3637                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3638                 blocksize = btrfs_level_size(root, *level - 1);
3639                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3640
3641                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3642                 btrfs_tree_lock(next);
3643                 btrfs_set_lock_blocking(next);
3644
3645                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3646                                               &refs);
3647                 BUG_ON(ret);
3648                 if (refs > 1) {
3649                         parent = path->nodes[*level];
3650                         ret = btrfs_free_extent(trans, root, bytenr,
3651                                         blocksize, parent->start,
3652                                         btrfs_header_owner(parent),
3653                                         btrfs_header_generation(parent),
3654                                         *level - 1, 1);
3655                         BUG_ON(ret);
3656                         path->slots[*level]++;
3657                         btrfs_tree_unlock(next);
3658                         free_extent_buffer(next);
3659                         continue;
3660                 }
3661
3662                 *level = btrfs_header_level(next);
3663                 path->nodes[*level] = next;
3664                 path->slots[*level] = 0;
3665                 path->locks[*level] = 1;
3666                 cond_resched();
3667         }
3668 out:
3669         parent = path->nodes[*level + 1];
3670         bytenr = path->nodes[*level]->start;
3671         blocksize = path->nodes[*level]->len;
3672
3673         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3674                         parent->start, btrfs_header_owner(parent),
3675                         btrfs_header_generation(parent), *level, 1);
3676         BUG_ON(ret);
3677
3678         if (path->locks[*level]) {
3679                 btrfs_tree_unlock(path->nodes[*level]);
3680                 path->locks[*level] = 0;
3681         }
3682         free_extent_buffer(path->nodes[*level]);
3683         path->nodes[*level] = NULL;
3684         *level += 1;
3685         cond_resched();
3686         return 0;
3687 }
3688
3689 /*
3690  * helper for dropping snapshots.  This walks back up the tree in the path
3691  * to find the first node higher up where we haven't yet gone through
3692  * all the slots
3693  */
3694 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3695                                  struct btrfs_root *root,
3696                                  struct btrfs_path *path,
3697                                  int *level, int max_level)
3698 {
3699         u64 root_owner;
3700         u64 root_gen;
3701         struct btrfs_root_item *root_item = &root->root_item;
3702         int i;
3703         int slot;
3704         int ret;
3705
3706         for (i = *level; i < max_level && path->nodes[i]; i++) {
3707                 slot = path->slots[i];
3708                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3709                         struct extent_buffer *node;
3710                         struct btrfs_disk_key disk_key;
3711
3712                         /*
3713                          * there is more work to do in this level.
3714                          * Update the drop_progress marker to reflect
3715                          * the work we've done so far, and then bump
3716                          * the slot number
3717                          */
3718                         node = path->nodes[i];
3719                         path->slots[i]++;
3720                         *level = i;
3721                         WARN_ON(*level == 0);
3722                         btrfs_node_key(node, &disk_key, path->slots[i]);
3723                         memcpy(&root_item->drop_progress,
3724                                &disk_key, sizeof(disk_key));
3725                         root_item->drop_level = i;
3726                         return 0;
3727                 } else {
3728                         struct extent_buffer *parent;
3729
3730                         /*
3731                          * this whole node is done, free our reference
3732                          * on it and go up one level
3733                          */
3734                         if (path->nodes[*level] == root->node)
3735                                 parent = path->nodes[*level];
3736                         else
3737                                 parent = path->nodes[*level + 1];
3738
3739                         root_owner = btrfs_header_owner(parent);
3740                         root_gen = btrfs_header_generation(parent);
3741
3742                         clean_tree_block(trans, root, path->nodes[*level]);
3743                         ret = btrfs_free_extent(trans, root,
3744                                                 path->nodes[*level]->start,
3745                                                 path->nodes[*level]->len,
3746                                                 parent->start, root_owner,
3747                                                 root_gen, *level, 1);
3748                         BUG_ON(ret);
3749                         if (path->locks[*level]) {
3750                                 btrfs_tree_unlock(path->nodes[*level]);
3751                                 path->locks[*level] = 0;
3752                         }
3753                         free_extent_buffer(path->nodes[*level]);
3754                         path->nodes[*level] = NULL;
3755                         *level = i + 1;
3756                 }
3757         }
3758         return 1;
3759 }
3760
3761 /*
3762  * drop the reference count on the tree rooted at 'snap'.  This traverses
3763  * the tree freeing any blocks that have a ref count of zero after being
3764  * decremented.
3765  */
3766 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3767                         *root)
3768 {
3769         int ret = 0;
3770         int wret;
3771         int level;
3772         struct btrfs_path *path;
3773         int i;
3774         int orig_level;
3775         int update_count;
3776         struct btrfs_root_item *root_item = &root->root_item;
3777
3778         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3779         path = btrfs_alloc_path();
3780         BUG_ON(!path);
3781
3782         level = btrfs_header_level(root->node);
3783         orig_level = level;
3784         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3785                 path->nodes[level] = root->node;
3786                 extent_buffer_get(root->node);
3787                 path->slots[level] = 0;
3788         } else {
3789                 struct btrfs_key key;
3790                 struct btrfs_disk_key found_key;
3791                 struct extent_buffer *node;
3792
3793                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3794                 level = root_item->drop_level;
3795                 path->lowest_level = level;
3796                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3797                 if (wret < 0) {
3798                         ret = wret;
3799                         goto out;
3800                 }
3801                 node = path->nodes[level];
3802                 btrfs_node_key(node, &found_key, path->slots[level]);
3803                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3804                                sizeof(found_key)));
3805                 /*
3806                  * unlock our path, this is safe because only this
3807                  * function is allowed to delete this snapshot
3808                  */
3809                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3810                         if (path->nodes[i] && path->locks[i]) {
3811                                 path->locks[i] = 0;
3812                                 btrfs_tree_unlock(path->nodes[i]);
3813                         }
3814                 }
3815         }
3816         while (1) {
3817                 unsigned long update;
3818                 wret = walk_down_tree(trans, root, path, &level);
3819                 if (wret > 0)
3820                         break;
3821                 if (wret < 0)
3822                         ret = wret;
3823
3824                 wret = walk_up_tree(trans, root, path, &level,
3825                                     BTRFS_MAX_LEVEL);
3826                 if (wret > 0)
3827                         break;
3828                 if (wret < 0)
3829                         ret = wret;
3830                 if (trans->transaction->in_commit ||
3831                     trans->transaction->delayed_refs.flushing) {
3832                         ret = -EAGAIN;
3833                         break;
3834                 }
3835                 atomic_inc(&root->fs_info->throttle_gen);
3836                 wake_up(&root->fs_info->transaction_throttle);
3837                 for (update_count = 0; update_count < 16; update_count++) {
3838                         update = trans->delayed_ref_updates;
3839                         trans->delayed_ref_updates = 0;
3840                         if (update)
3841                                 btrfs_run_delayed_refs(trans, root, update);
3842                         else
3843                                 break;
3844                 }
3845         }
3846         for (i = 0; i <= orig_level; i++) {
3847                 if (path->nodes[i]) {
3848                         free_extent_buffer(path->nodes[i]);
3849                         path->nodes[i] = NULL;
3850                 }
3851         }
3852 out:
3853         btrfs_free_path(path);
3854         return ret;
3855 }
3856
3857 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3858                         struct btrfs_root *root,
3859                         struct extent_buffer *node,
3860                         struct extent_buffer *parent)
3861 {
3862         struct btrfs_path *path;
3863         int level;
3864         int parent_level;
3865         int ret = 0;
3866         int wret;
3867
3868         path = btrfs_alloc_path();
3869         BUG_ON(!path);
3870
3871         btrfs_assert_tree_locked(parent);
3872         parent_level = btrfs_header_level(parent);
3873         extent_buffer_get(parent);
3874         path->nodes[parent_level] = parent;
3875         path->slots[parent_level] = btrfs_header_nritems(parent);
3876
3877         btrfs_assert_tree_locked(node);
3878         level = btrfs_header_level(node);
3879         extent_buffer_get(node);
3880         path->nodes[level] = node;
3881         path->slots[level] = 0;
3882
3883         while (1) {
3884                 wret = walk_down_subtree(trans, root, path, &level);
3885                 if (wret < 0)
3886                         ret = wret;
3887                 if (wret != 0)
3888                         break;
3889
3890                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3891                 if (wret < 0)
3892                         ret = wret;
3893                 if (wret != 0)
3894                         break;
3895         }
3896
3897         btrfs_free_path(path);
3898         return ret;
3899 }
3900
3901 static unsigned long calc_ra(unsigned long start, unsigned long last,
3902                              unsigned long nr)
3903 {
3904         return min(last, start + nr - 1);
3905 }
3906
3907 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3908                                          u64 len)
3909 {
3910         u64 page_start;
3911         u64 page_end;
3912         unsigned long first_index;
3913         unsigned long last_index;
3914         unsigned long i;
3915         struct page *page;
3916         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3917         struct file_ra_state *ra;
3918         struct btrfs_ordered_extent *ordered;
3919         unsigned int total_read = 0;
3920         unsigned int total_dirty = 0;
3921         int ret = 0;
3922
3923         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3924
3925         mutex_lock(&inode->i_mutex);
3926         first_index = start >> PAGE_CACHE_SHIFT;
3927         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3928
3929         /* make sure the dirty trick played by the caller work */
