SAFE public projects git trees. - safe/jmp/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 "hash.h"
  23 #include "crc32c.h"
  24 #include "ctree.h"
  25 #include "disk-io.h"
  26 #include "print-tree.h"
  27 #include "transaction.h"
  28 #include "volumes.h"
  29 #include "locking.h"
  30 #include "ref-cache.h"
  31
  32 #define PENDING_EXTENT_INSERT 0
  33 #define PENDING_EXTENT_DELETE 1
  34 #define PENDING_BACKREF_UPDATE 2
  35
  36 struct pending_extent_op {
  37         int type;
  38         u64 bytenr;
  39         u64 num_bytes;
  40         u64 parent;
  41         u64 orig_parent;
  42         u64 generation;
  43         u64 orig_generation;
  44         int level;
  45 };
  46
  47 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  48                                  btrfs_root *extent_root, int all);
  49 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  50                                btrfs_root *extent_root, int all);
  51 static struct btrfs_block_group_cache *
  52 __btrfs_find_block_group(struct btrfs_root *root,
  53                          struct btrfs_block_group_cache *hint,
  54                          u64 search_start, int data, int owner);
  55
  56 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  57 {
  58         return (cache->flags & bits) == bits;
  59 }
  60
  61 /*
  62  * this adds the block group to the fs_info rb tree for the block group
  63  * cache
  64  */
  65 int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  66                                 struct btrfs_block_group_cache *block_group)
  67 {
  68         struct rb_node **p;
  69         struct rb_node *parent = NULL;
  70         struct btrfs_block_group_cache *cache;
  71
  72         spin_lock(&info->block_group_cache_lock);
  73         p = &info->block_group_cache_tree.rb_node;
  74
  75         while (*p) {
  76                 parent = *p;
  77                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  78                                  cache_node);
  79                 if (block_group->key.objectid < cache->key.objectid) {
  80                         p = &(*p)->rb_left;
  81                 } else if (block_group->key.objectid > cache->key.objectid) {
  82                         p = &(*p)->rb_right;
  83                 } else {
  84                         spin_unlock(&info->block_group_cache_lock);
  85                         return -EEXIST;
  86                 }
  87         }
  88
  89         rb_link_node(&block_group->cache_node, parent, p);
  90         rb_insert_color(&block_group->cache_node,
  91                         &info->block_group_cache_tree);
  92         spin_unlock(&info->block_group_cache_lock);
  93
  94         return 0;
  95 }
  96
  97 /*
  98  * This will return the block group at or after bytenr if contains is 0, else
  99  * it will return the block group that contains the bytenr
 100  */
 101 static struct btrfs_block_group_cache *
 102 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 103                               int contains)
 104 {
 105         struct btrfs_block_group_cache *cache, *ret = NULL;
 106         struct rb_node *n;
 107         u64 end, start;
 108
 109         spin_lock(&info->block_group_cache_lock);
 110         n = info->block_group_cache_tree.rb_node;
 111
 112         while (n) {
 113                 cache = rb_entry(n, struct btrfs_block_group_cache,
 114                                  cache_node);
 115                 end = cache->key.objectid + cache->key.offset - 1;
 116                 start = cache->key.objectid;
 117
 118                 if (bytenr < start) {
 119                         if (!contains && (!ret || start < ret->key.objectid))
 120                                 ret = cache;
 121                         n = n->rb_left;
 122                 } else if (bytenr > start) {
 123                         if (contains && bytenr <= end) {
 124                                 ret = cache;
 125                                 break;
 126                         }
 127                         n = n->rb_right;
 128                 } else {
 129                         ret = cache;
 130                         break;
 131                 }
 132         }
 133         spin_unlock(&info->block_group_cache_lock);
 134
 135         return ret;
 136 }
 137
 138 /*
 139  * this is only called by cache_block_group, since we could have freed extents
 140  * we need to check the pinned_extents for any extents that can't be used yet
 141  * since their free space will be released as soon as the transaction commits.
 142  */
 143 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 144                               struct btrfs_fs_info *info, u64 start, u64 end)
 145 {
 146         u64 extent_start, extent_end, size;
 147         int ret;
 148
 149         mutex_lock(&info->pinned_mutex);
 150         while (start < end) {
 151                 ret = find_first_extent_bit(&info->pinned_extents, start,
 152                                             &extent_start, &extent_end,
 153                                             EXTENT_DIRTY);
 154                 if (ret)
 155                         break;
 156
 157                 if (extent_start == start) {
 158                         start = extent_end + 1;
 159                 } else if (extent_start > start && extent_start < end) {
 160                         size = extent_start - start;
 161                         ret = btrfs_add_free_space_lock(block_group, start,
 162                                                         size);
 163                         BUG_ON(ret);
 164                         start = extent_end + 1;
 165                 } else {
 166                         break;
 167                 }
 168         }
 169
 170         if (start < end) {
 171                 size = end - start;
 172                 ret = btrfs_add_free_space_lock(block_group, start, size);
 173                 BUG_ON(ret);
 174         }
 175         mutex_unlock(&info->pinned_mutex);
 176
 177         return 0;
 178 }
 179
 180 static int cache_block_group(struct btrfs_root *root,
 181                              struct btrfs_block_group_cache *block_group)
 182 {
 183         struct btrfs_path *path;
 184         int ret = 0;
 185         struct btrfs_key key;
 186         struct extent_buffer *leaf;
 187         int slot;
 188         u64 last = 0;
 189         u64 first_free;
 190         int found = 0;
 191
 192         if (!block_group)
 193                 return 0;
 194
 195         root = root->fs_info->extent_root;
 196
 197         if (block_group->cached)
 198                 return 0;
 199
 200         path = btrfs_alloc_path();
 201         if (!path)
 202                 return -ENOMEM;
 203
 204         path->reada = 2;
 205         /*
 206          * we get into deadlocks with paths held by callers of this function.
 207          * since the alloc_mutex is protecting things right now, just
 208          * skip the locking here
 209          */
 210         path->skip_locking = 1;
 211         first_free = max_t(u64, block_group->key.objectid,
 212                            BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 213         key.objectid = block_group->key.objectid;
 214         key.offset = 0;
 215         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 216         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 217         if (ret < 0)
 218                 goto err;
 219         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
 220         if (ret < 0)
 221                 goto err;
 222         if (ret == 0) {
 223                 leaf = path->nodes[0];
 224                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 225                 if (key.objectid + key.offset > first_free)
 226                         first_free = key.objectid + key.offset;
 227         }
 228         while(1) {
 229                 leaf = path->nodes[0];
 230                 slot = path->slots[0];
 231                 if (slot >= btrfs_header_nritems(leaf)) {
 232                         ret = btrfs_next_leaf(root, path);
 233                         if (ret < 0)
 234                                 goto err;
 235                         if (ret == 0)
 236                                 continue;
 237                         else
 238                                 break;
 239                 }
 240                 btrfs_item_key_to_cpu(leaf, &key, slot);
 241                 if (key.objectid < block_group->key.objectid)
 242                         goto next;
 243
 244                 if (key.objectid >= block_group->key.objectid +
 245                     block_group->key.offset)
 246                         break;
 247
 248                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 249                         if (!found) {
 250                                 last = first_free;
 251                                 found = 1;
 252                         }
 253
 254                         add_new_free_space(block_group, root->fs_info, last,
 255                                            key.objectid);
 256
 257                         last = key.objectid + key.offset;
 258                 }
 259 next:
 260                 path->slots[0]++;
 261         }
 262
 263         if (!found)
 264                 last = first_free;
 265
 266         add_new_free_space(block_group, root->fs_info, last,
 267                            block_group->key.objectid +
 268                            block_group->key.offset);
 269
 270         block_group->cached = 1;
 271         ret = 0;
 272 err:
 273         btrfs_free_path(path);
 274         return ret;
 275 }
 276
 277 /*
 278  * return the block group that starts at or after bytenr
 279  */
 280 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 281                                                        btrfs_fs_info *info,
 282                                                          u64 bytenr)
 283 {
 284         struct btrfs_block_group_cache *cache;
 285
 286         cache = block_group_cache_tree_search(info, bytenr, 0);
 287
 288         return cache;
 289 }
 290
 291 /*
 292  * return the block group that contains teh given bytenr
 293  */
 294 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 295                                                          btrfs_fs_info *info,
 296                                                          u64 bytenr)
 297 {
 298         struct btrfs_block_group_cache *cache;
 299
 300         cache = block_group_cache_tree_search(info, bytenr, 1);
 301
 302         return cache;
 303 }
 304
 305 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 306                                                   u64 flags)
 307 {
 308         struct list_head *head = &info->space_info;
 309         struct list_head *cur;
 310         struct btrfs_space_info *found;
 311         list_for_each(cur, head) {
 312                 found = list_entry(cur, struct btrfs_space_info, list);
 313                 if (found->flags == flags)
 314                         return found;
 315         }
 316         return NULL;
 317 }
 318
 319 static u64 div_factor(u64 num, int factor)
 320 {
 321         if (factor == 10)
 322                 return num;
 323         num *= factor;
 324         do_div(num, 10);
 325         return num;
 326 }
 327
 328 static struct btrfs_block_group_cache *
 329 __btrfs_find_block_group(struct btrfs_root *root,
 330                          struct btrfs_block_group_cache *hint,
 331                          u64 search_start, int data, int owner)
 332 {
 333         struct btrfs_block_group_cache *cache;
 334         struct btrfs_block_group_cache *found_group = NULL;
 335         struct btrfs_fs_info *info = root->fs_info;
 336         u64 used;
 337         u64 last = 0;
 338         u64 free_check;
 339         int full_search = 0;
 340         int factor = 10;
 341         int wrapped = 0;
 342
 343         if (data & BTRFS_BLOCK_GROUP_METADATA)
 344                 factor = 9;
 345
 346         if (search_start) {
 347                 struct btrfs_block_group_cache *shint;
 348                 shint = btrfs_lookup_first_block_group(info, search_start);
 349                 if (shint && block_group_bits(shint, data) && !shint->ro) {
 350                         spin_lock(&shint->lock);
 351                         used = btrfs_block_group_used(&shint->item);
 352                         if (used + shint->pinned + shint->reserved <
 353                             div_factor(shint->key.offset, factor)) {
 354                                 spin_unlock(&shint->lock);
 355                                 return shint;
 356                         }
 357                         spin_unlock(&shint->lock);
 358                 }
 359         }
 360         if (hint && !hint->ro && block_group_bits(hint, data)) {
 361                 spin_lock(&hint->lock);
 362                 used = btrfs_block_group_used(&hint->item);
 363                 if (used + hint->pinned + hint->reserved <
 364                     div_factor(hint->key.offset, factor)) {
 365                         spin_unlock(&hint->lock);
 366                         return hint;
 367                 }
 368                 spin_unlock(&hint->lock);
 369                 last = hint->key.objectid + hint->key.offset;
 370         } else {
 371                 if (hint)
 372                         last = max(hint->key.objectid, search_start);
 373                 else
 374                         last = search_start;
 375         }
 376 again:
 377         while (1) {
 378                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 379                 if (!cache)
 380                         break;
 381
 382                 spin_lock(&cache->lock);
 383                 last = cache->key.objectid + cache->key.offset;
 384                 used = btrfs_block_group_used(&cache->item);
 385
 386                 if (!cache->ro && block_group_bits(cache, data)) {
 387                         free_check = div_factor(cache->key.offset, factor);
 388                         if (used + cache->pinned + cache->reserved <
 389                             free_check) {
 390                                 found_group = cache;
 391                                 spin_unlock(&cache->lock);
 392                                 goto found;
 393                         }
 394                 }
 395                 spin_unlock(&cache->lock);
 396                 cond_resched();
 397         }
 398         if (!wrapped) {
 399                 last = search_start;
 400                 wrapped = 1;
 401                 goto again;
 402         }
 403         if (!full_search && factor < 10) {
 404                 last = search_start;
 405                 full_search = 1;
 406                 factor = 10;
 407                 goto again;
 408         }
 409 found:
 410         return found_group;
 411 }
 412
 413 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 414                                                  struct btrfs_block_group_cache
 415                                                  *hint, u64 search_start,
 416                                                  int data, int owner)
 417 {
 418
 419         struct btrfs_block_group_cache *ret;
 420         ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
 421         return ret;
 422 }
 423
 424 /* simple helper to search for an existing extent at a given offset */
 425 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 426 {
 427         int ret;
 428         struct btrfs_key key;
 429         struct btrfs_path *path;
 430
 431         path = btrfs_alloc_path();
 432         BUG_ON(!path);
 433         key.objectid = start;
 434         key.offset = len;
 435         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 436         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 437                                 0, 0);
 438         btrfs_free_path(path);
 439         return ret;
 440 }
 441
 442 /*
 443  * Back reference rules.  Back refs have three main goals:
 444  *
 445  * 1) differentiate between all holders of references to an extent so that
 446  *    when a reference is dropped we can make sure it was a valid reference
 447  *    before freeing the extent.
 448  *
 449  * 2) Provide enough information to quickly find the holders of an extent
 450  *    if we notice a given block is corrupted or bad.
 451  *
 452  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 453  *    maintenance.  This is actually the same as #2, but with a slightly
 454  *    different use case.
 455  *
 456  * File extents can be referenced by:
 457  *
 458  * - multiple snapshots, subvolumes, or different generations in one subvol
 459  * - different files inside a single subvolume
 460  * - different offsets inside a file (bookend extents in file.c)
 461  *
 462  * The extent ref structure has fields for:
 463  *
 464  * - Objectid of the subvolume root
 465  * - Generation number of the tree holding the reference
 466  * - objectid of the file holding the reference
 467  * - number of references holding by parent node (alway 1 for tree blocks)
 468  *
 469  * Btree leaf may hold multiple references to a file extent. In most cases,
 470  * these references are from same file and the corresponding offsets inside
 471  * the file are close together.
 472  *
 473  * When a file extent is allocated the fields are filled in:
 474  *     (root_key.objectid, trans->transid, inode objectid, 1)
 475  *
 476  * When a leaf is cow'd new references are added for every file extent found
 477  * in the leaf.  It looks similar to the create case, but trans->transid will
 478  * be different when the block is cow'd.
 479  *
 480  *     (root_key.objectid, trans->transid, inode objectid,
 481  *      number of references in the leaf)
 482  *
 483  * When a file extent is removed either during snapshot deletion or
 484  * file truncation, we find the corresponding back reference and check
 485  * the following fields:
 486  *
 487  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 488  *      inode objectid)
 489  *
 490  * Btree extents can be referenced by:
 491  *
 492  * - Different subvolumes
 493  * - Different generations of the same subvolume
 494  *
 495  * When a tree block is created, back references are inserted:
 496  *
 497  * (root->root_key.objectid, trans->transid, level, 1)
 498  *
 499  * When a tree block is cow'd, new back references are added for all the
 500  * blocks it points to. If the tree block isn't in reference counted root,
 501  * the old back references are removed. These new back references are of
 502  * the form (trans->transid will have increased since creation):
 503  *
 504  * (root->root_key.objectid, trans->transid, level, 1)
 505  *
 506  * When a backref is in deleting, the following fields are checked:
 507  *
 508  * if backref was for a tree root:
 509  *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
 510  * else
 511  *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
 512  *
 513  * Back Reference Key composing:
 514  *
 515  * The key objectid corresponds to the first byte in the extent, the key
 516  * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
 517  * byte of parent extent. If a extent is tree root, the key offset is set
 518  * to the key objectid.