3930         ret = invalidate_inode_pages2_range(inode->i_mapping,
3931                                             first_index, last_index);
3932         if (ret)
3933                 goto out_unlock;
3934
3935         file_ra_state_init(ra, inode->i_mapping);
3936
3937         for (i = first_index ; i <= last_index; i++) {
3938                 if (total_read % ra->ra_pages == 0) {
3939                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3940                                        calc_ra(i, last_index, ra->ra_pages));
3941                 }
3942                 total_read++;
3943 again:
3944                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3945                         BUG_ON(1);
3946                 page = grab_cache_page(inode->i_mapping, i);
3947                 if (!page) {
3948                         ret = -ENOMEM;
3949                         goto out_unlock;
3950                 }
3951                 if (!PageUptodate(page)) {
3952                         btrfs_readpage(NULL, page);
3953                         lock_page(page);
3954                         if (!PageUptodate(page)) {
3955                                 unlock_page(page);
3956                                 page_cache_release(page);
3957                                 ret = -EIO;
3958                                 goto out_unlock;
3959                         }
3960                 }
3961                 wait_on_page_writeback(page);
3962
3963                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3964                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3965                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3966
3967                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3968                 if (ordered) {
3969                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3970                         unlock_page(page);
3971                         page_cache_release(page);
3972                         btrfs_start_ordered_extent(inode, ordered, 1);
3973                         btrfs_put_ordered_extent(ordered);
3974                         goto again;
3975                 }
3976                 set_page_extent_mapped(page);
3977
3978                 if (i == first_index)
3979                         set_extent_bits(io_tree, page_start, page_end,
3980                                         EXTENT_BOUNDARY, GFP_NOFS);
3981                 btrfs_set_extent_delalloc(inode, page_start, page_end);
3982
3983                 set_page_dirty(page);
3984                 total_dirty++;
3985
3986                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3987                 unlock_page(page);
3988                 page_cache_release(page);
3989         }
3990
3991 out_unlock:
3992         kfree(ra);
3993         mutex_unlock(&inode->i_mutex);
3994         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3995         return ret;
3996 }
3997
3998 static noinline int relocate_data_extent(struct inode *reloc_inode,
3999                                          struct btrfs_key *extent_key,
4000                                          u64 offset)
4001 {
4002         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4003         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4004         struct extent_map *em;
4005         u64 start = extent_key->objectid - offset;
4006         u64 end = start + extent_key->offset - 1;
4007
4008         em = alloc_extent_map(GFP_NOFS);
4009         BUG_ON(!em || IS_ERR(em));
4010
4011         em->start = start;
4012         em->len = extent_key->offset;
4013         em->block_len = extent_key->offset;
4014         em->block_start = extent_key->objectid;
4015         em->bdev = root->fs_info->fs_devices->latest_bdev;
4016         set_bit(EXTENT_FLAG_PINNED, &em->flags);
4017
4018         /* setup extent map to cheat btrfs_readpage */
4019         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4020         while (1) {
4021                 int ret;
4022                 spin_lock(&em_tree->lock);
4023                 ret = add_extent_mapping(em_tree, em);
4024                 spin_unlock(&em_tree->lock);
4025                 if (ret != -EEXIST) {
4026                         free_extent_map(em);
4027                         break;
4028                 }
4029                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4030         }
4031         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4032
4033         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4034 }
4035
4036 struct btrfs_ref_path {
4037         u64 extent_start;
4038         u64 nodes[BTRFS_MAX_LEVEL];
4039         u64 root_objectid;
4040         u64 root_generation;
4041         u64 owner_objectid;
4042         u32 num_refs;
4043         int lowest_level;
4044         int current_level;
4045         int shared_level;
4046
4047         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4048         u64 new_nodes[BTRFS_MAX_LEVEL];
4049 };
4050
4051 struct disk_extent {
4052         u64 ram_bytes;
4053         u64 disk_bytenr;
4054         u64 disk_num_bytes;
4055         u64 offset;
4056         u64 num_bytes;
4057         u8 compression;
4058         u8 encryption;
4059         u16 other_encoding;
4060 };
4061
4062 static int is_cowonly_root(u64 root_objectid)
4063 {
4064         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4065             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4066             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4067             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4068             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4069             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4070                 return 1;
4071         return 0;
4072 }
4073
4074 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4075                                     struct btrfs_root *extent_root,
4076                                     struct btrfs_ref_path *ref_path,
4077                                     int first_time)
4078 {
4079         struct extent_buffer *leaf;
4080         struct btrfs_path *path;
4081         struct btrfs_extent_ref *ref;
4082         struct btrfs_key key;
4083         struct btrfs_key found_key;
4084         u64 bytenr;
4085         u32 nritems;
4086         int level;
4087         int ret = 1;
4088
4089         path = btrfs_alloc_path();
4090         if (!path)
4091                 return -ENOMEM;
4092
4093         if (first_time) {
4094                 ref_path->lowest_level = -1;
4095                 ref_path->current_level = -1;
4096                 ref_path->shared_level = -1;
4097                 goto walk_up;
4098         }
4099 walk_down:
4100         level = ref_path->current_level - 1;
4101         while (level >= -1) {
4102                 u64 parent;
4103                 if (level < ref_path->lowest_level)
4104                         break;
4105
4106                 if (level >= 0)
4107                         bytenr = ref_path->nodes[level];
4108                 else
4109                         bytenr = ref_path->extent_start;
4110                 BUG_ON(bytenr == 0);
4111
4112                 parent = ref_path->nodes[level + 1];
4113                 ref_path->nodes[level + 1] = 0;
4114                 ref_path->current_level = level;
4115                 BUG_ON(parent == 0);
4116
4117                 key.objectid = bytenr;
4118                 key.offset = parent + 1;
4119                 key.type = BTRFS_EXTENT_REF_KEY;
4120
4121                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4122                 if (ret < 0)
4123                         goto out;
4124                 BUG_ON(ret == 0);
4125
4126                 leaf = path->nodes[0];
4127                 nritems = btrfs_header_nritems(leaf);
4128                 if (path->slots[0] >= nritems) {
4129                         ret = btrfs_next_leaf(extent_root, path);
4130                         if (ret < 0)
4131                                 goto out;
4132                         if (ret > 0)
4133                                 goto next;
4134                         leaf = path->nodes[0];
4135                 }
4136
4137                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4138                 if (found_key.objectid == bytenr &&
4139                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4140                         if (level < ref_path->shared_level)
4141                                 ref_path->shared_level = level;
4142                         goto found;
4143                 }
4144 next:
4145                 level--;
4146                 btrfs_release_path(extent_root, path);
4147                 cond_resched();
4148         }
4149         /* reached lowest level */
4150         ret = 1;
4151         goto out;
4152 walk_up:
4153         level = ref_path->current_level;
4154         while (level < BTRFS_MAX_LEVEL - 1) {
4155                 u64 ref_objectid;
4156
4157                 if (level >= 0)
4158                         bytenr = ref_path->nodes[level];
4159                 else
4160                         bytenr = ref_path->extent_start;
4161
4162                 BUG_ON(bytenr == 0);
4163
4164                 key.objectid = bytenr;
4165                 key.offset = 0;
4166                 key.type = BTRFS_EXTENT_REF_KEY;
4167
4168                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4169                 if (ret < 0)
4170                         goto out;
4171
4172                 leaf = path->nodes[0];
4173                 nritems = btrfs_header_nritems(leaf);
4174                 if (path->slots[0] >= nritems) {
4175                         ret = btrfs_next_leaf(extent_root, path);
4176                         if (ret < 0)
4177                                 goto out;
4178                         if (ret > 0) {
4179                                 /* the extent was freed by someone */
4180                                 if (ref_path->lowest_level == level)
4181                                         goto out;
4182                                 btrfs_release_path(extent_root, path);
4183                                 goto walk_down;
4184                         }
4185                         leaf = path->nodes[0];
4186                 }
4187
4188                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4189                 if (found_key.objectid != bytenr ||
4190                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4191                         /* the extent was freed by someone */
4192                         if (ref_path->lowest_level == level) {
4193                                 ret = 1;
4194                                 goto out;
4195                         }
4196                         btrfs_release_path(extent_root, path);
4197                         goto walk_down;
4198                 }
4199 found:
4200                 ref = btrfs_item_ptr(leaf, path->slots[0],
4201                                 struct btrfs_extent_ref);
4202                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4203                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4204                         if (first_time) {
4205                                 level = (int)ref_objectid;
4206                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4207                                 ref_path->lowest_level = level;
4208                                 ref_path->current_level = level;
4209                                 ref_path->nodes[level] = bytenr;
4210                         } else {
4211                                 WARN_ON(ref_objectid != level);
4212                         }
4213                 } else {
4214                         WARN_ON(level != -1);
4215                 }
4216                 first_time = 0;
4217
4218                 if (ref_path->lowest_level == level) {
4219                         ref_path->owner_objectid = ref_objectid;
4220                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4221                 }
4222
4223                 /*
4224                  * the block is tree root or the block isn't in reference
4225                  * counted tree.
4226                  */
4227                 if (found_key.objectid == found_key.offset ||
4228                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4229                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4230                         ref_path->root_generation =
4231                                 btrfs_ref_generation(leaf, ref);
4232                         if (level < 0) {
4233                                 /* special reference from the tree log */
4234                                 ref_path->nodes[0] = found_key.offset;
4235                                 ref_path->current_level = 0;
4236                         }
4237                         ret = 0;
4238                         goto out;
4239                 }
4240
4241                 level++;
4242                 BUG_ON(ref_path->nodes[level] != 0);
4243                 ref_path->nodes[level] = found_key.offset;
4244                 ref_path->current_level = level;
4245
4246                 /*
4247                  * the reference was created in the running transaction,
4248                  * no need to continue walking up.
4249                  */
4250                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4251                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4252                         ref_path->root_generation =
4253                                 btrfs_ref_generation(leaf, ref);
4254                         ret = 0;
4255                         goto out;
4256                 }
4257
4258                 btrfs_release_path(extent_root, path);
4259                 cond_resched();
4260         }
4261         /* reached max tree level, but no tree root found. */
4262         BUG();
4263 out:
4264         btrfs_free_path(path);
4265         return ret;
4266 }
4267
4268 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4269                                 struct btrfs_root *extent_root,
4270                                 struct btrfs_ref_path *ref_path,
4271                                 u64 extent_start)
4272 {
4273         memset(ref_path, 0, sizeof(*ref_path));
4274         ref_path->extent_start = extent_start;
4275
4276         return __next_ref_path(trans, extent_root, ref_path, 1);
4277 }
4278
4279 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4280                                struct btrfs_root *extent_root,
4281                                struct btrfs_ref_path *ref_path)
4282 {
4283         return __next_ref_path(trans, extent_root, ref_path, 0);
4284 }
4285
4286 static noinline int get_new_locations(struct inode *reloc_inode,
4287                                       struct btrfs_key *extent_key,
4288                                       u64 offset, int no_fragment,
4289                                       struct disk_extent **extents,
4290                                       int *nr_extents)
4291 {
4292         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4293         struct btrfs_path *path;
4294         struct btrfs_file_extent_item *fi;
4295         struct extent_buffer *leaf;
4296         struct disk_extent *exts = *extents;
4297         struct btrfs_key found_key;
4298         u64 cur_pos;
4299         u64 last_byte;
4300         u32 nritems;
4301         int nr = 0;
4302         int max = *nr_extents;
4303         int ret;
4304
4305         WARN_ON(!no_fragment && *extents);
4306         if (!exts) {
4307                 max = 1;
4308                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4309                 if (!exts)
4310                         return -ENOMEM;
4311         }
4312
4313         path = btrfs_alloc_path();
4314         BUG_ON(!path);
4315
4316         cur_pos = extent_key->objectid - offset;
4317         last_byte = extent_key->objectid + extent_key->offset;
4318         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4319                                        cur_pos, 0);
4320         if (ret < 0)
4321                 goto out;
4322         if (ret > 0) {
4323                 ret = -ENOENT;
4324                 goto out;
4325         }
4326
4327         while (1) {
4328                 leaf = path->nodes[0];
4329                 nritems = btrfs_header_nritems(leaf);
4330                 if (path->slots[0] >= nritems) {
4331                         ret = btrfs_next_leaf(root, path);
4332                         if (ret < 0)
4333                                 goto out;
4334                         if (ret > 0)
4335                                 break;
4336                         leaf = path->nodes[0];
4337                 }
4338
4339                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4340                 if (found_key.offset != cur_pos ||
4341                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4342                     found_key.objectid != reloc_inode->i_ino)
4343                         break;
4344
4345                 fi = btrfs_item_ptr(leaf, path->slots[0],
4346                                     struct btrfs_file_extent_item);
4347                 if (btrfs_file_extent_type(leaf, fi) !=
4348                     BTRFS_FILE_EXTENT_REG ||
4349                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4350                         break;
4351
4352                 if (nr == max) {
4353                         struct disk_extent *old = exts;
4354                         max *= 2;
4355                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4356                         memcpy(exts, old, sizeof(*exts) * nr);
4357                         if (old != *extents)
4358                                 kfree(old);
4359                 }
4360
4361                 exts[nr].disk_bytenr =
4362                         btrfs_file_extent_disk_bytenr(leaf, fi);
4363                 exts[nr].disk_num_bytes =
4364                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4365                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4366                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4367                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4368                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4369                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4370                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4371                                                                            fi);
4372                 BUG_ON(exts[nr].offset > 0);
4373                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4374                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4375
4376                 cur_pos += exts[nr].num_bytes;
4377                 nr++;
4378
4379                 if (cur_pos + offset >= last_byte)
4380                         break;
4381
4382                 if (no_fragment) {
4383                         ret = 1;
4384                         goto out;
4385                 }
4386                 path->slots[0]++;
4387         }
4388
4389         BUG_ON(cur_pos + offset > last_byte);
4390         if (cur_pos + offset < last_byte) {
4391                 ret = -ENOENT;
4392                 goto out;
4393         }
4394         ret = 0;
4395 out:
4396         btrfs_free_path(path);
4397         if (ret) {
4398                 if (exts != *extents)
4399                         kfree(exts);
4400         } else {
4401                 *extents = exts;
4402                 *nr_extents = nr;
4403         }
4404         return ret;
4405 }
4406
4407 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4408                                         struct btrfs_root *root,
4409                                         struct btrfs_path *path,
4410                                         struct btrfs_key *extent_key,
4411                                         struct btrfs_key *leaf_key,
4412                                         struct btrfs_ref_path *ref_path,
4413                                         struct disk_extent *new_extents,
4414                                         int nr_extents)
4415 {
4416         struct extent_buffer *leaf;
4417         struct btrfs_file_extent_item *fi;
4418         struct inode *inode = NULL;
4419         struct btrfs_key key;
4420         u64 lock_start = 0;
4421         u64 lock_end = 0;
4422         u64 num_bytes;
4423         u64 ext_offset;
4424         u64 search_end = (u64)-1;
4425         u32 nritems;
4426         int nr_scaned = 0;
4427         int extent_locked = 0;
4428         int extent_type;
4429         int ret;
4430
4431         memcpy(&key, leaf_key, sizeof(key));
4432         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4433                 if (key.objectid < ref_path->owner_objectid ||
4434                     (key.objectid == ref_path->owner_objectid &&
4435                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4436                         key.objectid = ref_path->owner_objectid;
4437                         key.type = BTRFS_EXTENT_DATA_KEY;
4438                         key.offset = 0;
4439                 }
4440         }
4441
4442         while (1) {
4443                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4444                 if (ret < 0)
4445                         goto out;
4446
4447                 leaf = path->nodes[0];
4448                 nritems = btrfs_header_nritems(leaf);
4449 next:
4450                 if (extent_locked && ret > 0) {
4451                         /*
4452                          * the file extent item was modified by someone
4453                          * before the extent got locked.
4454                          */
4455                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4456                                       lock_end, GFP_NOFS);
4457                         extent_locked = 0;
4458                 }
4459
4460                 if (path->slots[0] >= nritems) {
4461                         if (++nr_scaned > 2)
4462                                 break;
4463
4464                         BUG_ON(extent_locked);
4465                         ret = btrfs_next_leaf(root, path);
4466                         if (ret < 0)
4467                                 goto out;
4468                         if (ret > 0)
4469                                 break;
4470                         leaf = path->nodes[0];
4471                         nritems = btrfs_header_nritems(leaf);
4472                 }
4473
4474                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4475
4476                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4477                         if ((key.objectid > ref_path->owner_objectid) ||
4478                             (key.objectid == ref_path->owner_objectid &&
4479                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4480                             key.offset >= search_end)
4481                                 break;
4482                 }
4483
4484                 if (inode && key.objectid != inode->i_ino) {
4485                         BUG_ON(extent_locked);
4486                         btrfs_release_path(root, path);
4487                         mutex_unlock(&inode->i_mutex);
4488                         iput(inode);
4489                         inode = NULL;
4490                         continue;
4491                 }
4492
4493                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4494                         path->slots[0]++;
4495                         ret = 1;
4496                         goto next;
4497                 }
4498                 fi = btrfs_item_ptr(leaf, path->slots[0],
4499                                     struct btrfs_file_extent_item);
4500                 extent_type = btrfs_file_extent_type(leaf, fi);
4501                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4502                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4503                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4504                      extent_key->objectid)) {
4505                         path->slots[0]++;
4506                         ret = 1;
4507                         goto next;
4508                 }
4509
4510                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4511                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4512
4513                 if (search_end == (u64)-1) {
4514                         search_end = key.offset - ext_offset +
4515                                 btrfs_file_extent_ram_bytes(leaf, fi);
4516                 }
4517
4518                 if (!extent_locked) {
4519                         lock_start = key.offset;
4520                         lock_end = lock_start + num_bytes - 1;
4521                 } else {
4522                         if (lock_start > key.offset ||
4523                             lock_end + 1 < key.offset + num_bytes) {
4524                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4525                                               lock_start, lock_end, GFP_NOFS);
4526                                 extent_locked = 0;
4527                         }
4528                 }
4529
4530                 if (!inode) {
4531                         btrfs_release_path(root, path);
4532
4533                         inode = btrfs_iget_locked(root->fs_info->sb,
4534                                                   key.objectid, root);
4535                         if (inode->i_state & I_NEW) {
4536                                 BTRFS_I(inode)->root = root;
4537                                 BTRFS_I(inode)->location.objectid =
4538                                         key.objectid;
4539                                 BTRFS_I(inode)->location.type =
4540                                         BTRFS_INODE_ITEM_KEY;
4541                                 BTRFS_I(inode)->location.offset = 0;
4542                                 btrfs_read_locked_inode(inode);
4543                                 unlock_new_inode(inode);
4544                         }
4545                         /*
4546                          * some code call btrfs_commit_transaction while
4547                          * holding the i_mutex, so we can't use mutex_lock
4548                          * here.