 519  */
 520
 521 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 522                                           struct btrfs_root *root,
 523                                           struct btrfs_path *path,
 524                                           u64 bytenr, u64 parent,
 525                                           u64 ref_root, u64 ref_generation,
 526                                           u64 owner_objectid, int del)
 527 {
 528         struct btrfs_key key;
 529         struct btrfs_extent_ref *ref;
 530         struct extent_buffer *leaf;
 531         u64 ref_objectid;
 532         int ret;
 533
 534         key.objectid = bytenr;
 535         key.type = BTRFS_EXTENT_REF_KEY;
 536         key.offset = parent;
 537
 538         ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
 539         if (ret < 0)
 540                 goto out;
 541         if (ret > 0) {
 542                 ret = -ENOENT;
 543                 goto out;
 544         }
 545
 546         leaf = path->nodes[0];
 547         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 548         ref_objectid = btrfs_ref_objectid(leaf, ref);
 549         if (btrfs_ref_root(leaf, ref) != ref_root ||
 550             btrfs_ref_generation(leaf, ref) != ref_generation ||
 551             (ref_objectid != owner_objectid &&
 552              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 553                 ret = -EIO;
 554                 WARN_ON(1);
 555                 goto out;
 556         }
 557         ret = 0;
 558 out:
 559         return ret;
 560 }
 561
 562 static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
 563                                           struct btrfs_root *root,
 564                                           struct btrfs_path *path,
 565                                           u64 bytenr, u64 parent,
 566                                           u64 ref_root, u64 ref_generation,
 567                                           u64 owner_objectid)
 568 {
 569         struct btrfs_key key;
 570         struct extent_buffer *leaf;
 571         struct btrfs_extent_ref *ref;
 572         u32 num_refs;
 573         int ret;
 574
 575         key.objectid = bytenr;
 576         key.type = BTRFS_EXTENT_REF_KEY;
 577         key.offset = parent;
 578
 579         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
 580         if (ret == 0) {
 581                 leaf = path->nodes[0];
 582                 ref = btrfs_item_ptr(leaf, path->slots[0],
 583                                      struct btrfs_extent_ref);
 584                 btrfs_set_ref_root(leaf, ref, ref_root);
 585                 btrfs_set_ref_generation(leaf, ref, ref_generation);
 586                 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
 587                 btrfs_set_ref_num_refs(leaf, ref, 1);
 588         } else if (ret == -EEXIST) {
 589                 u64 existing_owner;
 590                 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
 591                 leaf = path->nodes[0];
 592                 ref = btrfs_item_ptr(leaf, path->slots[0],
 593                                      struct btrfs_extent_ref);
 594                 if (btrfs_ref_root(leaf, ref) != ref_root ||
 595                     btrfs_ref_generation(leaf, ref) != ref_generation) {
 596                         ret = -EIO;
 597                         WARN_ON(1);
 598                         goto out;
 599                 }
 600
 601                 num_refs = btrfs_ref_num_refs(leaf, ref);
 602                 BUG_ON(num_refs == 0);
 603                 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
 604
 605                 existing_owner = btrfs_ref_objectid(leaf, ref);
 606                 if (existing_owner != owner_objectid &&
 607                     existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
 608                         btrfs_set_ref_objectid(leaf, ref,
 609                                         BTRFS_MULTIPLE_OBJECTIDS);
 610                 }
 611                 ret = 0;
 612         } else {
 613                 goto out;
 614         }
 615         btrfs_mark_buffer_dirty(path->nodes[0]);
 616 out:
 617         btrfs_release_path(root, path);
 618         return ret;
 619 }
 620
 621 static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
 622                                           struct btrfs_root *root,
 623                                           struct btrfs_path *path)
 624 {
 625         struct extent_buffer *leaf;
 626         struct btrfs_extent_ref *ref;
 627         u32 num_refs;
 628         int ret = 0;
 629
 630         leaf = path->nodes[0];
 631         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 632         num_refs = btrfs_ref_num_refs(leaf, ref);
 633         BUG_ON(num_refs == 0);
 634         num_refs -= 1;
 635         if (num_refs == 0) {
 636                 ret = btrfs_del_item(trans, root, path);
 637         } else {
 638                 btrfs_set_ref_num_refs(leaf, ref, num_refs);
 639                 btrfs_mark_buffer_dirty(leaf);
 640         }
 641         btrfs_release_path(root, path);
 642         return ret;
 643 }
 644
 645 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
 646                                      struct btrfs_root *root, u64 bytenr,
 647                                      u64 orig_parent, u64 parent,
 648                                      u64 orig_root, u64 ref_root,
 649                                      u64 orig_generation, u64 ref_generation,
 650                                      u64 owner_objectid)
 651 {
 652         int ret;
 653         struct btrfs_root *extent_root = root->fs_info->extent_root;
 654         struct btrfs_path *path;
 655
 656         if (root == root->fs_info->extent_root) {
 657                 struct pending_extent_op *extent_op;
 658                 u64 num_bytes;
 659
 660                 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
 661                 num_bytes = btrfs_level_size(root, (int)owner_objectid);
 662                 mutex_lock(&root->fs_info->extent_ins_mutex);
 663                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
 664                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
 665                         u64 priv;
 666                         ret = get_state_private(&root->fs_info->extent_ins,
 667                                                 bytenr, &priv);
 668                         BUG_ON(ret);
 669                         extent_op = (struct pending_extent_op *)
 670                                                         (unsigned long)priv;
 671                         BUG_ON(extent_op->parent != orig_parent);
 672                         BUG_ON(extent_op->generation != orig_generation);
 673
 674                         extent_op->parent = parent;
 675                         extent_op->generation = ref_generation;
 676                 } else {
 677                         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
 678                         BUG_ON(!extent_op);
 679
 680                         extent_op->type = PENDING_BACKREF_UPDATE;
 681                         extent_op->bytenr = bytenr;
 682                         extent_op->num_bytes = num_bytes;
 683                         extent_op->parent = parent;
 684                         extent_op->orig_parent = orig_parent;
 685                         extent_op->generation = ref_generation;
 686                         extent_op->orig_generation = orig_generation;
 687                         extent_op->level = (int)owner_objectid;
 688
 689                         set_extent_bits(&root->fs_info->extent_ins,
 690                                         bytenr, bytenr + num_bytes - 1,
 691                                         EXTENT_WRITEBACK, GFP_NOFS);
 692                         set_state_private(&root->fs_info->extent_ins,
 693                                           bytenr, (unsigned long)extent_op);
 694                 }
 695                 mutex_unlock(&root->fs_info->extent_ins_mutex);
 696                 return 0;
 697         }
 698
 699         path = btrfs_alloc_path();
 700         if (!path)
 701                 return -ENOMEM;
 702         ret = lookup_extent_backref(trans, extent_root, path,
 703                                     bytenr, orig_parent, orig_root,
 704                                     orig_generation, owner_objectid, 1);
 705         if (ret)
 706                 goto out;
 707         ret = remove_extent_backref(trans, extent_root, path);
 708         if (ret)
 709                 goto out;
 710         ret = insert_extent_backref(trans, extent_root, path, bytenr,
 711                                     parent, ref_root, ref_generation,
 712                                     owner_objectid);
 713         BUG_ON(ret);
 714         finish_current_insert(trans, extent_root, 0);
 715         del_pending_extents(trans, extent_root, 0);
 716 out:
 717         btrfs_free_path(path);
 718         return ret;
 719 }
 720
 721 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
 722                             struct btrfs_root *root, u64 bytenr,
 723                             u64 orig_parent, u64 parent,
 724                             u64 ref_root, u64 ref_generation,
 725                             u64 owner_objectid)
 726 {
 727         int ret;
 728         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
 729             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
 730                 return 0;
 731         ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
 732                                         parent, ref_root, ref_root,
 733                                         ref_generation, ref_generation,
 734                                         owner_objectid);
 735         return ret;
 736 }
 737
 738 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 739                                   struct btrfs_root *root, u64 bytenr,
 740                                   u64 orig_parent, u64 parent,
 741                                   u64 orig_root, u64 ref_root,
 742                                   u64 orig_generation, u64 ref_generation,
 743                                   u64 owner_objectid)
 744 {
 745         struct btrfs_path *path;
 746         int ret;
 747         struct btrfs_key key;
 748         struct extent_buffer *l;
 749         struct btrfs_extent_item *item;
 750         u32 refs;
 751
 752         path = btrfs_alloc_path();
 753         if (!path)
 754                 return -ENOMEM;
 755
 756         path->reada = 1;
 757         key.objectid = bytenr;
 758         key.type = BTRFS_EXTENT_ITEM_KEY;
 759         key.offset = (u64)-1;
 760
 761         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 762                                 0, 1);
 763         if (ret < 0)
 764                 return ret;
 765         BUG_ON(ret == 0 || path->slots[0] == 0);
 766
 767         path->slots[0]--;
 768         l = path->nodes[0];
 769
 770         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 771         if (key.objectid != bytenr) {
 772                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
 773                 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
 774                 BUG();
 775         }
 776         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
 777
 778         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 779         refs = btrfs_extent_refs(l, item);
 780         btrfs_set_extent_refs(l, item, refs + 1);
 781         btrfs_mark_buffer_dirty(path->nodes[0]);
 782
 783         btrfs_release_path(root->fs_info->extent_root, path);
 784
 785         path->reada = 1;
 786         ret = insert_extent_backref(trans, root->fs_info->extent_root,
 787                                     path, bytenr, parent,
 788                                     ref_root, ref_generation,
 789                                     owner_objectid);
 790         BUG_ON(ret);
 791         finish_current_insert(trans, root->fs_info->extent_root, 0);
 792         del_pending_extents(trans, root->fs_info->extent_root, 0);
 793
 794         btrfs_free_path(path);
 795         return 0;
 796 }
 797
 798 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 799                          struct btrfs_root *root,
 800                          u64 bytenr, u64 num_bytes, u64 parent,
 801                          u64 ref_root, u64 ref_generation,
 802                          u64 owner_objectid)
 803 {
 804         int ret;
 805         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
 806             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
 807                 return 0;
 808         ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
 809                                      0, ref_root, 0, ref_generation,
 810                                      owner_objectid);
 811         return ret;
 812 }
 813
 814 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 815                          struct btrfs_root *root)
 816 {
 817         finish_current_insert(trans, root->fs_info->extent_root, 1);
 818         del_pending_extents(trans, root->fs_info->extent_root, 1);
 819         return 0;
 820 }
 821
 822 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
 823                             struct btrfs_root *root, u64 bytenr,
 824                             u64 num_bytes, u32 *refs)
 825 {
 826         struct btrfs_path *path;
 827         int ret;
 828         struct btrfs_key key;
 829         struct extent_buffer *l;
 830         struct btrfs_extent_item *item;
 831
 832         WARN_ON(num_bytes < root->sectorsize);
 833         path = btrfs_alloc_path();
 834         path->reada = 1;
 835         key.objectid = bytenr;
 836         key.offset = num_bytes;
 837         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 838         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 839                                 0, 0);
 840         if (ret < 0)
 841                 goto out;
 842         if (ret != 0) {
 843                 btrfs_print_leaf(root, path->nodes[0]);
 844                 printk("failed to find block number %Lu\n", bytenr);
 845                 BUG();
 846         }
 847         l = path->nodes[0];
 848         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 849         *refs = btrfs_extent_refs(l, item);
 850 out:
 851         btrfs_free_path(path);
 852         return 0;
 853 }
 854
 855 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
 856                           struct btrfs_root *root, u64 bytenr)
 857 {
 858         struct btrfs_root *extent_root = root->fs_info->extent_root;
 859         struct btrfs_path *path;
 860         struct extent_buffer *leaf;
 861         struct btrfs_extent_ref *ref_item;
 862         struct btrfs_key key;
 863         struct btrfs_key found_key;
 864         u64 ref_root;
 865         u64 last_snapshot;
 866         u32 nritems;
 867         int ret;
 868
 869         key.objectid = bytenr;
 870         key.offset = (u64)-1;
 871         key.type = BTRFS_EXTENT_ITEM_KEY;
 872
 873         path = btrfs_alloc_path();
 874         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 875         if (ret < 0)
 876                 goto out;
 877         BUG_ON(ret == 0);
 878
 879         ret = -ENOENT;
 880         if (path->slots[0] == 0)
 881                 goto out;
 882
 883         path->slots[0]--;
 884         leaf = path->nodes[0];
 885         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 886
 887         if (found_key.objectid != bytenr ||
 888             found_key.type != BTRFS_EXTENT_ITEM_KEY)
 889                 goto out;
 890
 891         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
 892         while (1) {
 893                 leaf = path->nodes[0];
 894                 nritems = btrfs_header_nritems(leaf);
 895                 if (path->slots[0] >= nritems) {
 896                         ret = btrfs_next_leaf(extent_root, path);
 897                         if (ret < 0)
 898                                 goto out;
 899                         if (ret == 0)
 900                                 continue;
 901                         break;
 902                 }
 903                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 904                 if (found_key.objectid != bytenr)
 905                         break;
 906
 907                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 908                         path->slots[0]++;
 909                         continue;
 910                 }
 911
 912                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
 913                                           struct btrfs_extent_ref);
 914                 ref_root = btrfs_ref_root(leaf, ref_item);
 915                 if (ref_root != root->root_key.objectid &&
 916                     ref_root != BTRFS_TREE_LOG_OBJECTID) {
 917                         ret = 1;
 918                         goto out;
 919                 }
 920                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
 921                         ret = 1;
 922                         goto out;
 923                 }
 924
 925                 path->slots[0]++;
 926         }
 927         ret = 0;
 928 out:
 929         btrfs_free_path(path);
 930         return ret;
 931 }
 932
 933 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 934                     struct extent_buffer *buf, u32 nr_extents)
 935 {
 936         struct btrfs_key key;
 937         struct btrfs_file_extent_item *fi;
 938         u64 root_gen;
 939         u32 nritems;
 940         int i;
 941         int level;
 942         int ret = 0;
 943         int shared = 0;
 944
 945         if (!root->ref_cows)
 946                 return 0;
 947
 948         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
 949                 shared = 0;
 950                 root_gen = root->root_key.offset;
 951         } else {
 952                 shared = 1;
 953                 root_gen = trans->transid - 1;
 954         }
 955
 956         level = btrfs_header_level(buf);
 957         nritems = btrfs_header_nritems(buf);
 958
 959         if (level == 0) {
 960                 struct btrfs_leaf_ref *ref;
 961                 struct btrfs_extent_info *info;
 962
 963                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
 964                 if (!ref) {
 965                         ret = -ENOMEM;
 966                         goto out;
 967                 }
 968
 969                 ref->root_gen = root_gen;
 970                 ref->bytenr = buf->start;
 971                 ref->owner = btrfs_header_owner(buf);
 972                 ref->generation = btrfs_header_generation(buf);
 973                 ref->nritems = nr_extents;
 974                 info = ref->extents;
 975
 976                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
 977                         u64 disk_bytenr;
 978                         btrfs_item_key_to_cpu(buf, &key, i);
 979                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 980                                 continue;
 981                         fi = btrfs_item_ptr(buf, i,
 982                                             struct btrfs_file_extent_item);
 983                         if (btrfs_file_extent_type(buf, fi) ==
 984                             BTRFS_FILE_EXTENT_INLINE)
 985                                 continue;
 986                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 987                         if (disk_bytenr == 0)
 988                                 continue;
 989
 990                         info->bytenr = disk_bytenr;
 991                         info->num_bytes =
 992                                 btrfs_file_extent_disk_num_bytes(buf, fi);
 993                         info->objectid = key.objectid;
 994                         info->offset = key.offset;
 995                         info++;
 996                 }
 997
 998                 ret = btrfs_add_leaf_ref(root, ref, shared);
 999                 if (ret == -EEXIST && shared) {
1000                         struct btrfs_leaf_ref *old;
1001                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1002                         BUG_ON(!old);
1003                         btrfs_remove_leaf_ref(root, old);
1004                         btrfs_free_leaf_ref(root, old);
1005                         ret = btrfs_add_leaf_ref(root, ref, shared);
1006                 }
1007                 WARN_ON(ret);
1008                 btrfs_free_leaf_ref(root, ref);
1009         }
1010 out:
1011         return ret;
1012 }
1013
1014 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1015                   struct extent_buffer *orig_buf, struct extent_buffer *buf,
1016                   u32 *nr_extents)
1017 {
1018         u64 bytenr;
1019         u64 ref_root;
1020         u64 orig_root;
1021         u64 ref_generation;
1022         u64 orig_generation;
1023         u32 nritems;
1024         u32 nr_file_extents = 0;
1025         struct btrfs_key key;
1026         struct btrfs_file_extent_item *fi;
1027         int i;
1028         int level;
1029         int ret = 0;
1030         int faili = 0;
1031         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1032                             u64, u64, u64, u64, u64, u64, u64, u64);
1033
1034         ref_root = btrfs_header_owner(buf);
1035         ref_generation = btrfs_header_generation(buf);
1036         orig_root = btrfs_header_owner(orig_buf);
1037         orig_generation = btrfs_header_generation(orig_buf);
1038
1039         nritems = btrfs_header_nritems(buf);
1040         level = btrfs_header_level(buf);
1041
1042         if (root->ref_cows) {
1043                 process_func = __btrfs_inc_extent_ref;
1044         } else {
1045                 if (level == 0 &&
1046                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1047                         goto out;
1048                 if (level != 0 &&
1049                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1050                         goto out;
1051                 process_func = __btrfs_update_extent_ref;
1052         }
1053
1054         for (i = 0; i < nritems; i++) {
1055                 cond_resched();
1056                 if (level == 0) {
1057                         btrfs_item_key_to_cpu(buf, &key, i);
1058                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1059                                 continue;
1060                         fi = btrfs_item_ptr(buf, i,
1061                                             struct btrfs_file_extent_item);
1062                         if (btrfs_file_extent_type(buf, fi) ==
1063                             BTRFS_FILE_EXTENT_INLINE)
1064                                 continue;
1065                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1066                         if (bytenr == 0)
1067                                 continue;
1068
1069                         nr_file_extents++;
1070
1071                         ret = process_func(trans, root, bytenr,
1072                                            orig_buf->start, buf->start,
1073                                            orig_root, ref_root,
1074                                            orig_generation, ref_generation,
1075                                            key.objectid);
1076
1077                         if (ret) {
1078                                 faili = i;
1079                                 WARN_ON(1);
1080                                 goto fail;
1081                         }
1082                 } else {
1083                         bytenr = btrfs_node_blockptr(buf, i);
1084                         ret = process_func(trans, root, bytenr,
1085                                            orig_buf->start, buf->start,
1086                                            orig_root, ref_root,
1087                                            orig_generation, ref_generation,
1088                                            level - 1);
1089                         if (ret) {
1090                                 faili = i;
1091                                 WARN_ON(1);
1092                                 goto fail;
1093                         }
1094                 }
1095         }
1096 out:
1097         if (nr_extents) {
1098                 if (level == 0)
1099                         *nr_extents = nr_file_extents;
1100                 else
1101                         *nr_extents = nritems;
1102         }
1103         return 0;
1104 fail:
1105         WARN_ON(1);
1106         return ret;
1107 }
1108
1109 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1110                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1111                      struct extent_buffer *buf, int start_slot, int nr)
1112
1113 {
1114         u64 bytenr;
1115         u64 ref_root;
1116         u64 orig_root;
1117         u64 ref_generation;
1118         u64 orig_generation;
1119         struct btrfs_key key;
1120         struct btrfs_file_extent_item *fi;
1121         int i;
1122         int ret;
1123         int slot;
1124         int level;
1125
1126         BUG_ON(start_slot < 0);
1127         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1128