4549                          */
4550                         if (is_bad_inode(inode) ||
4551                             !mutex_trylock(&inode->i_mutex)) {
4552                                 iput(inode);
4553                                 inode = NULL;
4554                                 key.offset = (u64)-1;
4555                                 goto skip;
4556                         }
4557                 }
4558
4559                 if (!extent_locked) {
4560                         struct btrfs_ordered_extent *ordered;
4561
4562                         btrfs_release_path(root, path);
4563
4564                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4565                                     lock_end, GFP_NOFS);
4566                         ordered = btrfs_lookup_first_ordered_extent(inode,
4567                                                                     lock_end);
4568                         if (ordered &&
4569                             ordered->file_offset <= lock_end &&
4570                             ordered->file_offset + ordered->len > lock_start) {
4571                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4572                                               lock_start, lock_end, GFP_NOFS);
4573                                 btrfs_start_ordered_extent(inode, ordered, 1);
4574                                 btrfs_put_ordered_extent(ordered);
4575                                 key.offset += num_bytes;
4576                                 goto skip;
4577                         }
4578                         if (ordered)
4579                                 btrfs_put_ordered_extent(ordered);
4580
4581                         extent_locked = 1;
4582                         continue;
4583                 }
4584
4585                 if (nr_extents == 1) {
4586                         /* update extent pointer in place */
4587                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4588                                                 new_extents[0].disk_bytenr);
4589                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4590                                                 new_extents[0].disk_num_bytes);
4591                         btrfs_mark_buffer_dirty(leaf);
4592
4593                         btrfs_drop_extent_cache(inode, key.offset,
4594                                                 key.offset + num_bytes - 1, 0);
4595
4596                         ret = btrfs_inc_extent_ref(trans, root,
4597                                                 new_extents[0].disk_bytenr,
4598                                                 new_extents[0].disk_num_bytes,
4599                                                 leaf->start,
4600                                                 root->root_key.objectid,
4601                                                 trans->transid,
4602                                                 key.objectid);
4603                         BUG_ON(ret);
4604
4605                         ret = btrfs_free_extent(trans, root,
4606                                                 extent_key->objectid,
4607                                                 extent_key->offset,
4608                                                 leaf->start,
4609                                                 btrfs_header_owner(leaf),
4610                                                 btrfs_header_generation(leaf),
4611                                                 key.objectid, 0);
4612                         BUG_ON(ret);
4613
4614                         btrfs_release_path(root, path);
4615                         key.offset += num_bytes;
4616                 } else {
4617                         BUG_ON(1);
4618 #if 0
4619                         u64 alloc_hint;
4620                         u64 extent_len;
4621                         int i;
4622                         /*
4623                          * drop old extent pointer at first, then insert the
4624                          * new pointers one bye one
4625                          */
4626                         btrfs_release_path(root, path);
4627                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4628                                                  key.offset + num_bytes,
4629                                                  key.offset, &alloc_hint);
4630                         BUG_ON(ret);
4631
4632                         for (i = 0; i < nr_extents; i++) {
4633                                 if (ext_offset >= new_extents[i].num_bytes) {
4634                                         ext_offset -= new_extents[i].num_bytes;
4635                                         continue;
4636                                 }
4637                                 extent_len = min(new_extents[i].num_bytes -
4638                                                  ext_offset, num_bytes);
4639
4640                                 ret = btrfs_insert_empty_item(trans, root,
4641                                                               path, &key,
4642                                                               sizeof(*fi));
4643                                 BUG_ON(ret);
4644
4645                                 leaf = path->nodes[0];
4646                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4647                                                 struct btrfs_file_extent_item);
4648                                 btrfs_set_file_extent_generation(leaf, fi,
4649                                                         trans->transid);
4650                                 btrfs_set_file_extent_type(leaf, fi,
4651                                                         BTRFS_FILE_EXTENT_REG);
4652                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4653                                                 new_extents[i].disk_bytenr);
4654                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4655                                                 new_extents[i].disk_num_bytes);
4656                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4657                                                 new_extents[i].ram_bytes);
4658
4659                                 btrfs_set_file_extent_compression(leaf, fi,
4660                                                 new_extents[i].compression);
4661                                 btrfs_set_file_extent_encryption(leaf, fi,
4662                                                 new_extents[i].encryption);
4663                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4664                                                 new_extents[i].other_encoding);
4665
4666                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4667                                                         extent_len);
4668                                 ext_offset += new_extents[i].offset;
4669                                 btrfs_set_file_extent_offset(leaf, fi,
4670                                                         ext_offset);
4671                                 btrfs_mark_buffer_dirty(leaf);
4672
4673                                 btrfs_drop_extent_cache(inode, key.offset,
4674                                                 key.offset + extent_len - 1, 0);
4675
4676                                 ret = btrfs_inc_extent_ref(trans, root,
4677                                                 new_extents[i].disk_bytenr,
4678                                                 new_extents[i].disk_num_bytes,
4679                                                 leaf->start,
4680                                                 root->root_key.objectid,
4681                                                 trans->transid, key.objectid);
4682                                 BUG_ON(ret);
4683                                 btrfs_release_path(root, path);
4684
4685                                 inode_add_bytes(inode, extent_len);
4686
4687                                 ext_offset = 0;
4688                                 num_bytes -= extent_len;
4689                                 key.offset += extent_len;
4690
4691                                 if (num_bytes == 0)
4692                                         break;
4693                         }
4694                         BUG_ON(i >= nr_extents);
4695 #endif
4696                 }
4697
4698                 if (extent_locked) {
4699                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4700                                       lock_end, GFP_NOFS);
4701                         extent_locked = 0;
4702                 }
4703 skip:
4704                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4705                     key.offset >= search_end)
4706                         break;
4707
4708                 cond_resched();
4709         }
4710         ret = 0;
4711 out:
4712         btrfs_release_path(root, path);
4713         if (inode) {
4714                 mutex_unlock(&inode->i_mutex);
4715                 if (extent_locked) {
4716                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4717                                       lock_end, GFP_NOFS);
4718                 }
4719                 iput(inode);
4720         }
4721         return ret;
4722 }
4723
4724 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4725                                struct btrfs_root *root,
4726                                struct extent_buffer *buf, u64 orig_start)
4727 {
4728         int level;
4729         int ret;
4730
4731         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4732         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4733
4734         level = btrfs_header_level(buf);
4735         if (level == 0) {
4736                 struct btrfs_leaf_ref *ref;
4737                 struct btrfs_leaf_ref *orig_ref;
4738
4739                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4740                 if (!orig_ref)
4741                         return -ENOENT;
4742
4743                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4744                 if (!ref) {
4745                         btrfs_free_leaf_ref(root, orig_ref);
4746                         return -ENOMEM;
4747                 }
4748
4749                 ref->nritems = orig_ref->nritems;
4750                 memcpy(ref->extents, orig_ref->extents,
4751                         sizeof(ref->extents[0]) * ref->nritems);
4752
4753                 btrfs_free_leaf_ref(root, orig_ref);
4754
4755                 ref->root_gen = trans->transid;
4756                 ref->bytenr = buf->start;
4757                 ref->owner = btrfs_header_owner(buf);
4758                 ref->generation = btrfs_header_generation(buf);
4759
4760                 ret = btrfs_add_leaf_ref(root, ref, 0);
4761                 WARN_ON(ret);
4762                 btrfs_free_leaf_ref(root, ref);
4763         }
4764         return 0;
4765 }
4766
4767 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4768                                         struct extent_buffer *leaf,
4769                                         struct btrfs_block_group_cache *group,
4770                                         struct btrfs_root *target_root)
4771 {
4772         struct btrfs_key key;
4773         struct inode *inode = NULL;
4774         struct btrfs_file_extent_item *fi;
4775         u64 num_bytes;
4776         u64 skip_objectid = 0;
4777         u32 nritems;
4778         u32 i;
4779
4780         nritems = btrfs_header_nritems(leaf);
4781         for (i = 0; i < nritems; i++) {
4782                 btrfs_item_key_to_cpu(leaf, &key, i);
4783                 if (key.objectid == skip_objectid ||
4784                     key.type != BTRFS_EXTENT_DATA_KEY)
4785                         continue;
4786                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4787                 if (btrfs_file_extent_type(leaf, fi) ==
4788                     BTRFS_FILE_EXTENT_INLINE)
4789                         continue;
4790                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4791                         continue;
4792                 if (!inode || inode->i_ino != key.objectid) {
4793                         iput(inode);
4794                         inode = btrfs_ilookup(target_root->fs_info->sb,
4795                                               key.objectid, target_root, 1);
4796                 }
4797                 if (!inode) {
4798                         skip_objectid = key.objectid;
4799                         continue;
4800                 }
4801                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4802
4803                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4804                             key.offset + num_bytes - 1, GFP_NOFS);
4805                 btrfs_drop_extent_cache(inode, key.offset,
4806                                         key.offset + num_bytes - 1, 1);
4807                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4808                               key.offset + num_bytes - 1, GFP_NOFS);
4809                 cond_resched();
4810         }
4811         iput(inode);
4812         return 0;
4813 }
4814
4815 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4816                                         struct btrfs_root *root,
4817                                         struct extent_buffer *leaf,
4818                                         struct btrfs_block_group_cache *group,
4819                                         struct inode *reloc_inode)
4820 {
4821         struct btrfs_key key;
4822         struct btrfs_key extent_key;
4823         struct btrfs_file_extent_item *fi;
4824         struct btrfs_leaf_ref *ref;
4825         struct disk_extent *new_extent;
4826         u64 bytenr;
4827         u64 num_bytes;
4828         u32 nritems;
4829         u32 i;
4830         int ext_index;
4831         int nr_extent;
4832         int ret;
4833
4834         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4835         BUG_ON(!new_extent);
4836
4837         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4838         BUG_ON(!ref);
4839
4840         ext_index = -1;
4841         nritems = btrfs_header_nritems(leaf);
4842         for (i = 0; i < nritems; i++) {
4843                 btrfs_item_key_to_cpu(leaf, &key, i);
4844                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4845                         continue;
4846                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4847                 if (btrfs_file_extent_type(leaf, fi) ==
4848                     BTRFS_FILE_EXTENT_INLINE)
4849                         continue;
4850                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4851                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4852                 if (bytenr == 0)
4853                         continue;
4854
4855                 ext_index++;
4856                 if (bytenr >= group->key.objectid + group->key.offset ||
4857                     bytenr + num_bytes <= group->key.objectid)
4858                         continue;
4859
4860                 extent_key.objectid = bytenr;
4861                 extent_key.offset = num_bytes;
4862                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4863                 nr_extent = 1;
4864                 ret = get_new_locations(reloc_inode, &extent_key,
4865                                         group->key.objectid, 1,
4866                                         &new_extent, &nr_extent);
4867                 if (ret > 0)
4868                         continue;
4869                 BUG_ON(ret < 0);
4870
4871                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4872                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4873                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4874                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4875
4876                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4877                                                 new_extent->disk_bytenr);
4878                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4879                                                 new_extent->disk_num_bytes);
4880                 btrfs_mark_buffer_dirty(leaf);
4881
4882                 ret = btrfs_inc_extent_ref(trans, root,
4883                                         new_extent->disk_bytenr,
4884                                         new_extent->disk_num_bytes,
4885                                         leaf->start,
4886                                         root->root_key.objectid,
4887                                         trans->transid, key.objectid);
4888                 BUG_ON(ret);
4889
4890                 ret = btrfs_free_extent(trans, root,
4891                                         bytenr, num_bytes, leaf->start,
4892                                         btrfs_header_owner(leaf),
4893                                         btrfs_header_generation(leaf),
4894                                         key.objectid, 0);
4895                 BUG_ON(ret);
4896                 cond_resched();
4897         }
4898         kfree(new_extent);
4899         BUG_ON(ext_index + 1 != ref->nritems);
4900         btrfs_free_leaf_ref(root, ref);
4901         return 0;
4902 }
4903
4904 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4905                           struct btrfs_root *root)
4906 {
4907         struct btrfs_root *reloc_root;
4908         int ret;
4909
4910         if (root->reloc_root) {
4911                 reloc_root = root->reloc_root;
4912                 root->reloc_root = NULL;
4913                 list_add(&reloc_root->dead_list,
4914                          &root->fs_info->dead_reloc_roots);
4915
4916                 btrfs_set_root_bytenr(&reloc_root->root_item,
4917                                       reloc_root->node->start);
4918                 btrfs_set_root_level(&root->root_item,
4919                                      btrfs_header_level(reloc_root->node));
4920                 memset(&reloc_root->root_item.drop_progress, 0,
4921                         sizeof(struct btrfs_disk_key));
4922                 reloc_root->root_item.drop_level = 0;
4923
4924                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4925                                         &reloc_root->root_key,
4926                                         &reloc_root->root_item);
4927                 BUG_ON(ret);
4928         }
4929         return 0;
4930 }
4931
4932 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4933 {
4934         struct btrfs_trans_handle *trans;
4935         struct btrfs_root *reloc_root;
4936         struct btrfs_root *prev_root = NULL;
4937         struct list_head dead_roots;
4938         int ret;
4939         unsigned long nr;
4940
4941         INIT_LIST_HEAD(&dead_roots);
4942         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4943
4944         while (!list_empty(&dead_roots)) {
4945                 reloc_root = list_entry(dead_roots.prev,
4946                                         struct btrfs_root, dead_list);
4947                 list_del_init(&reloc_root->dead_list);
4948
4949                 BUG_ON(reloc_root->commit_root != NULL);
4950                 while (1) {
4951                         trans = btrfs_join_transaction(root, 1);
4952                         BUG_ON(!trans);
4953
4954                         mutex_lock(&root->fs_info->drop_mutex);
4955                         ret = btrfs_drop_snapshot(trans, reloc_root);
4956                         if (ret != -EAGAIN)
4957                                 break;
4958                         mutex_unlock(&root->fs_info->drop_mutex);
4959
4960                         nr = trans->blocks_used;
4961                         ret = btrfs_end_transaction(trans, root);
4962                         BUG_ON(ret);
4963                         btrfs_btree_balance_dirty(root, nr);
4964                 }
4965
4966                 free_extent_buffer(reloc_root->node);
4967
4968                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4969                                      &reloc_root->root_key);
4970                 BUG_ON(ret);
4971                 mutex_unlock(&root->fs_info->drop_mutex);
4972
4973                 nr = trans->blocks_used;
4974                 ret = btrfs_end_transaction(trans, root);
4975                 BUG_ON(ret);
4976                 btrfs_btree_balance_dirty(root, nr);
4977
4978                 kfree(prev_root);
4979                 prev_root = reloc_root;
4980         }
4981         if (prev_root) {
4982                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4983                 kfree(prev_root);
4984         }
4985         return 0;
4986 }
4987
4988 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4989 {
4990         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4991         return 0;
4992 }
4993
4994 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4995 {
4996         struct btrfs_root *reloc_root;
4997         struct btrfs_trans_handle *trans;
4998         struct btrfs_key location;
4999         int found;
5000         int ret;
5001
5002         mutex_lock(&root->fs_info->tree_reloc_mutex);
5003         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5004         BUG_ON(ret);
5005         found = !list_empty(&root->fs_info->dead_reloc_roots);
5006         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5007
5008         if (found) {
5009                 trans = btrfs_start_transaction(root, 1);
5010                 BUG_ON(!trans);
5011                 ret = btrfs_commit_transaction(trans, root);
5012                 BUG_ON(ret);
5013         }
5014
5015         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5016         location.offset = (u64)-1;
5017         location.type = BTRFS_ROOT_ITEM_KEY;
5018
5019         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5020         BUG_ON(!reloc_root);
5021         btrfs_orphan_cleanup(reloc_root);
5022         return 0;
5023 }
5024
5025 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5026                                     struct btrfs_root *root)
5027 {
5028         struct btrfs_root *reloc_root;
5029         struct extent_buffer *eb;
5030         struct btrfs_root_item *root_item;
5031         struct btrfs_key root_key;
5032         int ret;
5033
5034         BUG_ON(!root->ref_cows);
5035         if (root->reloc_root)
5036                 return 0;
5037
5038         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5039         BUG_ON(!root_item);
5040
5041         ret = btrfs_copy_root(trans, root, root->commit_root,
5042                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5043         BUG_ON(ret);
5044
5045         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5046         root_key.offset = root->root_key.objectid;
5047         root_key.type = BTRFS_ROOT_ITEM_KEY;
5048
5049         memcpy(root_item, &root->root_item, sizeof(root_item));
5050         btrfs_set_root_refs(root_item, 0);
5051         btrfs_set_root_bytenr(root_item, eb->start);
5052         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5053         btrfs_set_root_generation(root_item, trans->transid);
5054
5055         btrfs_tree_unlock(eb);
5056         free_extent_buffer(eb);
5057
5058         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5059                                 &root_key, root_item);
5060         BUG_ON(ret);
5061         kfree(root_item);
5062
5063         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5064                                                  &root_key);
5065         BUG_ON(!reloc_root);
5066         reloc_root->last_trans = trans->transid;
5067         reloc_root->commit_root = NULL;
5068         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5069
5070         root->reloc_root = reloc_root;
5071         return 0;
5072 }
5073
5074 /*
5075  * Core function of space balance.
5076  *
5077  * The idea is using reloc trees to relocate tree blocks in reference
5078  * counted roots. There is one reloc tree for each subvol, and all
5079  * reloc trees share same root key objectid. Reloc trees are snapshots
5080  * of the latest committed roots of subvols (root->commit_root).
5081  *
5082  * To relocate a tree block referenced by a subvol, there are two steps.