1129         ref_root = btrfs_header_owner(buf);
1130         ref_generation = btrfs_header_generation(buf);
1131         orig_root = btrfs_header_owner(orig_buf);
1132         orig_generation = btrfs_header_generation(orig_buf);
1133         level = btrfs_header_level(buf);
1134
1135         if (!root->ref_cows) {
1136                 if (level == 0 &&
1137                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1138                         return 0;
1139                 if (level != 0 &&
1140                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1141                         return 0;
1142         }
1143
1144         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1145                 cond_resched();
1146                 if (level == 0) {
1147                         btrfs_item_key_to_cpu(buf, &key, slot);
1148                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1149                                 continue;
1150                         fi = btrfs_item_ptr(buf, slot,
1151                                             struct btrfs_file_extent_item);
1152                         if (btrfs_file_extent_type(buf, fi) ==
1153                             BTRFS_FILE_EXTENT_INLINE)
1154                                 continue;
1155                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1156                         if (bytenr == 0)
1157                                 continue;
1158                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1159                                             orig_buf->start, buf->start,
1160                                             orig_root, ref_root,
1161                                             orig_generation, ref_generation,
1162                                             key.objectid);
1163                         if (ret)
1164                                 goto fail;
1165                 } else {
1166                         bytenr = btrfs_node_blockptr(buf, slot);
1167                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1168                                             orig_buf->start, buf->start,
1169                                             orig_root, ref_root,
1170                                             orig_generation, ref_generation,
1171                                             level - 1);
1172                         if (ret)
1173                                 goto fail;
1174                 }
1175         }
1176         return 0;
1177 fail:
1178         WARN_ON(1);
1179         return -1;
1180 }
1181
1182 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1183                                  struct btrfs_root *root,
1184                                  struct btrfs_path *path,
1185                                  struct btrfs_block_group_cache *cache)
1186 {
1187         int ret;
1188         int pending_ret;
1189         struct btrfs_root *extent_root = root->fs_info->extent_root;
1190         unsigned long bi;
1191         struct extent_buffer *leaf;
1192
1193         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1194         if (ret < 0)
1195                 goto fail;
1196         BUG_ON(ret);
1197
1198         leaf = path->nodes[0];
1199         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1200         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1201         btrfs_mark_buffer_dirty(leaf);
1202         btrfs_release_path(extent_root, path);
1203 fail:
1204         finish_current_insert(trans, extent_root, 0);
1205         pending_ret = del_pending_extents(trans, extent_root, 0);
1206         if (ret)
1207                 return ret;
1208         if (pending_ret)
1209                 return pending_ret;
1210         return 0;
1211
1212 }
1213
1214 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1215                                    struct btrfs_root *root)
1216 {
1217         struct btrfs_block_group_cache *cache, *entry;
1218         struct rb_node *n;
1219         int err = 0;
1220         int werr = 0;
1221         struct btrfs_path *path;
1222         u64 last = 0;
1223
1224         path = btrfs_alloc_path();
1225         if (!path)
1226                 return -ENOMEM;
1227
1228         while(1) {
1229                 cache = NULL;
1230                 spin_lock(&root->fs_info->block_group_cache_lock);
1231                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1232                      n; n = rb_next(n)) {
1233                         entry = rb_entry(n, struct btrfs_block_group_cache,
1234                                          cache_node);
1235                         if (entry->dirty) {
1236                                 cache = entry;
1237                                 break;
1238                         }
1239                 }
1240                 spin_unlock(&root->fs_info->block_group_cache_lock);
1241
1242                 if (!cache)
1243                         break;
1244
1245                 cache->dirty = 0;
1246                 last += cache->key.offset;
1247
1248                 err = write_one_cache_group(trans, root,
1249                                             path, cache);
1250                 /*
1251                  * if we fail to write the cache group, we want
1252                  * to keep it marked dirty in hopes that a later
1253                  * write will work
1254                  */
1255                 if (err) {
1256                         werr = err;
1257                         continue;
1258                 }
1259         }
1260         btrfs_free_path(path);
1261         return werr;
1262 }
1263
1264 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1265                              u64 total_bytes, u64 bytes_used,
1266                              struct btrfs_space_info **space_info)
1267 {
1268         struct btrfs_space_info *found;
1269
1270         found = __find_space_info(info, flags);
1271         if (found) {
1272                 spin_lock(&found->lock);
1273                 found->total_bytes += total_bytes;
1274                 found->bytes_used += bytes_used;
1275                 found->full = 0;
1276                 spin_unlock(&found->lock);
1277                 *space_info = found;
1278                 return 0;
1279         }
1280         found = kmalloc(sizeof(*found), GFP_NOFS);
1281         if (!found)
1282                 return -ENOMEM;
1283
1284         list_add(&found->list, &info->space_info);
1285         INIT_LIST_HEAD(&found->block_groups);
1286         init_rwsem(&found->groups_sem);
1287         spin_lock_init(&found->lock);
1288         found->flags = flags;
1289         found->total_bytes = total_bytes;
1290         found->bytes_used = bytes_used;
1291         found->bytes_pinned = 0;
1292         found->bytes_reserved = 0;
1293         found->full = 0;
1294         found->force_alloc = 0;
1295         *space_info = found;
1296         return 0;
1297 }
1298
1299 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1300 {
1301         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1302                                    BTRFS_BLOCK_GROUP_RAID1 |
1303                                    BTRFS_BLOCK_GROUP_RAID10 |
1304                                    BTRFS_BLOCK_GROUP_DUP);
1305         if (extra_flags) {
1306                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1307                         fs_info->avail_data_alloc_bits |= extra_flags;
1308                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1309                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1310                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1311                         fs_info->avail_system_alloc_bits |= extra_flags;
1312         }
1313 }
1314
1315 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1316 {
1317         u64 num_devices = root->fs_info->fs_devices->num_devices;
1318
1319         if (num_devices == 1)
1320                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1321         if (num_devices < 4)
1322                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1323
1324         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1325             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1326                       BTRFS_BLOCK_GROUP_RAID10))) {
1327                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1328         }
1329
1330         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1331             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1332                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1333         }
1334
1335         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1336             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1337              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1338              (flags & BTRFS_BLOCK_GROUP_DUP)))
1339                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1340         return flags;
1341 }
1342
1343 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1344                           struct btrfs_root *extent_root, u64 alloc_bytes,
1345                           u64 flags, int force)
1346 {
1347         struct btrfs_space_info *space_info;
1348         u64 thresh;
1349         u64 start;
1350         u64 num_bytes;
1351         int ret = 0, waited = 0;
1352
1353         flags = reduce_alloc_profile(extent_root, flags);
1354
1355         space_info = __find_space_info(extent_root->fs_info, flags);
1356         if (!space_info) {
1357                 ret = update_space_info(extent_root->fs_info, flags,
1358                                         0, 0, &space_info);
1359                 BUG_ON(ret);
1360         }
1361         BUG_ON(!space_info);
1362
1363         spin_lock(&space_info->lock);
1364         if (space_info->force_alloc) {
1365                 force = 1;
1366                 space_info->force_alloc = 0;
1367         }
1368         if (space_info->full) {
1369                 spin_unlock(&space_info->lock);
1370                 goto out;
1371         }
1372
1373         thresh = div_factor(space_info->total_bytes, 6);
1374         if (!force &&
1375            (space_info->bytes_used + space_info->bytes_pinned +
1376             space_info->bytes_reserved + alloc_bytes) < thresh) {
1377                 spin_unlock(&space_info->lock);
1378                 goto out;
1379         }
1380
1381         spin_unlock(&space_info->lock);
1382
1383         ret = mutex_trylock(&extent_root->fs_info->chunk_mutex);
1384         if (!ret && !force) {
1385                 goto out;
1386         } else if (!ret) {
1387                 mutex_lock(&extent_root->fs_info->chunk_mutex);
1388                 waited = 1;
1389         }
1390
1391         if (waited) {
1392                 spin_lock(&space_info->lock);
1393                 if (space_info->full) {
1394                         spin_unlock(&space_info->lock);
1395                         goto out_unlock;
1396                 }
1397                 spin_unlock(&space_info->lock);
1398         }
1399
1400         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1401         if (ret) {
1402 printk("space info full %Lu\n", flags);
1403                 space_info->full = 1;
1404                 goto out_unlock;
1405         }
1406
1407         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1408                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1409         BUG_ON(ret);
1410 out_unlock:
1411         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1412 out:
1413         return ret;
1414 }
1415
1416 static int update_block_group(struct btrfs_trans_handle *trans,
1417                               struct btrfs_root *root,
1418                               u64 bytenr, u64 num_bytes, int alloc,
1419                               int mark_free)
1420 {
1421         struct btrfs_block_group_cache *cache;
1422         struct btrfs_fs_info *info = root->fs_info;
1423         u64 total = num_bytes;
1424         u64 old_val;
1425         u64 byte_in_group;
1426
1427         while(total) {
1428                 cache = btrfs_lookup_block_group(info, bytenr);
1429                 if (!cache) {
1430                         return -1;
1431                 }
1432                 byte_in_group = bytenr - cache->key.objectid;
1433                 WARN_ON(byte_in_group > cache->key.offset);
1434
1435                 spin_lock(&cache->space_info->lock);
1436                 spin_lock(&cache->lock);
1437                 cache->dirty = 1;
1438                 old_val = btrfs_block_group_used(&cache->item);
1439                 num_bytes = min(total, cache->key.offset - byte_in_group);
1440                 if (alloc) {
1441                         old_val += num_bytes;
1442                         cache->space_info->bytes_used += num_bytes;
1443                         btrfs_set_block_group_used(&cache->item, old_val);
1444                         spin_unlock(&cache->lock);
1445                         spin_unlock(&cache->space_info->lock);
1446                 } else {
1447                         old_val -= num_bytes;
1448                         cache->space_info->bytes_used -= num_bytes;
1449                         btrfs_set_block_group_used(&cache->item, old_val);
1450                         spin_unlock(&cache->lock);
1451                         spin_unlock(&cache->space_info->lock);
1452                         if (mark_free) {
1453                                 int ret;
1454                                 ret = btrfs_add_free_space(cache, bytenr,
1455                                                            num_bytes);
1456                                 if (ret)
1457                                         return -1;
1458                         }
1459                 }
1460                 total -= num_bytes;
1461                 bytenr += num_bytes;
1462         }
1463         return 0;
1464 }
1465
1466 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1467 {
1468         struct btrfs_block_group_cache *cache;
1469
1470         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1471         if (!cache)
1472                 return 0;
1473
1474         return cache->key.objectid;
1475 }
1476
1477 int btrfs_update_pinned_extents(struct btrfs_root *root,
1478                                 u64 bytenr, u64 num, int pin)
1479 {
1480         u64 len;
1481         struct btrfs_block_group_cache *cache;
1482         struct btrfs_fs_info *fs_info = root->fs_info;
1483
1484         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
1485         if (pin) {
1486                 set_extent_dirty(&fs_info->pinned_extents,
1487                                 bytenr, bytenr + num - 1, GFP_NOFS);
1488         } else {
1489                 clear_extent_dirty(&fs_info->pinned_extents,
1490                                 bytenr, bytenr + num - 1, GFP_NOFS);
1491         }
1492         while (num > 0) {
1493                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1494                 BUG_ON(!cache);
1495                 len = min(num, cache->key.offset -
1496                           (bytenr - cache->key.objectid));
1497                 if (pin) {
1498                         spin_lock(&cache->space_info->lock);
1499                         spin_lock(&cache->lock);
1500                         cache->pinned += len;
1501                         cache->space_info->bytes_pinned += len;
1502                         spin_unlock(&cache->lock);
1503                         spin_unlock(&cache->space_info->lock);
1504                         fs_info->total_pinned += len;
1505                 } else {
1506                         spin_lock(&cache->space_info->lock);
1507                         spin_lock(&cache->lock);
1508                         cache->pinned -= len;
1509                         cache->space_info->bytes_pinned -= len;
1510                         spin_unlock(&cache->lock);
1511                         spin_unlock(&cache->space_info->lock);
1512                         fs_info->total_pinned -= len;
1513                 }
1514                 bytenr += len;
1515                 num -= len;
1516         }
1517         return 0;
1518 }
1519
1520 static int update_reserved_extents(struct btrfs_root *root,
1521                                    u64 bytenr, u64 num, int reserve)
1522 {
1523         u64 len;
1524         struct btrfs_block_group_cache *cache;
1525         struct btrfs_fs_info *fs_info = root->fs_info;
1526
1527         while (num > 0) {
1528                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1529                 BUG_ON(!cache);
1530                 len = min(num, cache->key.offset -
1531                           (bytenr - cache->key.objectid));
1532
1533                 spin_lock(&cache->space_info->lock);
1534                 spin_lock(&cache->lock);
1535                 if (reserve) {
1536                         cache->reserved += len;
1537                         cache->space_info->bytes_reserved += len;
1538                 } else {
1539                         cache->reserved -= len;
1540                         cache->space_info->bytes_reserved -= len;
1541                 }
1542                 spin_unlock(&cache->lock);
1543                 spin_unlock(&cache->space_info->lock);
1544                 bytenr += len;
1545                 num -= len;
1546         }
1547         return 0;
1548 }
1549
1550 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1551 {
1552         u64 last = 0;
1553         u64 start;
1554         u64 end;
1555         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1556         int ret;
1557
1558         mutex_lock(&root->fs_info->pinned_mutex);
1559         while(1) {
1560                 ret = find_first_extent_bit(pinned_extents, last,
1561                                             &start, &end, EXTENT_DIRTY);
1562                 if (ret)
1563                         break;
1564                 set_extent_dirty(copy, start, end, GFP_NOFS);
1565                 last = end + 1;
1566         }
1567         mutex_unlock(&root->fs_info->pinned_mutex);
1568         return 0;
1569 }
1570
1571 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1572                                struct btrfs_root *root,
1573                                struct extent_io_tree *unpin)
1574 {
1575         u64 start;
1576         u64 end;
1577         int ret;
1578         struct btrfs_block_group_cache *cache;
1579
1580         mutex_lock(&root->fs_info->pinned_mutex);
1581         while(1) {
1582                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1583                                             EXTENT_DIRTY);
1584                 if (ret)
1585                         break;
1586                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1587                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1588                 cache = btrfs_lookup_block_group(root->fs_info, start);
1589                 if (cache->cached)
1590                         btrfs_add_free_space(cache, start, end - start + 1);
1591                 if (need_resched()) {
1592                         mutex_unlock(&root->fs_info->pinned_mutex);
1593                         cond_resched();
1594                         mutex_lock(&root->fs_info->pinned_mutex);
1595                 }
1596         }
1597         mutex_unlock(&root->fs_info->pinned_mutex);
1598         return 0;
1599 }
1600
1601 static int finish_current_insert(struct btrfs_trans_handle *trans,
1602                                  struct btrfs_root *extent_root, int all)
1603 {
1604         u64 start;
1605         u64 end;
1606         u64 priv;
1607         u64 search = 0;
1608         struct btrfs_fs_info *info = extent_root->fs_info;
1609         struct btrfs_path *path;
1610         struct btrfs_extent_ref *ref;
1611         struct pending_extent_op *extent_op;
1612         struct btrfs_key key;
1613         struct btrfs_extent_item extent_item;
1614         int ret;
1615         int err = 0;
1616
1617         btrfs_set_stack_extent_refs(&extent_item, 1);
1618         path = btrfs_alloc_path();
1619
1620         while(1) {
1621                 mutex_lock(&info->extent_ins_mutex);
1622                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
1623                                             &end, EXTENT_WRITEBACK);
1624                 if (ret) {
1625                         mutex_unlock(&info->extent_ins_mutex);
1626                         if (search && all) {
1627                                 search = 0;
1628                                 continue;
1629                         }
1630                         break;
1631                 }
1632
1633                 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
1634                 if (!ret) {
1635                         search = end + 1;
1636                         mutex_unlock(&info->extent_ins_mutex);
1637                         cond_resched();
1638                         continue;
1639                 }
1640                 BUG_ON(ret < 0);
1641
1642                 ret = get_state_private(&info->extent_ins, start, &priv);
1643                 BUG_ON(ret);
1644                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1645
1646                 mutex_unlock(&info->extent_ins_mutex);
1647
1648                 if (extent_op->type == PENDING_EXTENT_INSERT) {
1649                         key.objectid = start;
1650                         key.offset = end + 1 - start;
1651                         key.type = BTRFS_EXTENT_ITEM_KEY;
1652                         err = btrfs_insert_item(trans, extent_root, &key,
1653                                         &extent_item, sizeof(extent_item));
1654                         BUG_ON(err);
1655
1656                         mutex_lock(&info->extent_ins_mutex);
1657                         clear_extent_bits(&info->extent_ins, start, end,
1658                                           EXTENT_WRITEBACK, GFP_NOFS);
1659                         mutex_unlock(&info->extent_ins_mutex);
1660
1661                         err = insert_extent_backref(trans, extent_root, path,
1662                                                 start, extent_op->parent,
1663                                                 extent_root->root_key.objectid,
1664                                                 extent_op->generation,
1665                                                 extent_op->level);
1666                         BUG_ON(err);
1667                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1668                         err = lookup_extent_backref(trans, extent_root, path,
1669                                                 start, extent_op->orig_parent,
1670                                                 extent_root->root_key.objectid,
1671                                                 extent_op->orig_generation,
1672                                                 extent_op->level, 0);
1673                         BUG_ON(err);
1674
1675                         mutex_lock(&info->extent_ins_mutex);
1676                         clear_extent_bits(&info->extent_ins, start, end,
1677                                           EXTENT_WRITEBACK, GFP_NOFS);
1678                         mutex_unlock(&info->extent_ins_mutex);
1679
1680                         key.objectid = start;
1681                         key.offset = extent_op->parent;
1682                         key.type = BTRFS_EXTENT_REF_KEY;
1683                         err = btrfs_set_item_key_safe(trans, extent_root, path,
1684                                                       &key);
1685                         BUG_ON(err);
1686                         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1687                                              struct btrfs_extent_ref);
1688                         btrfs_set_ref_generation(path->nodes[0], ref,
1689                                                  extent_op->generation);
1690                         btrfs_mark_buffer_dirty(path->nodes[0]);
1691                         btrfs_release_path(extent_root, path);
1692                 } else {
1693                         BUG_ON(1);
1694                 }
1695                 kfree(extent_op);
1696                 unlock_extent(&info->extent_ins, start, end, GFP_NOFS);
1697                 if (all)
1698                         search = 0;
1699                 else
1700                         search = end + 1;
1701
1702                 cond_resched();
1703         }
1704         btrfs_free_path(path);
1705         return 0;
1706 }
1707
1708 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1709                           struct btrfs_root *root,
1710                           u64 bytenr, u64 num_bytes, int is_data)
1711 {
1712         int err = 0;
1713         struct extent_buffer *buf;
1714
1715         if (is_data)
1716                 goto pinit;
1717
1718         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1719         if (!buf)
1720                 goto pinit;
1721
1722         /* we can reuse a block if it hasn't been written
1723          * and it is from this transaction.  We can't
1724          * reuse anything from the tree log root because
1725          * it has tiny sub-transactions.