5083  * COW the block through subvol's reloc tree, then update block pointer
5084  * in the subvol to point to the new block. Since all reloc trees share
5085  * same root key objectid, doing special handing for tree blocks owned
5086  * by them is easy. Once a tree block has been COWed in one reloc tree,
5087  * we can use the resulting new block directly when the same block is
5088  * required to COW again through other reloc trees. By this way, relocated
5089  * tree blocks are shared between reloc trees, so they are also shared
5090  * between subvols.
5091  */
5092 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5093                                       struct btrfs_root *root,
5094                                       struct btrfs_path *path,
5095                                       struct btrfs_key *first_key,
5096                                       struct btrfs_ref_path *ref_path,
5097                                       struct btrfs_block_group_cache *group,
5098                                       struct inode *reloc_inode)
5099 {
5100         struct btrfs_root *reloc_root;
5101         struct extent_buffer *eb = NULL;
5102         struct btrfs_key *keys;
5103         u64 *nodes;
5104         int level;
5105         int shared_level;
5106         int lowest_level = 0;
5107         int ret;
5108
5109         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5110                 lowest_level = ref_path->owner_objectid;
5111
5112         if (!root->ref_cows) {
5113                 path->lowest_level = lowest_level;
5114                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5115                 BUG_ON(ret < 0);
5116                 path->lowest_level = 0;
5117                 btrfs_release_path(root, path);
5118                 return 0;
5119         }
5120
5121         mutex_lock(&root->fs_info->tree_reloc_mutex);
5122         ret = init_reloc_tree(trans, root);
5123         BUG_ON(ret);
5124         reloc_root = root->reloc_root;
5125
5126         shared_level = ref_path->shared_level;
5127         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5128
5129         keys = ref_path->node_keys;
5130         nodes = ref_path->new_nodes;
5131         memset(&keys[shared_level + 1], 0,
5132                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5133         memset(&nodes[shared_level + 1], 0,
5134                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5135
5136         if (nodes[lowest_level] == 0) {
5137                 path->lowest_level = lowest_level;
5138                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5139                                         0, 1);
5140                 BUG_ON(ret);
5141                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5142                         eb = path->nodes[level];
5143                         if (!eb || eb == reloc_root->node)
5144                                 break;
5145                         nodes[level] = eb->start;
5146                         if (level == 0)
5147                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5148                         else
5149                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5150                 }
5151                 if (nodes[0] &&
5152                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5153                         eb = path->nodes[0];
5154                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5155                                                       group, reloc_inode);
5156                         BUG_ON(ret);
5157                 }
5158                 btrfs_release_path(reloc_root, path);
5159         } else {
5160                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5161                                        lowest_level);
5162                 BUG_ON(ret);
5163         }
5164
5165         /*
5166          * replace tree blocks in the fs tree with tree blocks in
5167          * the reloc tree.
5168          */
5169         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5170         BUG_ON(ret < 0);
5171
5172         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5173                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5174                                         0, 0);
5175                 BUG_ON(ret);
5176                 extent_buffer_get(path->nodes[0]);
5177                 eb = path->nodes[0];
5178                 btrfs_release_path(reloc_root, path);
5179                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5180                 BUG_ON(ret);
5181                 free_extent_buffer(eb);
5182         }
5183
5184         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5185         path->lowest_level = 0;
5186         return 0;
5187 }
5188
5189 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5190                                         struct btrfs_root *root,
5191                                         struct btrfs_path *path,
5192                                         struct btrfs_key *first_key,
5193                                         struct btrfs_ref_path *ref_path)
5194 {
5195         int ret;
5196
5197         ret = relocate_one_path(trans, root, path, first_key,
5198                                 ref_path, NULL, NULL);
5199         BUG_ON(ret);
5200
5201         return 0;
5202 }
5203
5204 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5205                                     struct btrfs_root *extent_root,
5206                                     struct btrfs_path *path,
5207                                     struct btrfs_key *extent_key)
5208 {
5209         int ret;
5210
5211         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5212         if (ret)
5213                 goto out;
5214         ret = btrfs_del_item(trans, extent_root, path);
5215 out:
5216         btrfs_release_path(extent_root, path);
5217         return ret;
5218 }
5219
5220 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5221                                                 struct btrfs_ref_path *ref_path)
5222 {
5223         struct btrfs_key root_key;
5224
5225         root_key.objectid = ref_path->root_objectid;
5226         root_key.type = BTRFS_ROOT_ITEM_KEY;
5227         if (is_cowonly_root(ref_path->root_objectid))
5228                 root_key.offset = 0;
5229         else
5230                 root_key.offset = (u64)-1;
5231
5232         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5233 }
5234
5235 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5236                                         struct btrfs_path *path,
5237                                         struct btrfs_key *extent_key,
5238                                         struct btrfs_block_group_cache *group,
5239                                         struct inode *reloc_inode, int pass)
5240 {
5241         struct btrfs_trans_handle *trans;
5242         struct btrfs_root *found_root;
5243         struct btrfs_ref_path *ref_path = NULL;
5244         struct disk_extent *new_extents = NULL;
5245         int nr_extents = 0;
5246         int loops;
5247         int ret;
5248         int level;
5249         struct btrfs_key first_key;
5250         u64 prev_block = 0;
5251
5252
5253         trans = btrfs_start_transaction(extent_root, 1);
5254         BUG_ON(!trans);
5255
5256         if (extent_key->objectid == 0) {
5257                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5258                 goto out;
5259         }
5260
5261         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5262         if (!ref_path) {
5263                 ret = -ENOMEM;
5264                 goto out;
5265         }
5266
5267         for (loops = 0; ; loops++) {
5268                 if (loops == 0) {
5269                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5270                                                    extent_key->objectid);
5271                 } else {
5272                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5273                 }
5274                 if (ret < 0)
5275                         goto out;
5276                 if (ret > 0)
5277                         break;
5278
5279                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5280                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5281                         continue;
5282
5283                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5284                 BUG_ON(!found_root);
5285                 /*
5286                  * for reference counted tree, only process reference paths
5287                  * rooted at the latest committed root.
5288                  */
5289                 if (found_root->ref_cows &&
5290                     ref_path->root_generation != found_root->root_key.offset)
5291                         continue;
5292
5293                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5294                         if (pass == 0) {
5295                                 /*
5296                                  * copy data extents to new locations
5297                                  */
5298                                 u64 group_start = group->key.objectid;
5299                                 ret = relocate_data_extent(reloc_inode,
5300                                                            extent_key,
5301                                                            group_start);
5302                                 if (ret < 0)
5303                                         goto out;
5304                                 break;
5305                         }
5306                         level = 0;
5307                 } else {
5308                         level = ref_path->owner_objectid;
5309                 }
5310
5311                 if (prev_block != ref_path->nodes[level]) {
5312                         struct extent_buffer *eb;
5313                         u64 block_start = ref_path->nodes[level];
5314                         u64 block_size = btrfs_level_size(found_root, level);
5315
5316                         eb = read_tree_block(found_root, block_start,
5317                                              block_size, 0);
5318                         btrfs_tree_lock(eb);
5319                         BUG_ON(level != btrfs_header_level(eb));
5320
5321                         if (level == 0)
5322                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5323                         else
5324                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5325
5326                         btrfs_tree_unlock(eb);
5327                         free_extent_buffer(eb);
5328                         prev_block = block_start;
5329                 }
5330
5331                 mutex_lock(&extent_root->fs_info->trans_mutex);
5332                 btrfs_record_root_in_trans(found_root);
5333                 mutex_unlock(&extent_root->fs_info->trans_mutex);
5334                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5335                         /*
5336                          * try to update data extent references while
5337                          * keeping metadata shared between snapshots.
5338                          */
5339                         if (pass == 1) {
5340                                 ret = relocate_one_path(trans, found_root,
5341                                                 path, &first_key, ref_path,
5342                                                 group, reloc_inode);
5343                                 if (ret < 0)
5344                                         goto out;
5345                                 continue;
5346                         }
5347                         /*
5348                          * use fallback method to process the remaining
5349                          * references.
5350                          */
5351                         if (!new_extents) {
5352                                 u64 group_start = group->key.objectid;
5353                                 new_extents = kmalloc(sizeof(*new_extents),
5354                                                       GFP_NOFS);
5355                                 nr_extents = 1;
5356                                 ret = get_new_locations(reloc_inode,
5357                                                         extent_key,
5358                                                         group_start, 1,
5359                                                         &new_extents,
5360                                                         &nr_extents);
5361                                 if (ret)
5362                                         goto out;
5363                         }
5364                         ret = replace_one_extent(trans, found_root,
5365                                                 path, extent_key,
5366                                                 &first_key, ref_path,
5367                                                 new_extents, nr_extents);
5368                 } else {
5369                         ret = relocate_tree_block(trans, found_root, path,
5370                                                   &first_key, ref_path);
5371                 }
5372                 if (ret < 0)
5373                         goto out;
5374         }
5375         ret = 0;
5376 out:
5377         btrfs_end_transaction(trans, extent_root);
5378         kfree(new_extents);
5379         kfree(ref_path);
5380         return ret;
5381 }
5382
5383 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5384 {
5385         u64 num_devices;
5386         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5387                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5388
5389         num_devices = root->fs_info->fs_devices->rw_devices;
5390         if (num_devices == 1) {
5391                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5392                 stripped = flags & ~stripped;
5393
5394                 /* turn raid0 into single device chunks */
5395                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5396                         return stripped;
5397
5398                 /* turn mirroring into duplication */
5399                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5400                              BTRFS_BLOCK_GROUP_RAID10))
5401                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5402                 return flags;
5403         } else {
5404                 /* they already had raid on here, just return */
5405                 if (flags & stripped)
5406                         return flags;
5407
5408                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5409                 stripped = flags & ~stripped;
5410
5411                 /* switch duplicated blocks with raid1 */
5412                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5413                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5414
5415                 /* turn single device chunks into raid0 */
5416                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5417         }
5418         return flags;
5419 }
5420
5421 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5422                      struct btrfs_block_group_cache *shrink_block_group,
5423                      int force)
5424 {
5425         struct btrfs_trans_handle *trans;
5426         u64 new_alloc_flags;
5427         u64 calc;
5428
5429         spin_lock(&shrink_block_group->lock);
5430         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5431                 spin_unlock(&shrink_block_group->lock);
5432
5433                 trans = btrfs_start_transaction(root, 1);
5434                 spin_lock(&shrink_block_group->lock);
5435
5436                 new_alloc_flags = update_block_group_flags(root,
5437                                                    shrink_block_group->flags);
5438                 if (new_alloc_flags != shrink_block_group->flags) {
5439                         calc =
5440                              btrfs_block_group_used(&shrink_block_group->item);
5441                 } else {
5442                         calc = shrink_block_group->key.offset;
5443                 }
5444                 spin_unlock(&shrink_block_group->lock);
5445
5446                 do_chunk_alloc(trans, root->fs_info->extent_root,
5447                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5448
5449                 btrfs_end_transaction(trans, root);
5450         } else
5451                 spin_unlock(&shrink_block_group->lock);
5452         return 0;
5453 }
5454
5455 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5456                                  struct btrfs_root *root,
5457                                  u64 objectid, u64 size)
5458 {
5459         struct btrfs_path *path;
5460         struct btrfs_inode_item *item;
5461         struct extent_buffer *leaf;
5462         int ret;
5463
5464         path = btrfs_alloc_path();
5465         if (!path)
5466                 return -ENOMEM;
5467
5468         path->leave_spinning = 1;
5469         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5470         if (ret)
5471                 goto out;
5472
5473         leaf = path->nodes[0];
5474         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5475         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5476         btrfs_set_inode_generation(leaf, item, 1);
5477         btrfs_set_inode_size(leaf, item, size);
5478         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5479         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5480         btrfs_mark_buffer_dirty(leaf);
5481         btrfs_release_path(root, path);
5482 out:
5483         btrfs_free_path(path);
5484         return ret;
5485 }
5486
5487 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5488                                         struct btrfs_block_group_cache *group)
5489 {
5490         struct inode *inode = NULL;
5491         struct btrfs_trans_handle *trans;
5492         struct btrfs_root *root;
5493         struct btrfs_key root_key;
5494         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5495         int err = 0;
5496
5497         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5498         root_key.type = BTRFS_ROOT_ITEM_KEY;
5499         root_key.offset = (u64)-1;
5500         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5501         if (IS_ERR(root))
5502                 return ERR_CAST(root);
5503
5504         trans = btrfs_start_transaction(root, 1);
5505         BUG_ON(!trans);
5506
5507         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5508         if (err)
5509                 goto out;
5510
5511         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5512         BUG_ON(err);
5513
5514         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5515                                        group->key.offset, 0, group->key.offset,
5516                                        0, 0, 0);
5517         BUG_ON(err);
5518
5519         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5520         if (inode->i_state & I_NEW) {
5521                 BTRFS_I(inode)->root = root;
5522                 BTRFS_I(inode)->location.objectid = objectid;
5523                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5524                 BTRFS_I(inode)->location.offset = 0;
5525                 btrfs_read_locked_inode(inode);
5526                 unlock_new_inode(inode);
5527                 BUG_ON(is_bad_inode(inode));
5528         } else {
5529                 BUG_ON(1);
5530         }
5531         BTRFS_I(inode)->index_cnt = group->key.objectid;
5532
5533         err = btrfs_orphan_add(trans, inode);
5534 out:
5535         btrfs_end_transaction(trans, root);
5536         if (err) {
5537                 if (inode)
5538                         iput(inode);
5539                 inode = ERR_PTR(err);
5540         }
5541         return inode;
5542 }
5543
5544 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5545 {
5546
5547         struct btrfs_ordered_sum *sums;
5548         struct btrfs_sector_sum *sector_sum;
5549         struct btrfs_ordered_extent *ordered;
5550         struct btrfs_root *root = BTRFS_I(inode)->root;
5551         struct list_head list;
5552         size_t offset;
5553         int ret;
5554         u64 disk_bytenr;
5555
5556         INIT_LIST_HEAD(&list);
5557
5558         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5559         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5560
5561         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5562         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5563                                        disk_bytenr + len - 1, &list);
5564
5565         while (!list_empty(&list)) {
5566                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5567                 list_del_init(&sums->list);
5568
5569                 sector_sum = sums->sums;
5570                 sums->bytenr = ordered->start;
5571
5572                 offset = 0;
5573                 while (offset < sums->len) {
5574                         sector_sum->bytenr += ordered->start - disk_bytenr;
5575                         sector_sum++;
5576                         offset += root->sectorsize;
5577                 }
5578
5579                 btrfs_add_ordered_sum(inode, ordered, sums);
5580         }
5581         btrfs_put_ordered_extent(ordered);
5582         return 0;
5583 }
5584
5585 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5586 {
5587         struct btrfs_trans_handle *trans;
5588         struct btrfs_path *path;
5589         struct btrfs_fs_info *info = root->fs_info;
5590         struct extent_buffer *leaf;
5591         struct inode *reloc_inode;
5592         struct btrfs_block_group_cache *block_group;
5593         struct btrfs_key key;
5594         u64 skipped;
5595         u64 cur_byte;
5596         u64 total_found;
5597         u32 nritems;
5598         int ret;
5599         int progress;
5600         int pass = 0;
5601
5602         root = root->fs_info->extent_root;
5603
5604         block_group = btrfs_lookup_block_group(info, group_start);
5605         BUG_ON(!block_group);
5606
5607         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5608                (unsigned long long)block_group->key.objectid,
5609                (unsigned long long)block_group->flags);
5610
5611         path = btrfs_alloc_path();
5612         BUG_ON(!path);
5613
5614         reloc_inode = create_reloc_inode(info, block_group);
5615         BUG_ON(IS_ERR(reloc_inode));
5616
5617         __alloc_chunk_for_shrink(root, block_group, 1);
5618         set_block_group_readonly(block_group);
5619
5620         btrfs_start_delalloc_inodes(info->tree_root);
5621         btrfs_wait_ordered_extents(info->tree_root, 0);
5622 again:
5623         skipped = 0;
5624         total_found = 0;
5625         progress = 0;
5626         key.objectid = block_group->key.objectid;
5627         key.offset = 0;
5628         key.type = 0;
5629         cur_byte = key.objectid;
5630
5631         trans = btrfs_start_transaction(info->tree_root, 1);
5632         btrfs_commit_transaction(trans, info->tree_root);
5633
5634         mutex_lock(&root->fs_info->cleaner_mutex);
5635         btrfs_clean_old_snapshots(info->tree_root);
5636         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5637         mutex_unlock(&root->fs_info->cleaner_mutex);
5638
5639         trans = btrfs_start_transaction(info->tree_root, 1);
5640         btrfs_commit_transaction(trans, info->tree_root);
5641
5642         while (1) {
5643                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5644                 if (ret < 0)
5645                         goto out;
5646 next:
5647                 leaf = path->nodes[0];
5648                 nritems = btrfs_header_nritems(leaf);
5649                 if (path->slots[0] >= nritems) {
5650                         ret = btrfs_next_leaf(root, path);
5651                         if (ret < 0)
5652                                 goto out;
5653                         if (ret == 1) {
5654                                 ret = 0;
5655                                 break;
5656                         }
5657                         leaf = path->nodes[0];
5658                         nritems = btrfs_header_nritems(leaf);
5659                 }
5660
5661                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5662
5663                 if (key.objectid >= block_group->key.objectid +