1726          */
1727         if (btrfs_buffer_uptodate(buf, 0) &&
1728             btrfs_try_tree_lock(buf)) {
1729                 u64 header_owner = btrfs_header_owner(buf);
1730                 u64 header_transid = btrfs_header_generation(buf);
1731                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1732                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
1733                     header_transid == trans->transid &&
1734                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1735                         clean_tree_block(NULL, root, buf);
1736                         btrfs_tree_unlock(buf);
1737                         free_extent_buffer(buf);
1738                         return 1;
1739                 }
1740                 btrfs_tree_unlock(buf);
1741         }
1742         free_extent_buffer(buf);
1743 pinit:
1744         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1745
1746         BUG_ON(err < 0);
1747         return 0;
1748 }
1749
1750 /*
1751  * remove an extent from the root, returns 0 on success
1752  */
1753 static int __free_extent(struct btrfs_trans_handle *trans,
1754                          struct btrfs_root *root,
1755                          u64 bytenr, u64 num_bytes, u64 parent,
1756                          u64 root_objectid, u64 ref_generation,
1757                          u64 owner_objectid, int pin, int mark_free)
1758 {
1759         struct btrfs_path *path;
1760         struct btrfs_key key;
1761         struct btrfs_fs_info *info = root->fs_info;
1762         struct btrfs_root *extent_root = info->extent_root;
1763         struct extent_buffer *leaf;
1764         int ret;
1765         int extent_slot = 0;
1766         int found_extent = 0;
1767         int num_to_del = 1;
1768         struct btrfs_extent_item *ei;
1769         u32 refs;
1770
1771         key.objectid = bytenr;
1772         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1773         key.offset = num_bytes;
1774         path = btrfs_alloc_path();
1775         if (!path)
1776                 return -ENOMEM;
1777
1778         path->reada = 1;
1779         ret = lookup_extent_backref(trans, extent_root, path,
1780                                     bytenr, parent, root_objectid,
1781                                     ref_generation, owner_objectid, 1);
1782         if (ret == 0) {
1783                 struct btrfs_key found_key;
1784                 extent_slot = path->slots[0];
1785                 while(extent_slot > 0) {
1786                         extent_slot--;
1787                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1788                                               extent_slot);
1789                         if (found_key.objectid != bytenr)
1790                                 break;
1791                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1792                             found_key.offset == num_bytes) {
1793                                 found_extent = 1;
1794                                 break;
1795                         }
1796                         if (path->slots[0] - extent_slot > 5)
1797                                 break;
1798                 }
1799                 if (!found_extent) {
1800                         ret = remove_extent_backref(trans, extent_root, path);
1801                         BUG_ON(ret);
1802                         btrfs_release_path(extent_root, path);
1803                         ret = btrfs_search_slot(trans, extent_root,
1804                                                 &key, path, -1, 1);
1805                         BUG_ON(ret);
1806                         extent_slot = path->slots[0];
1807                 }
1808         } else {
1809                 btrfs_print_leaf(extent_root, path->nodes[0]);
1810                 WARN_ON(1);
1811                 printk("Unable to find ref byte nr %Lu root %Lu "
1812                        "gen %Lu owner %Lu\n", bytenr,
1813                        root_objectid, ref_generation, owner_objectid);
1814         }
1815
1816         leaf = path->nodes[0];
1817         ei = btrfs_item_ptr(leaf, extent_slot,
1818                             struct btrfs_extent_item);
1819         refs = btrfs_extent_refs(leaf, ei);
1820         BUG_ON(refs == 0);
1821         refs -= 1;
1822         btrfs_set_extent_refs(leaf, ei, refs);
1823
1824         btrfs_mark_buffer_dirty(leaf);
1825
1826         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1827                 struct btrfs_extent_ref *ref;
1828                 ref = btrfs_item_ptr(leaf, path->slots[0],
1829                                      struct btrfs_extent_ref);
1830                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
1831                 /* if the back ref and the extent are next to each other
1832                  * they get deleted below in one shot
1833                  */
1834                 path->slots[0] = extent_slot;
1835                 num_to_del = 2;
1836         } else if (found_extent) {
1837                 /* otherwise delete the extent back ref */
1838                 ret = remove_extent_backref(trans, extent_root, path);
1839                 BUG_ON(ret);
1840                 /* if refs are 0, we need to setup the path for deletion */
1841                 if (refs == 0) {
1842                         btrfs_release_path(extent_root, path);
1843                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1844                                                 -1, 1);
1845                         BUG_ON(ret);
1846                 }
1847         }
1848
1849         if (refs == 0) {
1850                 u64 super_used;
1851                 u64 root_used;
1852 #ifdef BIO_RW_DISCARD
1853                 u64 map_length = num_bytes;
1854                 struct btrfs_multi_bio *multi = NULL;
1855 #endif
1856
1857                 if (pin) {
1858                         mutex_lock(&root->fs_info->pinned_mutex);
1859                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
1860                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
1861                         mutex_unlock(&root->fs_info->pinned_mutex);
1862                         if (ret > 0)
1863                                 mark_free = 1;
1864                         BUG_ON(ret < 0);
1865                 }
1866
1867                 /* block accounting for super block */
1868                 spin_lock_irq(&info->delalloc_lock);
1869                 super_used = btrfs_super_bytes_used(&info->super_copy);
1870                 btrfs_set_super_bytes_used(&info->super_copy,
1871                                            super_used - num_bytes);
1872                 spin_unlock_irq(&info->delalloc_lock);
1873
1874                 /* block accounting for root item */
1875                 root_used = btrfs_root_used(&root->root_item);
1876                 btrfs_set_root_used(&root->root_item,
1877                                            root_used - num_bytes);
1878                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1879                                       num_to_del);
1880                 BUG_ON(ret);
1881                 btrfs_release_path(extent_root, path);
1882                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1883                                          mark_free);
1884                 BUG_ON(ret);
1885
1886 #ifdef BIO_RW_DISCARD
1887                 /* Tell the block device(s) that the sectors can be discarded */
1888                 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1889                                       bytenr, &map_length, &multi, 0);
1890                 if (!ret) {
1891                         struct btrfs_bio_stripe *stripe = multi->stripes;
1892                         int i;
1893
1894                         if (map_length > num_bytes)
1895                                 map_length = num_bytes;
1896
1897                         for (i = 0; i < multi->num_stripes; i++, stripe++) {
1898                                 blkdev_issue_discard(stripe->dev->bdev,
1899                                                      stripe->physical >> 9,
1900                                                      map_length >> 9);
1901                         }
1902                         kfree(multi);
1903                 }
1904 #endif
1905         }
1906         btrfs_free_path(path);
1907         finish_current_insert(trans, extent_root, 0);
1908         return ret;
1909 }
1910
1911 /*
1912  * find all the blocks marked as pending in the radix tree and remove
1913  * them from the extent map
1914  */
1915 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1916                                btrfs_root *extent_root, int all)
1917 {
1918         int ret;
1919         int err = 0;
1920         u64 start;
1921         u64 end;
1922         u64 priv;
1923         u64 search = 0;
1924         struct extent_io_tree *pending_del;
1925         struct extent_io_tree *extent_ins;
1926         struct pending_extent_op *extent_op;
1927         struct btrfs_fs_info *info = extent_root->fs_info;
1928
1929         extent_ins = &extent_root->fs_info->extent_ins;
1930         pending_del = &extent_root->fs_info->pending_del;
1931
1932         while(1) {
1933                 mutex_lock(&info->extent_ins_mutex);
1934                 ret = find_first_extent_bit(pending_del, search, &start, &end,
1935                                             EXTENT_WRITEBACK);
1936                 if (ret) {
1937                         mutex_unlock(&info->extent_ins_mutex);
1938                         if (all && search) {
1939                                 search = 0;
1940                                 continue;
1941                         }
1942                         break;
1943                 }
1944
1945                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
1946                 if (!ret) {
1947                         search = end+1;
1948                         mutex_unlock(&info->extent_ins_mutex);
1949                         cond_resched();
1950                         continue;
1951                 }
1952                 BUG_ON(ret < 0);
1953
1954                 ret = get_state_private(pending_del, start, &priv);
1955                 BUG_ON(ret);
1956                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1957
1958                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1959                                   GFP_NOFS);
1960                 if (!test_range_bit(extent_ins, start, end,
1961                                     EXTENT_WRITEBACK, 0)) {
1962                         mutex_unlock(&info->extent_ins_mutex);
1963 free_extent:
1964                         ret = __free_extent(trans, extent_root,
1965                                             start, end + 1 - start,
1966                                             extent_op->orig_parent,
1967                                             extent_root->root_key.objectid,
1968                                             extent_op->orig_generation,
1969                                             extent_op->level, 1, 0);
1970                         kfree(extent_op);
1971                 } else {
1972                         kfree(extent_op);
1973
1974                         ret = get_state_private(&info->extent_ins, start,
1975                                                 &priv);
1976                         BUG_ON(ret);
1977                         extent_op = (struct pending_extent_op *)
1978                                                 (unsigned long)priv;
1979
1980                         clear_extent_bits(&info->extent_ins, start, end,
1981                                           EXTENT_WRITEBACK, GFP_NOFS);
1982
1983                         mutex_unlock(&info->extent_ins_mutex);
1984
1985                         if (extent_op->type == PENDING_BACKREF_UPDATE)
1986                                 goto free_extent;
1987
1988                         mutex_lock(&extent_root->fs_info->pinned_mutex);
1989                         ret = pin_down_bytes(trans, extent_root, start,
1990                                              end + 1 - start, 0);
1991                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
1992
1993                         ret = update_block_group(trans, extent_root, start,
1994                                                 end + 1 - start, 0, ret > 0);
1995
1996                         BUG_ON(ret);
1997                         kfree(extent_op);
1998                 }
1999                 if (ret)
2000                         err = ret;
2001                 unlock_extent(extent_ins, start, end, GFP_NOFS);
2002
2003                 if (all)
2004                         search = 0;
2005                 else
2006                         search = end + 1;
2007                 cond_resched();
2008         }
2009         return err;
2010 }
2011
2012 /*
2013  * remove an extent from the root, returns 0 on success
2014  */
2015 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2016                                struct btrfs_root *root,
2017                                u64 bytenr, u64 num_bytes, u64 parent,
2018                                u64 root_objectid, u64 ref_generation,
2019                                u64 owner_objectid, int pin)
2020 {
2021         struct btrfs_root *extent_root = root->fs_info->extent_root;
2022         int pending_ret;
2023         int ret;
2024
2025         WARN_ON(num_bytes < root->sectorsize);
2026         if (root == extent_root) {
2027                 struct pending_extent_op *extent_op;
2028
2029                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2030                 BUG_ON(!extent_op);
2031
2032                 extent_op->type = PENDING_EXTENT_DELETE;
2033                 extent_op->bytenr = bytenr;
2034                 extent_op->num_bytes = num_bytes;
2035                 extent_op->parent = parent;
2036                 extent_op->orig_parent = parent;
2037                 extent_op->generation = ref_generation;
2038                 extent_op->orig_generation = ref_generation;
2039                 extent_op->level = (int)owner_objectid;
2040
2041                 mutex_lock(&root->fs_info->extent_ins_mutex);
2042                 set_extent_bits(&root->fs_info->pending_del,
2043                                 bytenr, bytenr + num_bytes - 1,
2044                                 EXTENT_WRITEBACK, GFP_NOFS);
2045                 set_state_private(&root->fs_info->pending_del,
2046                                   bytenr, (unsigned long)extent_op);
2047                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2048                 return 0;
2049         }
2050         /* if metadata always pin */
2051         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2052                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2053                         struct btrfs_block_group_cache *cache;
2054
2055                         /* btrfs_free_reserved_extent */
2056                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2057                         BUG_ON(!cache);
2058                         btrfs_add_free_space(cache, bytenr, num_bytes);
2059                         update_reserved_extents(root, bytenr, num_bytes, 0);
2060                         return 0;
2061                 }
2062                 pin = 1;
2063         }
2064
2065         /* if data pin when any transaction has committed this */
2066         if (ref_generation != trans->transid)
2067                 pin = 1;
2068
2069         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2070                             root_objectid, ref_generation,
2071                             owner_objectid, pin, pin == 0);
2072
2073         finish_current_insert(trans, root->fs_info->extent_root, 0);
2074         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2075         return ret ? ret : pending_ret;
2076 }
2077
2078 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2079                       struct btrfs_root *root,
2080                       u64 bytenr, u64 num_bytes, u64 parent,
2081                       u64 root_objectid, u64 ref_generation,
2082                       u64 owner_objectid, int pin)
2083 {
2084         int ret;
2085
2086         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2087                                   root_objectid, ref_generation,
2088                                   owner_objectid, pin);
2089         return ret;
2090 }
2091
2092 static u64 stripe_align(struct btrfs_root *root, u64 val)
2093 {
2094         u64 mask = ((u64)root->stripesize - 1);
2095         u64 ret = (val + mask) & ~mask;
2096         return ret;
2097 }
2098
2099 /*
2100  * walks the btree of allocated extents and find a hole of a given size.
2101  * The key ins is changed to record the hole:
2102  * ins->objectid == block start
2103  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2104  * ins->offset == number of blocks
2105  * Any available blocks before search_start are skipped.
2106  */
2107 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2108                                      struct btrfs_root *orig_root,
2109                                      u64 num_bytes, u64 empty_size,
2110                                      u64 search_start, u64 search_end,
2111                                      u64 hint_byte, struct btrfs_key *ins,
2112                                      u64 exclude_start, u64 exclude_nr,
2113                                      int data)
2114 {
2115         int ret = 0;
2116         struct btrfs_root * root = orig_root->fs_info->extent_root;
2117         u64 total_needed = num_bytes;
2118         u64 *last_ptr = NULL;
2119         u64 last_wanted = 0;
2120         struct btrfs_block_group_cache *block_group = NULL;
2121         int chunk_alloc_done = 0;
2122         int empty_cluster = 2 * 1024 * 1024;
2123         int allowed_chunk_alloc = 0;
2124         struct list_head *head = NULL, *cur = NULL;
2125         int loop = 0;
2126         int extra_loop = 0;
2127         struct btrfs_space_info *space_info;
2128
2129         WARN_ON(num_bytes < root->sectorsize);
2130         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2131         ins->objectid = 0;
2132         ins->offset = 0;
2133
2134         if (orig_root->ref_cows || empty_size)
2135                 allowed_chunk_alloc = 1;
2136
2137         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2138                 last_ptr = &root->fs_info->last_alloc;
2139                 empty_cluster = 64 * 1024;
2140         }
2141
2142         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2143                 last_ptr = &root->fs_info->last_data_alloc;
2144
2145         if (last_ptr) {
2146                 if (*last_ptr) {
2147                         hint_byte = *last_ptr;
2148                         last_wanted = *last_ptr;
2149                 } else
2150                         empty_size += empty_cluster;
2151         } else {
2152                 empty_cluster = 0;
2153         }
2154         search_start = max(search_start, first_logical_byte(root, 0));
2155         search_start = max(search_start, hint_byte);
2156
2157         if (last_wanted && search_start != last_wanted) {
2158                 last_wanted = 0;
2159                 empty_size += empty_cluster;
2160         }
2161
2162         total_needed += empty_size;
2163         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2164         if (!block_group)
2165                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2166                                                              search_start);
2167         space_info = __find_space_info(root->fs_info, data);
2168
2169         down_read(&space_info->groups_sem);
2170         while (1) {
2171                 struct btrfs_free_space *free_space;
2172                 /*
2173                  * the only way this happens if our hint points to a block
2174                  * group thats not of the proper type, while looping this
2175                  * should never happen
2176                  */
2177                 if (empty_size)
2178                         extra_loop = 1;
2179
2180                 if (!block_group)
2181                         goto new_group_no_lock;
2182
2183                 mutex_lock(&block_group->alloc_mutex);
2184                 if (unlikely(!block_group_bits(block_group, data)))
2185                         goto new_group;
2186
2187                 ret = cache_block_group(root, block_group);
2188                 if (ret) {
2189                         mutex_unlock(&block_group->alloc_mutex);
2190                         break;
2191                 }
2192
2193                 if (block_group->ro)
2194                         goto new_group;
2195
2196                 free_space = btrfs_find_free_space(block_group, search_start,
2197                                                    total_needed);
2198                 if (free_space) {
2199                         u64 start = block_group->key.objectid;
2200                         u64 end = block_group->key.objectid +
2201                                 block_group->key.offset;
2202
2203                         search_start = stripe_align(root, free_space->offset);
2204
2205                         /* move on to the next group */
2206                         if (search_start + num_bytes >= search_end)
2207                                 goto new_group;
2208
2209                         /* move on to the next group */
2210                         if (search_start + num_bytes > end)
2211                                 goto new_group;
2212
2213                         if (last_wanted && search_start != last_wanted) {
2214                                 total_needed += empty_cluster;
2215                                 empty_size += empty_cluster;
2216                                 last_wanted = 0;
2217                                 /*
2218                                  * if search_start is still in this block group
2219                                  * then we just re-search this block group
2220                                  */
2221                                 if (search_start >= start &&
2222                                     search_start < end) {
2223                                         mutex_unlock(&block_group->alloc_mutex);
2224                                         continue;
2225                                 }
2226
2227                                 /* else we go to the next block group */
2228                                 goto new_group;
2229                         }
2230
2231                         if (exclude_nr > 0 &&
2232                             (search_start + num_bytes > exclude_start &&
2233                              search_start < exclude_start + exclude_nr)) {
2234                                 search_start = exclude_start + exclude_nr;
2235                                 /*
2236                                  * if search_start is still in this block group
2237                                  * then we just re-search this block group
2238                                  */
2239                                 if (search_start >= start &&
2240                                     search_start < end) {
2241                                         mutex_unlock(&block_group->alloc_mutex);
2242                                         last_wanted = 0;
2243                                         continue;
2244                                 }
2245
2246                                 /* else we go to the next block group */
2247                                 goto new_group;
2248                         }
2249
2250                         ins->objectid = search_start;
2251                         ins->offset = num_bytes;
2252
2253                         btrfs_remove_free_space_lock(block_group, search_start,
2254                                                      num_bytes);
2255                         /* we are all good, lets return */
2256                         mutex_unlock(&block_group->alloc_mutex);
2257                         break;
2258                 }
2259 new_group:
2260                 mutex_unlock(&block_group->alloc_mutex);
2261 new_group_no_lock:
2262                 /* don't try to compare new allocations against the
2263                  * last allocation any more
2264                  */
2265                 last_wanted = 0;
2266
2267                 /*
2268                  * Here's how this works.