5664                     block_group->key.offset)
5665                         break;
5666
5667                 if (progress && need_resched()) {
5668                         btrfs_release_path(root, path);
5669                         cond_resched();
5670                         progress = 0;
5671                         continue;
5672                 }
5673                 progress = 1;
5674
5675                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5676                     key.objectid + key.offset <= cur_byte) {
5677                         path->slots[0]++;
5678                         goto next;
5679                 }
5680
5681                 total_found++;
5682                 cur_byte = key.objectid + key.offset;
5683                 btrfs_release_path(root, path);
5684
5685                 __alloc_chunk_for_shrink(root, block_group, 0);
5686                 ret = relocate_one_extent(root, path, &key, block_group,
5687                                           reloc_inode, pass);
5688                 BUG_ON(ret < 0);
5689                 if (ret > 0)
5690                         skipped++;
5691
5692                 key.objectid = cur_byte;
5693                 key.type = 0;
5694                 key.offset = 0;
5695         }
5696
5697         btrfs_release_path(root, path);
5698
5699         if (pass == 0) {
5700                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5701                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5702         }
5703
5704         if (total_found > 0) {
5705                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5706                        (unsigned long long)total_found, pass);
5707                 pass++;
5708                 if (total_found == skipped && pass > 2) {
5709                         iput(reloc_inode);
5710                         reloc_inode = create_reloc_inode(info, block_group);
5711                         pass = 0;
5712                 }
5713                 goto again;
5714         }
5715
5716         /* delete reloc_inode */
5717         iput(reloc_inode);
5718
5719         /* unpin extents in this range */
5720         trans = btrfs_start_transaction(info->tree_root, 1);
5721         btrfs_commit_transaction(trans, info->tree_root);
5722
5723         spin_lock(&block_group->lock);
5724         WARN_ON(block_group->pinned > 0);
5725         WARN_ON(block_group->reserved > 0);
5726         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5727         spin_unlock(&block_group->lock);
5728         put_block_group(block_group);
5729         ret = 0;
5730 out:
5731         btrfs_free_path(path);
5732         return ret;
5733 }
5734
5735 static int find_first_block_group(struct btrfs_root *root,
5736                 struct btrfs_path *path, struct btrfs_key *key)
5737 {
5738         int ret = 0;
5739         struct btrfs_key found_key;
5740         struct extent_buffer *leaf;
5741         int slot;
5742
5743         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5744         if (ret < 0)
5745                 goto out;
5746
5747         while (1) {
5748                 slot = path->slots[0];
5749                 leaf = path->nodes[0];
5750                 if (slot >= btrfs_header_nritems(leaf)) {
5751                         ret = btrfs_next_leaf(root, path);
5752                         if (ret == 0)
5753                                 continue;
5754                         if (ret < 0)
5755                                 goto out;
5756                         break;
5757                 }
5758                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5759
5760                 if (found_key.objectid >= key->objectid &&
5761                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5762                         ret = 0;
5763                         goto out;
5764                 }
5765                 path->slots[0]++;
5766         }
5767         ret = -ENOENT;
5768 out:
5769         return ret;
5770 }
5771
5772 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5773 {
5774         struct btrfs_block_group_cache *block_group;
5775         struct btrfs_space_info *space_info;
5776         struct rb_node *n;
5777
5778         spin_lock(&info->block_group_cache_lock);
5779         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5780                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5781                                        cache_node);
5782                 rb_erase(&block_group->cache_node,
5783                          &info->block_group_cache_tree);
5784                 spin_unlock(&info->block_group_cache_lock);
5785
5786                 btrfs_remove_free_space_cache(block_group);
5787                 down_write(&block_group->space_info->groups_sem);
5788                 list_del(&block_group->list);
5789                 up_write(&block_group->space_info->groups_sem);
5790
5791                 WARN_ON(atomic_read(&block_group->count) != 1);
5792                 kfree(block_group);
5793
5794                 spin_lock(&info->block_group_cache_lock);
5795         }
5796         spin_unlock(&info->block_group_cache_lock);
5797
5798         /* now that all the block groups are freed, go through and
5799          * free all the space_info structs.  This is only called during
5800          * the final stages of unmount, and so we know nobody is
5801          * using them.  We call synchronize_rcu() once before we start,
5802          * just to be on the safe side.
5803          */
5804         synchronize_rcu();
5805
5806         while(!list_empty(&info->space_info)) {
5807                 space_info = list_entry(info->space_info.next,
5808                                         struct btrfs_space_info,
5809                                         list);
5810
5811                 list_del(&space_info->list);
5812                 kfree(space_info);
5813         }
5814         return 0;
5815 }
5816
5817 int btrfs_read_block_groups(struct btrfs_root *root)
5818 {
5819         struct btrfs_path *path;
5820         int ret;
5821         struct btrfs_block_group_cache *cache;
5822         struct btrfs_fs_info *info = root->fs_info;
5823         struct btrfs_space_info *space_info;
5824         struct btrfs_key key;
5825         struct btrfs_key found_key;
5826         struct extent_buffer *leaf;
5827
5828         root = info->extent_root;
5829         key.objectid = 0;
5830         key.offset = 0;
5831         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5832         path = btrfs_alloc_path();
5833         if (!path)
5834                 return -ENOMEM;
5835
5836         while (1) {
5837                 ret = find_first_block_group(root, path, &key);
5838                 if (ret > 0) {
5839                         ret = 0;
5840                         goto error;
5841                 }
5842                 if (ret != 0)
5843                         goto error;
5844
5845                 leaf = path->nodes[0];
5846                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5847                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5848                 if (!cache) {
5849                         ret = -ENOMEM;
5850                         break;
5851                 }
5852
5853                 atomic_set(&cache->count, 1);
5854                 spin_lock_init(&cache->lock);
5855                 mutex_init(&cache->alloc_mutex);
5856                 mutex_init(&cache->cache_mutex);
5857                 INIT_LIST_HEAD(&cache->list);
5858                 read_extent_buffer(leaf, &cache->item,
5859                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5860                                    sizeof(cache->item));
5861                 memcpy(&cache->key, &found_key, sizeof(found_key));
5862
5863                 key.objectid = found_key.objectid + found_key.offset;
5864                 btrfs_release_path(root, path);
5865                 cache->flags = btrfs_block_group_flags(&cache->item);
5866
5867                 ret = update_space_info(info, cache->flags, found_key.offset,
5868                                         btrfs_block_group_used(&cache->item),
5869                                         &space_info);
5870                 BUG_ON(ret);
5871                 cache->space_info = space_info;
5872                 down_write(&space_info->groups_sem);
5873                 list_add_tail(&cache->list, &space_info->block_groups);
5874                 up_write(&space_info->groups_sem);
5875
5876                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5877                 BUG_ON(ret);
5878
5879                 set_avail_alloc_bits(root->fs_info, cache->flags);
5880                 if (btrfs_chunk_readonly(root, cache->key.objectid))
5881                         set_block_group_readonly(cache);
5882         }
5883         ret = 0;
5884 error:
5885         btrfs_free_path(path);
5886         return ret;
5887 }
5888
5889 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5890                            struct btrfs_root *root, u64 bytes_used,
5891                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5892                            u64 size)
5893 {
5894         int ret;
5895         struct btrfs_root *extent_root;
5896         struct btrfs_block_group_cache *cache;
5897
5898         extent_root = root->fs_info->extent_root;
5899
5900         root->fs_info->last_trans_new_blockgroup = trans->transid;
5901
5902         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5903         if (!cache)
5904                 return -ENOMEM;
5905
5906         cache->key.objectid = chunk_offset;
5907         cache->key.offset = size;
5908         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5909         atomic_set(&cache->count, 1);
5910         spin_lock_init(&cache->lock);
5911         mutex_init(&cache->alloc_mutex);
5912         mutex_init(&cache->cache_mutex);
5913         INIT_LIST_HEAD(&cache->list);
5914
5915         btrfs_set_block_group_used(&cache->item, bytes_used);
5916         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5917         cache->flags = type;
5918         btrfs_set_block_group_flags(&cache->item, type);
5919
5920         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5921                                 &cache->space_info);
5922         BUG_ON(ret);
5923         down_write(&cache->space_info->groups_sem);
5924         list_add_tail(&cache->list, &cache->space_info->block_groups);
5925         up_write(&cache->space_info->groups_sem);
5926
5927         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5928         BUG_ON(ret);
5929
5930         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5931                                 sizeof(cache->item));
5932         BUG_ON(ret);
5933
5934         set_avail_alloc_bits(extent_root->fs_info, type);
5935
5936         return 0;
5937 }
5938
5939 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5940                              struct btrfs_root *root, u64 group_start)
5941 {
5942         struct btrfs_path *path;
5943         struct btrfs_block_group_cache *block_group;
5944         struct btrfs_key key;
5945         int ret;
5946
5947         root = root->fs_info->extent_root;
5948
5949         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5950         BUG_ON(!block_group);
5951         BUG_ON(!block_group->ro);
5952
5953         memcpy(&key, &block_group->key, sizeof(key));
5954
5955         path = btrfs_alloc_path();
5956         BUG_ON(!path);
5957
5958         spin_lock(&root->fs_info->block_group_cache_lock);
5959         rb_erase(&block_group->cache_node,
5960                  &root->fs_info->block_group_cache_tree);
5961         spin_unlock(&root->fs_info->block_group_cache_lock);
5962         btrfs_remove_free_space_cache(block_group);
5963         down_write(&block_group->space_info->groups_sem);
5964         list_del(&block_group->list);
5965         up_write(&block_group->space_info->groups_sem);
5966
5967         spin_lock(&block_group->space_info->lock);
5968         block_group->space_info->total_bytes -= block_group->key.offset;
5969         block_group->space_info->bytes_readonly -= block_group->key.offset;
5970         spin_unlock(&block_group->space_info->lock);
5971         block_group->space_info->full = 0;
5972
5973         put_block_group(block_group);
5974         put_block_group(block_group);
5975
5976         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5977         if (ret > 0)
5978                 ret = -EIO;
5979         if (ret < 0)
5980                 goto out;
5981
5982         ret = btrfs_del_item(trans, root, path);
5983 out:
5984         btrfs_free_path(path);
5985         return ret;
5986 }