2269                  * loop == 0: we were searching a block group via a hint
2270                  *              and didn't find anything, so we start at
2271                  *              the head of the block groups and keep searching
2272                  * loop == 1: we're searching through all of the block groups
2273                  *              if we hit the head again we have searched
2274                  *              all of the block groups for this space and we
2275                  *              need to try and allocate, if we cant error out.
2276                  * loop == 2: we allocated more space and are looping through
2277                  *              all of the block groups again.
2278                  */
2279                 if (loop == 0) {
2280                         head = &space_info->block_groups;
2281                         cur = head->next;
2282                         loop++;
2283                 } else if (loop == 1 && cur == head) {
2284                         int keep_going;
2285
2286                         /* at this point we give up on the empty_size
2287                          * allocations and just try to allocate the min
2288                          * space.
2289                          *
2290                          * The extra_loop field was set if an empty_size
2291                          * allocation was attempted above, and if this
2292                          * is try we need to try the loop again without
2293                          * the additional empty_size.
2294                          */
2295                         total_needed -= empty_size;
2296                         empty_size = 0;
2297                         keep_going = extra_loop;
2298                         loop++;
2299
2300                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2301                                 up_read(&space_info->groups_sem);
2302                                 ret = do_chunk_alloc(trans, root, num_bytes +
2303                                                      2 * 1024 * 1024, data, 1);
2304                                 down_read(&space_info->groups_sem);
2305                                 if (ret < 0)
2306                                         goto loop_check;
2307                                 head = &space_info->block_groups;
2308                                 /*
2309                                  * we've allocated a new chunk, keep
2310                                  * trying
2311                                  */
2312                                 keep_going = 1;
2313                                 chunk_alloc_done = 1;
2314                         } else if (!allowed_chunk_alloc) {
2315                                 space_info->force_alloc = 1;
2316                         }
2317 loop_check:
2318                         if (keep_going) {
2319                                 cur = head->next;
2320                                 extra_loop = 0;
2321                         } else {
2322                                 break;
2323                         }
2324                 } else if (cur == head) {
2325                         break;
2326                 }
2327
2328                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2329                                          list);
2330                 search_start = block_group->key.objectid;
2331                 cur = cur->next;
2332         }
2333
2334         /* we found what we needed */
2335         if (ins->objectid) {
2336                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2337                         trans->block_group = block_group;
2338
2339                 if (last_ptr)
2340                         *last_ptr = ins->objectid + ins->offset;
2341                 ret = 0;
2342         } else if (!ret) {
2343                 ret = -ENOSPC;
2344         }
2345
2346         up_read(&space_info->groups_sem);
2347         return ret;
2348 }
2349
2350 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2351 {
2352         struct btrfs_block_group_cache *cache;
2353         struct list_head *l;
2354
2355         printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
2356                info->total_bytes - info->bytes_used - info->bytes_pinned -
2357                info->bytes_reserved, (info->full) ? "" : "not ");
2358
2359         down_read(&info->groups_sem);
2360         list_for_each(l, &info->block_groups) {
2361                 cache = list_entry(l, struct btrfs_block_group_cache, list);
2362                 spin_lock(&cache->lock);
2363                 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
2364                        "%Lu pinned %Lu reserved\n",
2365                        cache->key.objectid, cache->key.offset,
2366                        btrfs_block_group_used(&cache->item),
2367                        cache->pinned, cache->reserved);
2368                 btrfs_dump_free_space(cache, bytes);
2369                 spin_unlock(&cache->lock);
2370         }
2371         up_read(&info->groups_sem);
2372 }
2373
2374 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2375                                   struct btrfs_root *root,
2376                                   u64 num_bytes, u64 min_alloc_size,
2377                                   u64 empty_size, u64 hint_byte,
2378                                   u64 search_end, struct btrfs_key *ins,
2379                                   u64 data)
2380 {
2381         int ret;
2382         u64 search_start = 0;
2383         u64 alloc_profile;
2384         struct btrfs_fs_info *info = root->fs_info;
2385
2386         if (data) {
2387                 alloc_profile = info->avail_data_alloc_bits &
2388                                 info->data_alloc_profile;
2389                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2390         } else if (root == root->fs_info->chunk_root) {
2391                 alloc_profile = info->avail_system_alloc_bits &
2392                                 info->system_alloc_profile;
2393                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2394         } else {
2395                 alloc_profile = info->avail_metadata_alloc_bits &
2396                                 info->metadata_alloc_profile;
2397                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2398         }
2399 again:
2400         data = reduce_alloc_profile(root, data);
2401         /*
2402          * the only place that sets empty_size is btrfs_realloc_node, which
2403          * is not called recursively on allocations
2404          */
2405         if (empty_size || root->ref_cows) {
2406                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2407                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2408                                      2 * 1024 * 1024,
2409                                      BTRFS_BLOCK_GROUP_METADATA |
2410                                      (info->metadata_alloc_profile &
2411                                       info->avail_metadata_alloc_bits), 0);
2412                 }
2413                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2414                                      num_bytes + 2 * 1024 * 1024, data, 0);
2415         }
2416
2417         WARN_ON(num_bytes < root->sectorsize);
2418         ret = find_free_extent(trans, root, num_bytes, empty_size,
2419                                search_start, search_end, hint_byte, ins,
2420                                trans->alloc_exclude_start,
2421                                trans->alloc_exclude_nr, data);
2422
2423         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2424                 num_bytes = num_bytes >> 1;
2425                 num_bytes = num_bytes & ~(root->sectorsize - 1);
2426                 num_bytes = max(num_bytes, min_alloc_size);
2427                 do_chunk_alloc(trans, root->fs_info->extent_root,
2428                                num_bytes, data, 1);
2429                 goto again;
2430         }
2431         if (ret) {
2432                 struct btrfs_space_info *sinfo;
2433
2434                 sinfo = __find_space_info(root->fs_info, data);
2435                 printk("allocation failed flags %Lu, wanted %Lu\n",
2436                        data, num_bytes);
2437                 dump_space_info(sinfo, num_bytes);
2438                 BUG();
2439         }
2440
2441         return ret;
2442 }
2443
2444 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2445 {
2446         struct btrfs_block_group_cache *cache;
2447
2448         cache = btrfs_lookup_block_group(root->fs_info, start);
2449         if (!cache) {
2450                 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
2451                 return -ENOSPC;
2452         }
2453         btrfs_add_free_space(cache, start, len);
2454         update_reserved_extents(root, start, len, 0);
2455         return 0;
2456 }
2457
2458 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2459                                   struct btrfs_root *root,
2460                                   u64 num_bytes, u64 min_alloc_size,
2461                                   u64 empty_size, u64 hint_byte,
2462                                   u64 search_end, struct btrfs_key *ins,
2463                                   u64 data)
2464 {
2465         int ret;
2466         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2467                                      empty_size, hint_byte, search_end, ins,
2468                                      data);
2469         update_reserved_extents(root, ins->objectid, ins->offset, 1);
2470         return ret;
2471 }
2472
2473 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2474                                          struct btrfs_root *root, u64 parent,
2475                                          u64 root_objectid, u64 ref_generation,
2476                                          u64 owner, struct btrfs_key *ins)
2477 {
2478         int ret;
2479         int pending_ret;
2480         u64 super_used;
2481         u64 root_used;
2482         u64 num_bytes = ins->offset;
2483         u32 sizes[2];
2484         struct btrfs_fs_info *info = root->fs_info;
2485         struct btrfs_root *extent_root = info->extent_root;
2486         struct btrfs_extent_item *extent_item;
2487         struct btrfs_extent_ref *ref;
2488         struct btrfs_path *path;
2489         struct btrfs_key keys[2];
2490
2491         if (parent == 0)
2492                 parent = ins->objectid;
2493
2494         /* block accounting for super block */
2495         spin_lock_irq(&info->delalloc_lock);
2496         super_used = btrfs_super_bytes_used(&info->super_copy);
2497         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2498         spin_unlock_irq(&info->delalloc_lock);
2499
2500         /* block accounting for root item */
2501         root_used = btrfs_root_used(&root->root_item);
2502         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2503
2504         if (root == extent_root) {
2505                 struct pending_extent_op *extent_op;
2506
2507                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2508                 BUG_ON(!extent_op);
2509
2510                 extent_op->type = PENDING_EXTENT_INSERT;
2511                 extent_op->bytenr = ins->objectid;
2512                 extent_op->num_bytes = ins->offset;
2513                 extent_op->parent = parent;
2514                 extent_op->orig_parent = 0;
2515                 extent_op->generation = ref_generation;
2516                 extent_op->orig_generation = 0;
2517                 extent_op->level = (int)owner;
2518
2519                 mutex_lock(&root->fs_info->extent_ins_mutex);
2520                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2521                                 ins->objectid + ins->offset - 1,
2522                                 EXTENT_WRITEBACK, GFP_NOFS);
2523                 set_state_private(&root->fs_info->extent_ins,
2524                                   ins->objectid, (unsigned long)extent_op);
2525                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2526                 goto update_block;
2527         }
2528
2529         memcpy(&keys[0], ins, sizeof(*ins));
2530         keys[1].objectid = ins->objectid;
2531         keys[1].type = BTRFS_EXTENT_REF_KEY;
2532         keys[1].offset = parent;
2533         sizes[0] = sizeof(*extent_item);
2534         sizes[1] = sizeof(*ref);
2535
2536         path = btrfs_alloc_path();
2537         BUG_ON(!path);
2538
2539         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2540                                        sizes, 2);
2541         BUG_ON(ret);
2542
2543         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2544                                      struct btrfs_extent_item);
2545         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2546         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2547                              struct btrfs_extent_ref);
2548
2549         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2550         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2551         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2552         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
2553
2554         btrfs_mark_buffer_dirty(path->nodes[0]);
2555
2556         trans->alloc_exclude_start = 0;
2557         trans->alloc_exclude_nr = 0;
2558         btrfs_free_path(path);
2559         finish_current_insert(trans, extent_root, 0);
2560         pending_ret = del_pending_extents(trans, extent_root, 0);
2561
2562         if (ret)
2563                 goto out;
2564         if (pending_ret) {
2565                 ret = pending_ret;
2566                 goto out;
2567         }
2568
2569 update_block:
2570         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2571         if (ret) {
2572                 printk("update block group failed for %Lu %Lu\n",
2573                        ins->objectid, ins->offset);
2574                 BUG();
2575         }
2576 out:
2577         return ret;
2578 }
2579
2580 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2581                                 struct btrfs_root *root, u64 parent,
2582                                 u64 root_objectid, u64 ref_generation,
2583                                 u64 owner, struct btrfs_key *ins)
2584 {
2585         int ret;
2586
2587         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2588                 return 0;
2589         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2590                                             ref_generation, owner, ins);
2591         update_reserved_extents(root, ins->objectid, ins->offset, 0);
2592         return ret;
2593 }
2594
2595 /*
2596  * this is used by the tree logging recovery code.  It records that
2597  * an extent has been allocated and makes sure to clear the free
2598  * space cache bits as well
2599  */
2600 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2601                                 struct btrfs_root *root, u64 parent,
2602                                 u64 root_objectid, u64 ref_generation,
2603                                 u64 owner, struct btrfs_key *ins)
2604 {
2605         int ret;
2606         struct btrfs_block_group_cache *block_group;
2607
2608         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2609         mutex_lock(&block_group->alloc_mutex);
2610         cache_block_group(root, block_group);
2611
2612         ret = btrfs_remove_free_space_lock(block_group, ins->objectid,
2613                                            ins->offset);
2614         mutex_unlock(&block_group->alloc_mutex);
2615         BUG_ON(ret);
2616         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2617                                             ref_generation, owner, ins);
2618         return ret;
2619 }
2620
2621 /*
2622  * finds a free extent and does all the dirty work required for allocation
2623  * returns the key for the extent through ins, and a tree buffer for
2624  * the first block of the extent through buf.
2625  *
2626  * returns 0 if everything worked, non-zero otherwise.
2627  */
2628 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2629                        struct btrfs_root *root,
2630                        u64 num_bytes, u64 parent, u64 min_alloc_size,
2631                        u64 root_objectid, u64 ref_generation,
2632                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
2633                        u64 search_end, struct btrfs_key *ins, u64 data)
2634 {
2635         int ret;
2636
2637         ret = __btrfs_reserve_extent(trans, root, num_bytes,
2638                                      min_alloc_size, empty_size, hint_byte,
2639                                      search_end, ins, data);
2640         BUG_ON(ret);
2641         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2642                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2643                                         root_objectid, ref_generation,
2644                                         owner_objectid, ins);
2645                 BUG_ON(ret);
2646
2647         } else {
2648                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2649         }
2650         return ret;
2651 }
2652
2653 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2654                                             struct btrfs_root *root,
2655                                             u64 bytenr, u32 blocksize)
2656 {
2657         struct extent_buffer *buf;
2658
2659         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2660         if (!buf)
2661                 return ERR_PTR(-ENOMEM);
2662         btrfs_set_header_generation(buf, trans->transid);
2663         btrfs_tree_lock(buf);
2664         clean_tree_block(trans, root, buf);
2665         btrfs_set_buffer_uptodate(buf);
2666         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2667                 set_extent_dirty(&root->dirty_log_pages, buf->start,
2668                          buf->start + buf->len - 1, GFP_NOFS);
2669         } else {
2670                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2671                          buf->start + buf->len - 1, GFP_NOFS);
2672         }
2673         trans->blocks_used++;
2674         return buf;
2675 }
2676
2677 /*
2678  * helper function to allocate a block for a given tree
2679  * returns the tree buffer or NULL.
2680  */
2681 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2682                                              struct btrfs_root *root,
2683                                              u32 blocksize, u64 parent,
2684                                              u64 root_objectid,
2685                                              u64 ref_generation,
2686                                              int level,
2687                                              u64 hint,
2688                                              u64 empty_size)
2689 {
2690         struct btrfs_key ins;
2691         int ret;
2692         struct extent_buffer *buf;
2693
2694         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
2695                                  root_objectid, ref_generation, level,
2696                                  empty_size, hint, (u64)-1, &ins, 0);
2697         if (ret) {
2698                 BUG_ON(ret > 0);
2699                 return ERR_PTR(ret);
2700         }
2701
2702         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2703         return buf;
2704 }
2705
2706 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2707                         struct btrfs_root *root, struct extent_buffer *leaf)
2708 {
2709         u64 leaf_owner;
2710         u64 leaf_generation;
2711         struct btrfs_key key;
2712         struct btrfs_file_extent_item *fi;
2713         int i;
2714         int nritems;
2715         int ret;
2716
2717         BUG_ON(!btrfs_is_leaf(leaf));
2718         nritems = btrfs_header_nritems(leaf);
2719         leaf_owner = btrfs_header_owner(leaf);
2720         leaf_generation = btrfs_header_generation(leaf);
2721
2722         for (i = 0; i < nritems; i++) {
2723                 u64 disk_bytenr;
2724                 cond_resched();
2725
2726                 btrfs_item_key_to_cpu(leaf, &key, i);
2727                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2728                         continue;
2729                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2730                 if (btrfs_file_extent_type(leaf, fi) ==
2731                     BTRFS_FILE_EXTENT_INLINE)
2732                         continue;
2733                 /*
2734                  * FIXME make sure to insert a trans record that
2735                  * repeats the snapshot del on crash
2736                  */
2737                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2738                 if (disk_bytenr == 0)
2739                         continue;
2740
2741                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2742                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2743                                 leaf->start, leaf_owner, leaf_generation,
2744                                 key.objectid, 0);
2745                 BUG_ON(ret);
2746
2747                 atomic_inc(&root->fs_info->throttle_gen);
2748                 wake_up(&root->fs_info->transaction_throttle);
2749                 cond_resched();
2750         }
2751         return 0;
2752 }
2753
2754 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2755                                         struct btrfs_root *root,
2756                                         struct btrfs_leaf_ref *ref)
2757 {
2758         int i;
2759         int ret;
2760         struct btrfs_extent_info *info = ref->extents;
2761
2762         for (i = 0; i < ref->nritems; i++) {
2763                 ret = __btrfs_free_extent(trans, root, info->bytenr,
2764                                           info->num_bytes, ref->bytenr,
2765                                           ref->owner, ref->generation,
2766                                           info->objectid, 0);
2767
2768                 atomic_inc(&root->fs_info->throttle_gen);
2769                 wake_up(&root->fs_info->transaction_throttle);
2770                 cond_resched();
2771
2772                 BUG_ON(ret);
2773                 info++;
2774         }
2775
2776         return 0;
2777 }
2778
2779 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2780                               u32 *refs)
2781 {
2782         int ret;
2783
2784         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
2785         BUG_ON(ret);
2786
2787 #if 0 // some debugging code in case we see problems here
2788         /* if the refs count is one, it won't get increased again.  But
2789          * if the ref count is > 1, someone may be decreasing it at
2790          * the same time we are.
2791          */
2792         if (*refs != 1) {
2793                 struct extent_buffer *eb = NULL;
2794                 eb = btrfs_find_create_tree_block(root, start, len);
2795                 if (eb)
2796                         btrfs_tree_lock(eb);
2797
2798                 mutex_lock(&root->fs_info->alloc_mutex);
2799                 ret = lookup_extent_ref(NULL, root, start, len, refs);
2800                 BUG_ON(ret);
2801                 mutex_unlock(&root->fs_info->alloc_mutex);
2802
2803                 if (eb) {
2804                         btrfs_tree_unlock(eb);
2805                         free_extent_buffer(eb);
2806                 }
2807                 if (*refs == 1) {
2808                         printk("block %llu went down to one during drop_snap\n",
2809                                (unsigned long long)start);
2810                 }
2811
2812         }
2813 #endif
2814
2815         cond_resched();
2816         return ret;
2817 }
2818
2819 /*
2820  * helper function for drop_snapshot, this walks down the tree dropping ref
2821  * counts as it goes.
2822  */
2823 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2824                                    struct btrfs_root *root,
2825                                    struct btrfs_path *path, int *level)
2826 {
2827         u64 root_owner;
2828         u64 root_gen;
2829         u64 bytenr;
2830         u64 ptr_gen;
2831         struct extent_buffer *next;
2832         struct extent_buffer *cur;
2833         struct extent_buffer *parent;
2834         struct btrfs_leaf_ref *ref;
2835         u32 blocksize;
2836         int ret;
2837         u32 refs;
2838
2839         WARN_ON(*level < 0);
2840         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2841         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2842                                 path->nodes[*level]->len, &refs);
2843         BUG_ON(ret);
2844         if (refs > 1)
2845                 goto out;
2846
2847         /*
2848          * walk down to the last node level and free all the leaves
2849          */
2850         while(*level >= 0) {
2851                 WARN_ON(*level < 0);
2852                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2853                 cur = path->nodes[*level];
2854
2855                 if (btrfs_header_level(cur) != *level)
2856                         WARN_ON(1);
2857
2858                 if (path->slots[*level] >=
2859                     btrfs_header_nritems(cur))
2860                         break;
2861                 if (*level == 0) {
2862                         ret = btrfs_drop_leaf_ref(trans, root, cur);
2863                         BUG_ON(ret);
2864                         break;
2865                 }
2866                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2867                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2868                 blocksize = btrfs_level_size(root, *level - 1);
2869
2870                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2871                 BUG_ON(ret);
2872                 if (refs != 1) {
2873                         parent = path->nodes[*level];
2874                         root_owner = btrfs_header_owner(parent);
2875                         root_gen = btrfs_header_generation(parent);
2876                         path->slots[*level]++;
2877
2878                         ret = __btrfs_free_extent(trans, root, bytenr,
2879                                                 blocksize, parent->start,
2880                                                 root_owner, root_gen,
2881                                                 *level - 1, 1);
2882                         BUG_ON(ret);
2883
2884                         atomic_inc(&root->fs_info->throttle_gen);
2885                         wake_up(&root->fs_info->transaction_throttle);
2886                         cond_resched();
2887
2888                         continue;
2889                 }
2890                 /*
2891                  * at this point, we have a single ref, and since the
2892                  * only place referencing this extent is a dead root
2893                  * the reference count should never go higher.
2894                  * So, we don't need to check it again
2895                  */
2896                 if (*level == 1) {
2897                         ref = btrfs_lookup_leaf_ref(root, bytenr);
2898                         if (ref && ref->generation != ptr_gen) {
2899                                 btrfs_free_leaf_ref(root, ref);
2900                                 ref = NULL;
2901                         }
2902                         if (ref) {
2903                                 ret = cache_drop_leaf_ref(trans, root, ref);
2904                                 BUG_ON(ret);
2905                                 btrfs_remove_leaf_ref(root, ref);
2906                                 btrfs_free_leaf_ref(root, ref);
2907                                 *level = 0;
2908                                 break;
2909                         }
2910                         if (printk_ratelimit()) {
2911                                 printk("leaf ref miss for bytenr %llu\n",
2912                                        (unsigned long long)bytenr);
2913                         }
2914                 }
2915                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2916                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2917                         free_extent_buffer(next);
2918
2919                         next = read_tree_block(root, bytenr, blocksize,
2920                                                ptr_gen);
2921                         cond_resched();
2922 #if 0
2923                         /*
2924                          * this is a debugging check and can go away
2925                          * the ref should never go all the way down to 1
2926                          * at this point
2927                          */
2928                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2929                                                 &refs);
2930                         BUG_ON(ret);
2931                         WARN_ON(refs != 1);
2932 #endif
2933                 }
2934                 WARN_ON(*level <= 0);
2935                 if (path->nodes[*level-1])
2936                         free_extent_buffer(path->nodes[*level-1]);
2937                 path->nodes[*level-1] = next;
2938                 *level = btrfs_header_level(next);
2939                 path->slots[*level] = 0;
2940                 cond_resched();
2941         }
2942 out:
2943         WARN_ON(*level < 0);
2944         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2945
2946         if (path->nodes[*level] == root->node) {
2947                 parent = path->nodes[*level];
2948                 bytenr = path->nodes[*level]->start;
2949         } else {
2950                 parent = path->nodes[*level + 1];
2951                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
2952         }
2953
2954         blocksize = btrfs_level_size(root, *level);
2955         root_owner = btrfs_header_owner(parent);
2956         root_gen = btrfs_header_generation(parent);
2957
2958         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
2959                                   parent->start, root_owner, root_gen,
2960                                   *level, 1);
2961         free_extent_buffer(path->nodes[*level]);
2962         path->nodes[*level] = NULL;
2963         *level += 1;
2964         BUG_ON(ret);
2965
2966         cond_resched();
2967         return 0;
2968 }
2969
2970 /*
2971  * helper function for drop_subtree, this function is similar to
2972  * walk_down_tree. The main difference is that it checks reference
2973  * counts while tree blocks are locked.
2974  */
2975 static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
2976                                       struct btrfs_root *root,
2977                                       struct btrfs_path *path, int *level)
2978 {
2979         struct extent_buffer *next;
2980         struct extent_buffer *cur;
2981         struct extent_buffer *parent;
2982         u64 bytenr;
2983         u64 ptr_gen;
2984         u32 blocksize;
2985         u32 refs;
2986         int ret;
2987
2988         cur = path->nodes[*level];
2989         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
2990                                       &refs);
2991         BUG_ON(ret);
2992         if (refs > 1)
2993                 goto out;
2994
2995         while (*level >= 0) {
2996                 cur = path->nodes[*level];
2997                 if (*level == 0) {
2998                         ret = btrfs_drop_leaf_ref(trans, root, cur);
2999                         BUG_ON(ret);
3000                         clean_tree_block(trans, root, cur);
3001                         break;
3002                 }
3003                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3004                         clean_tree_block(trans, root, cur);
3005                         break;
3006                 }
3007
3008                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3009                 blocksize = btrfs_level_size(root, *level - 1);
3010                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3011
3012                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3013                 btrfs_tree_lock(next);
3014
3015                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3016                                               &refs);
3017                 BUG_ON(ret);
3018                 if (refs > 1) {
3019                         parent = path->nodes[*level];
3020                         ret = btrfs_free_extent(trans, root, bytenr,
3021                                         blocksize, parent->start,
3022                                         btrfs_header_owner(parent),
3023                                         btrfs_header_generation(parent),
3024                                         *level - 1, 1);
3025                         BUG_ON(ret);
3026                         path->slots[*level]++;
3027                         btrfs_tree_unlock(next);
3028                         free_extent_buffer(next);
3029                         continue;
3030                 }
3031
3032                 *level = btrfs_header_level(next);
3033                 path->nodes[*level] = next;
3034                 path->slots[*level] = 0;
3035                 path->locks[*level] = 1;
3036                 cond_resched();
3037         }
3038 out:
3039         parent = path->nodes[*level + 1];
3040         bytenr = path->nodes[*level]->start;
3041         blocksize = path->nodes[*level]->len;
3042
3043         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3044                         parent->start, btrfs_header_owner(parent),
3045                         btrfs_header_generation(parent), *level, 1);
3046         BUG_ON(ret);
3047
3048         if (path->locks[*level]) {
3049                 btrfs_tree_unlock(path->nodes[*level]);
3050                 path->locks[*level] = 0;
3051         }
3052         free_extent_buffer(path->nodes[*level]);
3053         path->nodes[*level] = NULL;
3054         *level += 1;
3055         cond_resched();
3056         return 0;
3057 }
3058
3059 /*
3060  * helper for dropping snapshots.  This walks back up the tree in the path
3061  * to find the first node higher up where we haven't yet gone through
3062  * all the slots
3063  */
3064 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3065                                  struct btrfs_root *root,
3066                                  struct btrfs_path *path,
3067                                  int *level, int max_level)
3068 {
3069         u64 root_owner;
3070         u64 root_gen;
3071         struct btrfs_root_item *root_item = &root->root_item;
3072         int i;
3073         int slot;
3074         int ret;
3075
3076         for (i = *level; i < max_level && path->nodes[i]; i++) {
3077                 slot = path->slots[i];
3078                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3079                         struct extent_buffer *node;
3080                         struct btrfs_disk_key disk_key;
3081                         node = path->nodes[i];
3082                         path->slots[i]++;
3083                         *level = i;
3084                         WARN_ON(*level == 0);
3085                         btrfs_node_key(node, &disk_key, path->slots[i]);
3086                         memcpy(&root_item->drop_progress,
3087                                &disk_key, sizeof(disk_key));
3088                         root_item->drop_level = i;
3089                         return 0;
3090                 } else {
3091                         struct extent_buffer *parent;
3092                         if (path->nodes[*level] == root->node)
3093                                 parent = path->nodes[*level];
3094                         else
3095                                 parent = path->nodes[*level + 1];
3096
3097                         root_owner = btrfs_header_owner(parent);
3098                         root_gen = btrfs_header_generation(parent);
3099
3100                         clean_tree_block(trans, root, path->nodes[*level]);
3101                         ret = btrfs_free_extent(trans, root,
3102                                                 path->nodes[*level]->start,
3103                                                 path->nodes[*level]->len,
3104                                                 parent->start, root_owner,
3105                                                 root_gen, *level, 1);
3106                         BUG_ON(ret);
3107                         if (path->locks[*level]) {
3108                                 btrfs_tree_unlock(path->nodes[*level]);
3109                                 path->locks[*level] = 0;
3110                         }
3111                         free_extent_buffer(path->nodes[*level]);
3112                         path->nodes[*level] = NULL;
3113                         *level = i + 1;
3114                 }
3115         }
3116         return 1;
3117 }
3118
3119 /*
3120  * drop the reference count on the tree rooted at 'snap'.  This traverses
3121  * the tree freeing any blocks that have a ref count of zero after being
3122  * decremented.
3123  */
3124 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3125                         *root)
3126 {
3127         int ret = 0;
3128         int wret;
3129         int level;
3130         struct btrfs_path *path;
3131         int i;
3132         int orig_level;
3133         struct btrfs_root_item *root_item = &root->root_item;
3134
3135         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3136         path = btrfs_alloc_path();
3137         BUG_ON(!path);
3138
3139         level = btrfs_header_level(root->node);
3140         orig_level = level;
3141         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3142                 path->nodes[level] = root->node;
3143                 extent_buffer_get(root->node);
3144                 path->slots[level] = 0;
3145         } else {
3146                 struct btrfs_key key;
3147                 struct btrfs_disk_key found_key;
3148                 struct extent_buffer *node;
3149
3150                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3151                 level = root_item->drop_level;
3152                 path->lowest_level = level;
3153                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3154                 if (wret < 0) {
3155                         ret = wret;
3156                         goto out;
3157                 }
3158                 node = path->nodes[level];
3159                 btrfs_node_key(node, &found_key, path->slots[level]);
3160                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3161                                sizeof(found_key)));
3162                 /*
3163                  * unlock our path, this is safe because only this
3164                  * function is allowed to delete this snapshot
3165                  */
3166                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3167                         if (path->nodes[i] && path->locks[i]) {
3168                                 path->locks[i] = 0;
3169                                 btrfs_tree_unlock(path->nodes[i]);
3170                         }
3171                 }
3172         }
3173         while(1) {
3174                 wret = walk_down_tree(trans, root, path, &level);
3175                 if (wret > 0)
3176                         break;
3177                 if (wret < 0)
3178                         ret = wret;
3179
3180                 wret = walk_up_tree(trans, root, path, &level,
3181                                     BTRFS_MAX_LEVEL);
3182                 if (wret > 0)
3183                         break;
3184                 if (wret < 0)
3185                         ret = wret;
3186                 if (trans->transaction->in_commit) {
3187                         ret = -EAGAIN;
3188                         break;
3189                 }
3190                 atomic_inc(&root->fs_info->throttle_gen);
3191                 wake_up(&root->fs_info->transaction_throttle);
3192         }
3193         for (i = 0; i <= orig_level; i++) {
3194                 if (path->nodes[i]) {
3195                         free_extent_buffer(path->nodes[i]);
3196                         path->nodes[i] = NULL;
3197                 }
3198         }
3199 out:
3200         btrfs_free_path(path);
3201         return ret;
3202 }
3203
3204 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3205                         struct btrfs_root *root,
3206                         struct extent_buffer *node,
3207                         struct extent_buffer *parent)
3208 {
3209         struct btrfs_path *path;
3210         int level;
3211         int parent_level;
3212         int ret = 0;
3213         int wret;
3214
3215         path = btrfs_alloc_path();
3216         BUG_ON(!path);
3217
3218         BUG_ON(!btrfs_tree_locked(parent));
3219         parent_level = btrfs_header_level(parent);
3220         extent_buffer_get(parent);
3221         path->nodes[parent_level] = parent;
3222         path->slots[parent_level] = btrfs_header_nritems(parent);
3223
3224         BUG_ON(!btrfs_tree_locked(node));
3225         level = btrfs_header_level(node);
3226         extent_buffer_get(node);
3227         path->nodes[level] = node;
3228         path->slots[level] = 0;
3229
3230         while (1) {
3231                 wret = walk_down_subtree(trans, root, path, &level);
3232                 if (wret < 0)
3233                         ret = wret;
3234                 if (wret != 0)
3235                         break;
3236
3237                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3238                 if (wret < 0)
3239                         ret = wret;
3240                 if (wret != 0)
3241                         break;
3242         }
3243
3244         btrfs_free_path(path);
3245         return ret;
3246 }
3247
3248 static unsigned long calc_ra(unsigned long start, unsigned long last,
3249                              unsigned long nr)
3250 {
3251         return min(last, start + nr - 1);
3252 }
3253
3254 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3255                                          u64 len)
3256 {
3257         u64 page_start;
3258         u64 page_end;
3259         unsigned long first_index;
3260         unsigned long last_index;
3261         unsigned long i;
3262         struct page *page;
3263         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3264         struct file_ra_state *ra;
3265         struct btrfs_ordered_extent *ordered;
3266         unsigned int total_read = 0;
3267         unsigned int total_dirty = 0;
3268         int ret = 0;
3269
3270         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3271
3272         mutex_lock(&inode->i_mutex);
3273         first_index = start >> PAGE_CACHE_SHIFT;
3274         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3275
3276         /* make sure the dirty trick played by the caller work */
3277         ret = invalidate_inode_pages2_range(inode->i_mapping,
3278                                             first_index, last_index);
3279         if (ret)
3280                 goto out_unlock;
3281
3282         file_ra_state_init(ra, inode->i_mapping);
3283
3284         for (i = first_index ; i <= last_index; i++) {
3285                 if (total_read % ra->ra_pages == 0) {
3286                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3287                                        calc_ra(i, last_index, ra->ra_pages));
3288                 }
3289                 total_read++;
3290 again:
3291                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3292                         BUG_ON(1);
3293                 page = grab_cache_page(inode->i_mapping, i);
3294                 if (!page) {
3295                         ret = -ENOMEM;
3296                         goto out_unlock;
3297                 }
3298                 if (!PageUptodate(page)) {
3299                         btrfs_readpage(NULL, page);
3300                         lock_page(page);
3301                         if (!PageUptodate(page)) {
3302                                 unlock_page(page);
3303                                 page_cache_release(page);
3304                                 ret = -EIO;
3305                                 goto out_unlock;
3306                         }
3307                 }
3308                 wait_on_page_writeback(page);
3309
3310                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3311                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3312                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3313
3314                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3315                 if (ordered) {
3316                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3317                         unlock_page(page);
3318                         page_cache_release(page);
3319                         btrfs_start_ordered_extent(inode, ordered, 1);
3320                         btrfs_put_ordered_extent(ordered);
3321                         goto again;
3322                 }
3323                 set_page_extent_mapped(page);
3324
3325                 btrfs_set_extent_delalloc(inode, page_start, page_end);
3326                 if (i == first_index)
3327                         set_extent_bits(io_tree, page_start, page_end,
3328                                         EXTENT_BOUNDARY, GFP_NOFS);
3329
3330                 set_page_dirty(page);
3331                 total_dirty++;
3332
3333                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3334                 unlock_page(page);
3335                 page_cache_release(page);
3336         }
3337
3338 out_unlock:
3339         kfree(ra);
3340         mutex_unlock(&inode->i_mutex);
3341         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3342         return ret;
3343 }
3344
3345 static int noinline relocate_data_extent(struct inode *reloc_inode,
3346                                          struct btrfs_key *extent_key,
3347                                          u64 offset)
3348 {
3349         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3350         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3351         struct extent_map *em;
3352         u64 start = extent_key->objectid - offset;
3353         u64 end = start + extent_key->offset - 1;
3354
3355         em = alloc_extent_map(GFP_NOFS);
3356         BUG_ON(!em || IS_ERR(em));
3357
3358         em->start = start;
3359         em->len = extent_key->offset;
3360         em->block_len = extent_key->offset;
3361         em->block_start = extent_key->objectid;
3362         em->bdev = root->fs_info->fs_devices->latest_bdev;
3363         set_bit(EXTENT_FLAG_PINNED, &em->flags);
3364
3365         /* setup extent map to cheat btrfs_readpage */
3366         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3367         while (1) {
3368                 int ret;
3369                 spin_lock(&em_tree->lock);
3370                 ret = add_extent_mapping(em_tree, em);
3371                 spin_unlock(&em_tree->lock);
3372                 if (ret != -EEXIST) {
3373                         free_extent_map(em);
3374                         break;
3375                 }
3376                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
3377         }
3378         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3379
3380         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
3381 }
3382
3383 struct btrfs_ref_path {
3384         u64 extent_start;
3385         u64 nodes[BTRFS_MAX_LEVEL];
3386         u64 root_objectid;
3387         u64 root_generation;
3388         u64 owner_objectid;
3389         u32 num_refs;
3390         int lowest_level;
3391         int current_level;
3392         int shared_level;
3393
3394         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
3395         u64 new_nodes[BTRFS_MAX_LEVEL];
3396 };
3397
3398 struct disk_extent {
3399         u64 ram_bytes;
3400         u64 disk_bytenr;
3401         u64 disk_num_bytes;
3402         u64 offset;
3403         u64 num_bytes;
3404         u8 compression;
3405         u8 encryption;
3406         u16 other_encoding;
3407 };
3408
3409 static int is_cowonly_root(u64 root_objectid)
3410 {
3411         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3412             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3413             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3414             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3415             root_objectid == BTRFS_TREE_LOG_OBJECTID)
3416                 return 1;
3417         return 0;
3418 }
3419
3420 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
3421                                     struct btrfs_root *extent_root,
3422                                     struct btrfs_ref_path *ref_path,
3423                                     int first_time)
3424 {
3425         struct extent_buffer *leaf;
3426         struct btrfs_path *path;
3427         struct btrfs_extent_ref *ref;
3428         struct btrfs_key key;
3429         struct btrfs_key found_key;
3430         u64 bytenr;
3431         u32 nritems;
3432         int level;
3433         int ret = 1;
3434
3435         path = btrfs_alloc_path();
3436         if (!path)
3437                 return -ENOMEM;
3438
3439         if (first_time) {
3440                 ref_path->lowest_level = -1;
3441                 ref_path->current_level = -1;
3442                 ref_path->shared_level = -1;
3443                 goto walk_up;
3444         }
3445 walk_down:
3446         level = ref_path->current_level - 1;
3447         while (level >= -1) {
3448                 u64 parent;
3449                 if (level < ref_path->lowest_level)
3450                         break;
3451
3452                 if (level >= 0) {
3453                         bytenr = ref_path->nodes[level];
3454                 } else {
3455                         bytenr = ref_path->extent_start;
3456                 }
3457                 BUG_ON(bytenr == 0);
3458
3459                 parent = ref_path->nodes[level + 1];
3460                 ref_path->nodes[level + 1] = 0;
3461                 ref_path->current_level = level;
3462                 BUG_ON(parent == 0);
3463
3464                 key.objectid = bytenr;
3465                 key.offset = parent + 1;
3466                 key.type = BTRFS_EXTENT_REF_KEY;
3467
3468                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3469                 if (ret < 0)
3470                         goto out;
3471                 BUG_ON(ret == 0);
3472
3473                 leaf = path->nodes[0];
3474                 nritems = btrfs_header_nritems(leaf);
3475                 if (path->slots[0] >= nritems) {
3476                         ret = btrfs_next_leaf(extent_root, path);
3477                         if (ret < 0)
3478                                 goto out;
3479                         if (ret > 0)
3480                                 goto next;
3481                         leaf = path->nodes[0];
3482                 }
3483
3484                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3485                 if (found_key.objectid == bytenr &&
3486                     found_key.type == BTRFS_EXTENT_REF_KEY) {
3487                         if (level < ref_path->shared_level)
3488                                 ref_path->shared_level = level;
3489                         goto found;
3490                 }
3491 next:
3492                 level--;
3493                 btrfs_release_path(extent_root, path);
3494                 cond_resched();
3495         }
3496         /* reached lowest level */
3497         ret = 1;
3498         goto out;
3499 walk_up:
3500         level = ref_path->current_level;
3501         while (level < BTRFS_MAX_LEVEL - 1) {
3502                 u64 ref_objectid;
3503                 if (level >= 0) {
3504                         bytenr = ref_path->nodes[level];
3505                 } else {
3506                         bytenr = ref_path->extent_start;
3507                 }
3508                 BUG_ON(bytenr == 0);
3509
3510                 key.objectid = bytenr;
3511                 key.offset = 0;
3512                 key.type = BTRFS_EXTENT_REF_KEY;
3513
3514                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3515                 if (ret < 0)
3516                         goto out;
3517
3518                 leaf = path->nodes[0];
3519                 nritems = btrfs_header_nritems(leaf);
3520                 if (path->slots[0] >= nritems) {
3521                         ret = btrfs_next_leaf(extent_root, path);
3522                         if (ret < 0)
3523                                 goto out;
3524                         if (ret > 0) {
3525                                 /* the extent was freed by someone */
3526                                 if (ref_path->lowest_level == level)
3527                                         goto out;
3528                                 btrfs_release_path(extent_root, path);
3529                                 goto walk_down;
3530                         }
3531                         leaf = path->nodes[0];
3532                 }
3533
3534                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3535                 if (found_key.objectid != bytenr ||
3536                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
3537                         /* the extent was freed by someone */
3538                         if (ref_path->lowest_level == level) {
3539                                 ret = 1;
3540                                 goto out;
3541                         }
3542                         btrfs_release_path(extent_root, path);
3543                         goto walk_down;
3544                 }
3545 found:
3546                 ref = btrfs_item_ptr(leaf, path->slots[0],
3547                                 struct btrfs_extent_ref);
3548                 ref_objectid = btrfs_ref_objectid(leaf, ref);
3549                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3550                         if (first_time) {
3551                                 level = (int)ref_objectid;
3552                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
3553                                 ref_path->lowest_level = level;
3554                                 ref_path->current_level = level;
3555                                 ref_path->nodes[level] = bytenr;
3556                         } else {
3557                                 WARN_ON(ref_objectid != level);
3558                         }
3559                 } else {
3560                         WARN_ON(level != -1);
3561                 }
3562                 first_time = 0;
3563
3564                 if (ref_path->lowest_level == level) {
3565                         ref_path->owner_objectid = ref_objectid;
3566                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
3567                 }
3568
3569                 /*
3570                  * the block is tree root or the block isn't in reference
3571                  * counted tree.
3572                  */
3573                 if (found_key.objectid == found_key.offset ||
3574                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
3575                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3576                         ref_path->root_generation =
3577                                 btrfs_ref_generation(leaf, ref);
3578                         if (level < 0) {
3579                                 /* special reference from the tree log */
3580                                 ref_path->nodes[0] = found_key.offset;
3581                                 ref_path->current_level = 0;
3582                         }
3583                         ret = 0;
3584                         goto out;
3585                 }
3586
3587                 level++;
3588                 BUG_ON(ref_path->nodes[level] != 0);
3589                 ref_path->nodes[level] = found_key.offset;
3590                 ref_path->current_level = level;
3591
3592                 /*
3593                  * the reference was created in the running transaction,
3594                  * no need to continue walking up.
3595                  */
3596                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
3597                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3598                         ref_path->root_generation =
3599                                 btrfs_ref_generation(leaf, ref);
3600                         ret = 0;
3601                         goto out;
3602                 }
3603
3604                 btrfs_release_path(extent_root, path);
3605                 cond_resched();
3606         }
3607         /* reached max tree level, but no tree root found. */
3608         BUG();
3609 out:
3610         btrfs_free_path(path);
3611         return ret;
3612 }
3613
3614 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
3615                                 struct btrfs_root *extent_root,
3616                                 struct btrfs_ref_path *ref_path,
3617                                 u64 extent_start)
3618 {
3619         memset(ref_path, 0, sizeof(*ref_path));
3620         ref_path->extent_start = extent_start;
3621
3622         return __next_ref_path(trans, extent_root, ref_path, 1);
3623 }
3624
3625 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
3626                                struct btrfs_root *extent_root,
3627                                struct btrfs_ref_path *ref_path)
3628 {
3629         return __next_ref_path(trans, extent_root, ref_path, 0);
3630 }
3631
3632 static int noinline get_new_locations(struct inode *reloc_inode,
3633                                       struct btrfs_key *extent_key,
3634                                       u64 offset, int no_fragment,
3635                                       struct disk_extent **extents,
3636                                       int *nr_extents)
3637 {
3638         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3639         struct btrfs_path *path;
3640         struct btrfs_file_extent_item *fi;
3641         struct extent_buffer *leaf;
3642         struct disk_extent *exts = *extents;
3643         struct btrfs_key found_key;
3644         u64 cur_pos;
3645         u64 last_byte;
3646         u32 nritems;
3647         int nr = 0;
3648         int max = *nr_extents;
3649         int ret;
3650
3651         WARN_ON(!no_fragment && *extents);
3652         if (!exts) {
3653                 max = 1;
3654                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
3655                 if (!exts)
3656                         return -ENOMEM;
3657         }
3658
3659         path = btrfs_alloc_path();
3660         BUG_ON(!path);
3661
3662         cur_pos = extent_key->objectid - offset;
3663         last_byte = extent_key->objectid + extent_key->offset;
3664         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
3665                                        cur_pos, 0);
3666         if (ret < 0)
3667                 goto out;
3668         if (ret > 0) {
3669                 ret = -ENOENT;
3670                 goto out;
3671         }
3672
3673         while (1) {
3674                 leaf = path->nodes[0];
3675                 nritems = btrfs_header_nritems(leaf);
3676                 if (path->slots[0] >= nritems) {
3677                         ret = btrfs_next_leaf(root, path);
3678                         if (ret < 0)
3679                                 goto out;
3680                         if (ret > 0)
3681                                 break;
3682                         leaf = path->nodes[0];
3683                 }
3684
3685                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3686                 if (found_key.offset != cur_pos ||
3687                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
3688                     found_key.objectid != reloc_inode->i_ino)
3689                         break;
3690
3691                 fi = btrfs_item_ptr(leaf, path->slots[0],
3692                                     struct btrfs_file_extent_item);
3693                 if (btrfs_file_extent_type(leaf, fi) !=
3694                     BTRFS_FILE_EXTENT_REG ||
3695                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
3696                         break;
3697
3698                 if (nr == max) {
3699                         struct disk_extent *old = exts;
3700                         max *= 2;
3701                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
3702                         memcpy(exts, old, sizeof(*exts) * nr);
3703                         if (old != *extents)
3704                                 kfree(old);
3705                 }
3706
3707                 exts[nr].disk_bytenr =
3708                         btrfs_file_extent_disk_bytenr(leaf, fi);
3709                 exts[nr].disk_num_bytes =
3710                         btrfs_file_extent_disk_num_bytes(leaf, fi);
3711                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
3712                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3713                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
3714                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
3715                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
3716                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
3717                                                                            fi);
3718                 BUG_ON(exts[nr].offset > 0);
3719                 BUG_ON(exts[nr].compression || exts[nr].encryption);
3720                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
3721
3722                 cur_pos += exts[nr].num_bytes;
3723                 nr++;
3724
3725                 if (cur_pos + offset >= last_byte)
3726                         break;
3727
3728                 if (no_fragment) {
3729                         ret = 1;
3730                         goto out;
3731                 }
3732                 path->slots[0]++;
3733         }
3734
3735         WARN_ON(cur_pos + offset > last_byte);
3736         if (cur_pos + offset < last_byte) {
3737                 ret = -ENOENT;
3738                 goto out;
3739         }
3740         ret = 0;
3741 out:
3742         btrfs_free_path(path);
3743         if (ret) {
3744                 if (exts != *extents)
3745                         kfree(exts);
3746         } else {
3747                 *extents = exts;
3748                 *nr_extents = nr;
3749         }
3750         return ret;
3751 }
3752
3753 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
3754                                         struct btrfs_root *root,
3755                                         struct btrfs_path *path,
3756                                         struct btrfs_key *extent_key,
3757                                         struct btrfs_key *leaf_key,
3758                                         struct btrfs_ref_path *ref_path,
3759                                         struct disk_extent *new_extents,
3760                                         int nr_extents)
3761 {
3762         struct extent_buffer *leaf;
3763         struct btrfs_file_extent_item *fi;
3764         struct inode *inode = NULL;
3765         struct btrfs_key key;
3766         u64 lock_start = 0;
3767         u64 lock_end = 0;
3768         u64 num_bytes;
3769         u64 ext_offset;
3770         u64 first_pos;
3771         u32 nritems;
3772         int nr_scaned = 0;
3773         int extent_locked = 0;
3774         int extent_type;
3775         int ret;
3776
3777         memcpy(&key, leaf_key, sizeof(key));
3778         first_pos = INT_LIMIT(loff_t) - extent_key->offset;
3779         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3780                 if (key.objectid < ref_path->owner_objectid ||
3781                     (key.objectid == ref_path->owner_objectid &&
3782                      key.type < BTRFS_EXTENT_DATA_KEY)) {
3783                         key.objectid = ref_path->owner_objectid;
3784                         key.type = BTRFS_EXTENT_DATA_KEY;
3785                         key.offset = 0;
3786                 }
3787         }
3788
3789         while (1) {
3790                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3791                 if (ret < 0)
3792                         goto out;
3793
3794                 leaf = path->nodes[0];
3795                 nritems = btrfs_header_nritems(leaf);
3796 next:
3797                 if (extent_locked && ret > 0) {
3798                         /*
3799                          * the file extent item was modified by someone
3800                          * before the extent got locked.
3801                          */
3802                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3803                                       lock_end, GFP_NOFS);
3804                         extent_locked = 0;
3805                 }
3806
3807                 if (path->slots[0] >= nritems) {
3808                         if (++nr_scaned > 2)
3809                                 break;
3810
3811                         BUG_ON(extent_locked);
3812                         ret = btrfs_next_leaf(root, path);
3813                         if (ret < 0)
3814                                 goto out;
3815                         if (ret > 0)
3816                                 break;
3817                         leaf = path->nodes[0];
3818                         nritems = btrfs_header_nritems(leaf);
3819                 }
3820
3821                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3822
3823                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3824                         if ((key.objectid > ref_path->owner_objectid) ||
3825                             (key.objectid == ref_path->owner_objectid &&
3826                              key.type > BTRFS_EXTENT_DATA_KEY) ||
3827                             (key.offset >= first_pos + extent_key->offset))
3828                                 break;
3829                 }
3830
3831                 if (inode && key.objectid != inode->i_ino) {
3832                         BUG_ON(extent_locked);
3833                         btrfs_release_path(root, path);
3834                         mutex_unlock(&inode->i_mutex);
3835                         iput(inode);
3836                         inode = NULL;
3837                         continue;
3838                 }
3839
3840                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
3841                         path->slots[0]++;
3842                         ret = 1;
3843                         goto next;
3844                 }
3845                 fi = btrfs_item_ptr(leaf, path->slots[0],
3846                                     struct btrfs_file_extent_item);
3847                 extent_type = btrfs_file_extent_type(leaf, fi);
3848                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
3849                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
3850                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3851                      extent_key->objectid)) {
3852                         path->slots[0]++;
3853                         ret = 1;
3854                         goto next;
3855                 }
3856
3857                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3858                 ext_offset = btrfs_file_extent_offset(leaf, fi);
3859
3860                 if (first_pos > key.offset - ext_offset)
3861                         first_pos = key.offset - ext_offset;
3862
3863                 if (!extent_locked) {
3864                         lock_start = key.offset;
3865                         lock_end = lock_start + num_bytes - 1;
3866                 } else {
3867                         if (lock_start > key.offset ||
3868                             lock_end + 1 < key.offset + num_bytes) {
3869                                 unlock_extent(&BTRFS_I(inode)->io_tree,
3870                                               lock_start, lock_end, GFP_NOFS);
3871                                 extent_locked = 0;
3872                         }
3873                 }
3874
3875                 if (!inode) {
3876                         btrfs_release_path(root, path);
3877
3878                         inode = btrfs_iget_locked(root->fs_info->sb,
3879                                                   key.objectid, root);
3880                         if (inode->i_state & I_NEW) {
3881                                 BTRFS_I(inode)->root = root;
3882                                 BTRFS_I(inode)->location.objectid =
3883                                         key.objectid;
3884                                 BTRFS_I(inode)->location.type =
3885                                         BTRFS_INODE_ITEM_KEY;
3886                                 BTRFS_I(inode)->location.offset = 0;
3887                                 btrfs_read_locked_inode(inode);
3888                                 unlock_new_inode(inode);
3889                         }
3890                         /*
3891                          * some code call btrfs_commit_transaction while
3892                          * holding the i_mutex, so we can't use mutex_lock
3893                          * here.
3894                          */
3895                         if (is_bad_inode(inode) ||
3896                             !mutex_trylock(&inode->i_mutex)) {
3897                                 iput(inode);
3898                                 inode = NULL;
3899                                 key.offset = (u64)-1;
3900                                 goto skip;
3901                         }
3902                 }
3903
3904                 if (!extent_locked) {
3905                         struct btrfs_ordered_extent *ordered;
3906
3907                         btrfs_release_path(root, path);
3908
3909                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3910                                     lock_end, GFP_NOFS);
3911                         ordered = btrfs_lookup_first_ordered_extent(inode,
3912                                                                     lock_end);
3913                         if (ordered &&
3914                             ordered->file_offset <= lock_end &&
3915                             ordered->file_offset + ordered->len > lock_start) {
3916                                 unlock_extent(&BTRFS_I(inode)->io_tree,
3917                                               lock_start, lock_end, GFP_NOFS);
3918                                 btrfs_start_ordered_extent(inode, ordered, 1);
3919                                 btrfs_put_ordered_extent(ordered);
3920                                 key.offset += num_bytes;
3921                                 goto skip;
3922                         }
3923                         if (ordered)
3924                                 btrfs_put_ordered_extent(ordered);
3925
3926                         extent_locked = 1;
3927                         continue;
3928                 }
3929
3930                 if (nr_extents == 1) {
3931                         /* update extent pointer in place */
3932                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
3933                                                 new_extents[0].disk_bytenr);
3934                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3935                                                 new_extents[0].disk_num_bytes);
3936                         btrfs_mark_buffer_dirty(leaf);
3937
3938                         btrfs_drop_extent_cache(inode, key.offset,
3939                                                 key.offset + num_bytes - 1, 0);
3940
3941                         ret = btrfs_inc_extent_ref(trans, root,
3942                                                 new_extents[0].disk_bytenr,
3943                                                 new_extents[0].disk_num_bytes,
3944                                                 leaf->start,
3945                                                 root->root_key.objectid,
3946                                                 trans->transid,
3947                                                 key.objectid);
3948                         BUG_ON(ret);
3949
3950                         ret = btrfs_free_extent(trans, root,
3951                                                 extent_key->objectid,
3952                                                 extent_key->offset,
3953                                                 leaf->start,
3954                                                 btrfs_header_owner(leaf),
3955                                                 btrfs_header_generation(leaf),
3956                                                 key.objectid, 0);
3957                         BUG_ON(ret);
3958
3959                         btrfs_release_path(root, path);
3960                         key.offset += num_bytes;
3961                 } else {
3962                         BUG_ON(1);
3963 #if 0
3964                         u64 alloc_hint;
3965                         u64 extent_len;
3966                         int i;
3967                         /*
3968                          * drop old extent pointer at first, then insert the
3969                          * new pointers one bye one
3970                          */
3971                         btrfs_release_path(root, path);
3972                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
3973                                                  key.offset + num_bytes,
3974                                                  key.offset, &alloc_hint);
3975                         BUG_ON(ret);
3976
3977                         for (i = 0; i < nr_extents; i++) {
3978                                 if (ext_offset >= new_extents[i].num_bytes) {
3979                                         ext_offset -= new_extents[i].num_bytes;
3980                                         continue;
3981                                 }
3982                                 extent_len = min(new_extents[i].num_bytes -
3983                                                  ext_offset, num_bytes);
3984
3985                                 ret = btrfs_insert_empty_item(trans, root,
3986                                                               path, &key,
3987                                                               sizeof(*fi));
3988                                 BUG_ON(ret);
3989
3990                                 leaf = path->nodes[0];
3991                                 fi = btrfs_item_ptr(leaf, path->slots[0],
3992                                                 struct btrfs_file_extent_item);
3993                                 btrfs_set_file_extent_generation(leaf, fi,
3994                                                         trans->transid);
3995                                 btrfs_set_file_extent_type(leaf, fi,
3996                                                         BTRFS_FILE_EXTENT_REG);
3997                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3998                                                 new_extents[i].disk_bytenr);
3999                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4000                                                 new_extents[i].disk_num_bytes);
4001                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4002                                                 new_extents[i].ram_bytes);
4003
4004                                 btrfs_set_file_extent_compression(leaf, fi,
4005                                                 new_extents[i].compression);
4006                                 btrfs_set_file_extent_encryption(leaf, fi,
4007                                                 new_extents[i].encryption);
4008                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4009                                                 new_extents[i].other_encoding);
4010
4011                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4012                                                         extent_len);
4013                                 ext_offset += new_extents[i].offset;
4014                                 btrfs_set_file_extent_offset(leaf, fi,
4015                                                         ext_offset);
4016                                 btrfs_mark_buffer_dirty(leaf);
4017
4018                                 btrfs_drop_extent_cache(inode, key.offset,
4019                                                 key.offset + extent_len - 1, 0);
4020
4021                                 ret = btrfs_inc_extent_ref(trans, root,
4022                                                 new_extents[i].disk_bytenr,
4023                                                 new_extents[i].disk_num_bytes,
4024                                                 leaf->start,
4025                                                 root->root_key.objectid,
4026                                                 trans->transid, key.objectid);
4027                                 BUG_ON(ret);
4028                                 btrfs_release_path(root, path);
4029
4030                                 inode_add_bytes(inode, extent_len);
4031
4032                                 ext_offset = 0;
4033                                 num_bytes -= extent_len;
4034                                 key.offset += extent_len;
4035
4036                                 if (num_bytes == 0)
4037                                         break;
4038                         }
4039                         BUG_ON(i >= nr_extents);
4040 #endif
4041                 }
4042
4043                 if (extent_locked) {
4044                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4045                                       lock_end, GFP_NOFS);
4046                         extent_locked = 0;
4047                 }
4048 skip:
4049                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4050                     key.offset >= first_pos + extent_key->offset)
4051                         break;
4052
4053                 cond_resched();
4054         }
4055         ret = 0;
4056 out:
4057         btrfs_release_path(root, path);
4058         if (inode) {
4059                 mutex_unlock(&inode->i_mutex);
4060                 if (extent_locked) {
4061                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4062                                       lock_end, GFP_NOFS);
4063                 }
4064                 iput(inode);
4065         }
4066         return ret;
4067 }
4068
4069 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4070                                struct btrfs_root *root,
4071                                struct extent_buffer *buf, u64 orig_start)
4072 {
4073         int level;
4074         int ret;
4075
4076         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4077         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4078
4079         level = btrfs_header_level(buf);
4080         if (level == 0) {
4081                 struct btrfs_leaf_ref *ref;
4082                 struct btrfs_leaf_ref *orig_ref;
4083
4084                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4085                 if (!orig_ref)
4086                         return -ENOENT;
4087
4088                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4089                 if (!ref) {
4090                         btrfs_free_leaf_ref(root, orig_ref);
4091                         return -ENOMEM;
4092                 }
4093
4094                 ref->nritems = orig_ref->nritems;
4095                 memcpy(ref->extents, orig_ref->extents,
4096                         sizeof(ref->extents[0]) * ref->nritems);
4097
4098                 btrfs_free_leaf_ref(root, orig_ref);
4099
4100                 ref->root_gen = trans->transid;
4101                 ref->bytenr = buf->start;
4102                 ref->owner = btrfs_header_owner(buf);
4103                 ref->generation = btrfs_header_generation(buf);
4104                 ret = btrfs_add_leaf_ref(root, ref, 0);
4105                 WARN_ON(ret);
4106                 btrfs_free_leaf_ref(root, ref);
4107         }
4108         return 0;
4109 }
4110
4111 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4112                                         struct extent_buffer *leaf,
4113                                         struct btrfs_block_group_cache *group,
4114                                         struct btrfs_root *target_root)
4115 {
4116         struct btrfs_key key;
4117         struct inode *inode = NULL;
4118         struct btrfs_file_extent_item *fi;
4119         u64 num_bytes;
4120         u64 skip_objectid = 0;
4121         u32 nritems;
4122         u32 i;
4123
4124         nritems = btrfs_header_nritems(leaf);
4125         for (i = 0; i < nritems; i++) {
4126                 btrfs_item_key_to_cpu(leaf, &key, i);
4127                 if (key.objectid == skip_objectid ||
4128                     key.type != BTRFS_EXTENT_DATA_KEY)
4129                         continue;
4130                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4131                 if (btrfs_file_extent_type(leaf, fi) ==
4132                     BTRFS_FILE_EXTENT_INLINE)
4133                         continue;
4134                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4135                         continue;
4136                 if (!inode || inode->i_ino != key.objectid) {
4137                         iput(inode);
4138                         inode = btrfs_ilookup(target_root->fs_info->sb,
4139                                               key.objectid, target_root, 1);
4140                 }
4141                 if (!inode) {
4142                         skip_objectid = key.objectid;
4143                         continue;
4144                 }
4145                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4146
4147                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4148                             key.offset + num_bytes - 1, GFP_NOFS);
4149                 btrfs_drop_extent_cache(inode, key.offset,
4150                                         key.offset + num_bytes - 1, 1);
4151                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4152                               key.offset + num_bytes - 1, GFP_NOFS);
4153                 cond_resched();
4154         }
4155         iput(inode);
4156         return 0;
4157 }
4158
4159 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4160                                         struct btrfs_root *root,
4161                                         struct extent_buffer *leaf,
4162                                         struct btrfs_block_group_cache *group,
4163                                         struct inode *reloc_inode)
4164 {
4165         struct btrfs_key key;
4166         struct btrfs_key extent_key;
4167         struct btrfs_file_extent_item *fi;
4168         struct btrfs_leaf_ref *ref;
4169         struct disk_extent *new_extent;
4170         u64 bytenr;
4171         u64 num_bytes;
4172         u32 nritems;
4173         u32 i;
4174         int ext_index;
4175         int nr_extent;
4176         int ret;
4177
4178         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4179         BUG_ON(!new_extent);
4180
4181         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4182         BUG_ON(!ref);
4183
4184         ext_index = -1;
4185         nritems = btrfs_header_nritems(leaf);
4186         for (i = 0; i < nritems; i++) {
4187                 btrfs_item_key_to_cpu(leaf, &key, i);
4188                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4189                         continue;
4190                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4191                 if (btrfs_file_extent_type(leaf, fi) ==
4192                     BTRFS_FILE_EXTENT_INLINE)
4193                         continue;
4194                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4195                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4196                 if (bytenr == 0)
4197                         continue;
4198
4199                 ext_index++;
4200                 if (bytenr >= group->key.objectid + group->key.offset ||
4201                     bytenr + num_bytes <= group->key.objectid)
4202                         continue;
4203
4204                 extent_key.objectid = bytenr;
4205                 extent_key.offset = num_bytes;
4206                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4207                 nr_extent = 1;
4208                 ret = get_new_locations(reloc_inode, &extent_key,
4209                                         group->key.objectid, 1,
4210                                         &new_extent, &nr_extent);
4211                 if (ret > 0)
4212                         continue;
4213                 BUG_ON(ret < 0);
4214
4215                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4216                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4217                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4218                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4219
4220                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4221                                                 new_extent->disk_bytenr);
4222                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4223                                                 new_extent->disk_num_bytes);
4224                 btrfs_mark_buffer_dirty(leaf);
4225
4226                 ret = btrfs_inc_extent_ref(trans, root,
4227                                         new_extent->disk_bytenr,
4228                                         new_extent->disk_num_bytes,
4229                                         leaf->start,
4230                                         root->root_key.objectid,
4231                                         trans->transid, key.objectid);
4232                 BUG_ON(ret);
4233                 ret = btrfs_free_extent(trans, root,
4234                                         bytenr, num_bytes, leaf->start,
4235                                         btrfs_header_owner(leaf),
4236                                         btrfs_header_generation(leaf),
4237                                         key.objectid, 0);
4238                 BUG_ON(ret);
4239                 cond_resched();
4240         }
4241         kfree(new_extent);
4242         BUG_ON(ext_index + 1 != ref->nritems);
4243         btrfs_free_leaf_ref(root, ref);
4244         return 0;
4245 }
4246
4247 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4248                           struct btrfs_root *root)
4249 {
4250         struct btrfs_root *reloc_root;
4251         int ret;
4252
4253         if (root->reloc_root) {
4254                 reloc_root = root->reloc_root;
4255                 root->reloc_root = NULL;
4256                 list_add(&reloc_root->dead_list,
4257                          &root->fs_info->dead_reloc_roots);
4258
4259                 btrfs_set_root_bytenr(&reloc_root->root_item,
4260                                       reloc_root->node->start);
4261                 btrfs_set_root_level(&root->root_item,
4262                                      btrfs_header_level(reloc_root->node));
4263                 memset(&reloc_root->root_item.drop_progress, 0,
4264                         sizeof(struct btrfs_disk_key));
4265                 reloc_root->root_item.drop_level = 0;
4266
4267                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4268                                         &reloc_root->root_key,
4269                                         &reloc_root->root_item);
4270                 BUG_ON(ret);
4271         }
4272         return 0;
4273 }
4274
4275 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4276 {
4277         struct btrfs_trans_handle *trans;
4278         struct btrfs_root *reloc_root;
4279         struct btrfs_root *prev_root = NULL;
4280         struct list_head dead_roots;
4281         int ret;
4282         unsigned long nr;
4283
4284         INIT_LIST_HEAD(&dead_roots);
4285         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4286
4287         while (!list_empty(&dead_roots)) {
4288                 reloc_root = list_entry(dead_roots.prev,
4289                                         struct btrfs_root, dead_list);
4290                 list_del_init(&reloc_root->dead_list);
4291
4292                 BUG_ON(reloc_root->commit_root != NULL);
4293                 while (1) {
4294                         trans = btrfs_join_transaction(root, 1);
4295                         BUG_ON(!trans);
4296
4297                         mutex_lock(&root->fs_info->drop_mutex);
4298                         ret = btrfs_drop_snapshot(trans, reloc_root);
4299                         if (ret != -EAGAIN)
4300                                 break;
4301                         mutex_unlock(&root->fs_info->drop_mutex);
4302
4303                         nr = trans->blocks_used;
4304                         ret = btrfs_end_transaction(trans, root);
4305                         BUG_ON(ret);
4306                         btrfs_btree_balance_dirty(root, nr);
4307                 }
4308
4309                 free_extent_buffer(reloc_root->node);
4310
4311                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4312                                      &reloc_root->root_key);
4313                 BUG_ON(ret);
4314                 mutex_unlock(&root->fs_info->drop_mutex);
4315
4316                 nr = trans->blocks_used;
4317                 ret = btrfs_end_transaction(trans, root);
4318                 BUG_ON(ret);
4319                 btrfs_btree_balance_dirty(root, nr);
4320
4321                 kfree(prev_root);
4322                 prev_root = reloc_root;
4323         }
4324         if (prev_root) {
4325                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4326                 kfree(prev_root);
4327         }
4328         return 0;
4329 }
4330
4331 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4332 {
4333         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4334         return 0;
4335 }
4336
4337 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4338 {
4339         struct btrfs_root *reloc_root;
4340         struct btrfs_trans_handle *trans;
4341         struct btrfs_key location;
4342         int found;
4343         int ret;
4344
4345         mutex_lock(&root->fs_info->tree_reloc_mutex);
4346         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4347         BUG_ON(ret);
4348         found = !list_empty(&root->fs_info->dead_reloc_roots);
4349         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4350
4351         if (found) {
4352                 trans = btrfs_start_transaction(root, 1);
4353                 BUG_ON(!trans);
4354                 ret = btrfs_commit_transaction(trans, root);
4355                 BUG_ON(ret);
4356         }
4357
4358         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4359         location.offset = (u64)-1;
4360         location.type = BTRFS_ROOT_ITEM_KEY;
4361
4362         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4363         BUG_ON(!reloc_root);
4364         btrfs_orphan_cleanup(reloc_root);
4365         return 0;
4366 }
4367
4368 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
4369                                     struct btrfs_root *root)
4370 {
4371         struct btrfs_root *reloc_root;
4372         struct extent_buffer *eb;
4373         struct btrfs_root_item *root_item;
4374         struct btrfs_key root_key;
4375         int ret;
4376
4377         BUG_ON(!root->ref_cows);
4378         if (root->reloc_root)
4379                 return 0;
4380
4381         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4382         BUG_ON(!root_item);
4383
4384         ret = btrfs_copy_root(trans, root, root->commit_root,
4385                               &eb, BTRFS_TREE_RELOC_OBJECTID);
4386         BUG_ON(ret);
4387
4388         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4389         root_key.offset = root->root_key.objectid;
4390         root_key.type = BTRFS_ROOT_ITEM_KEY;
4391
4392         memcpy(root_item, &root->root_item, sizeof(root_item));
4393         btrfs_set_root_refs(root_item, 0);
4394         btrfs_set_root_bytenr(root_item, eb->start);
4395         btrfs_set_root_level(root_item, btrfs_header_level(eb));
4396         btrfs_set_root_generation(root_item, trans->transid);
4397
4398         btrfs_tree_unlock(eb);
4399         free_extent_buffer(eb);
4400
4401         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4402                                 &root_key, root_item);
4403         BUG_ON(ret);
4404         kfree(root_item);
4405
4406         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4407                                                  &root_key);
4408         BUG_ON(!reloc_root);
4409         reloc_root->last_trans = trans->transid;
4410         reloc_root->commit_root = NULL;
4411         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4412
4413         root->reloc_root = reloc_root;
4414         return 0;
4415 }
4416
4417 /*
4418  * Core function of space balance.
4419  *
4420  * The idea is using reloc trees to relocate tree blocks in reference
4421  * counted roots. There is one reloc tree for each subvol, and all
4422  * reloc trees share same root key objectid. Reloc trees are snapshots
4423  * of the latest committed roots of subvols (root->commit_root).
4424  *
4425  * To relocate a tree block referenced by a subvol, there are two steps.
4426  * COW the block through subvol's reloc tree, then update block pointer
4427  * in the subvol to point to the new block. Since all reloc trees share
4428  * same root key objectid, doing special handing for tree blocks owned
4429  * by them is easy. Once a tree block has been COWed in one reloc tree,
4430  * we can use the resulting new block directly when the same block is
4431  * required to COW again through other reloc trees. By this way, relocated
4432  * tree blocks are shared between reloc trees, so they are also shared
4433  * between subvols.
4434  */
4435 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
4436                                       struct btrfs_root *root,
4437                                       struct btrfs_path *path,
4438                                       struct btrfs_key *first_key,
4439                                       struct btrfs_ref_path *ref_path,
4440                                       struct btrfs_block_group_cache *group,
4441                                       struct inode *reloc_inode)
4442 {
4443         struct btrfs_root *reloc_root;
4444         struct extent_buffer *eb = NULL;
4445         struct btrfs_key *keys;
4446         u64 *nodes;
4447         int level;
4448         int shared_level;
4449         int lowest_level = 0;
4450         int ret;
4451
4452         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
4453                 lowest_level = ref_path->owner_objectid;
4454
4455         if (!root->ref_cows) {
4456                 path->lowest_level = lowest_level;
4457                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
4458                 BUG_ON(ret < 0);
4459                 path->lowest_level = 0;
4460                 btrfs_release_path(root, path);
4461                 return 0;
4462         }
4463
4464         mutex_lock(&root->fs_info->tree_reloc_mutex);
4465         ret = init_reloc_tree(trans, root);
4466         BUG_ON(ret);
4467         reloc_root = root->reloc_root;
4468
4469         shared_level = ref_path->shared_level;
4470         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
4471
4472         keys = ref_path->node_keys;
4473         nodes = ref_path->new_nodes;
4474         memset(&keys[shared_level + 1], 0,
4475                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
4476         memset(&nodes[shared_level + 1], 0,
4477                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
4478
4479         if (nodes[lowest_level] == 0) {
4480                 path->lowest_level = lowest_level;
4481                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4482                                         0, 1);
4483                 BUG_ON(ret);
4484                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
4485                         eb = path->nodes[level];
4486                         if (!eb || eb == reloc_root->node)
4487                                 break;
4488                         nodes[level] = eb->start;
4489                         if (level == 0)
4490                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4491                         else
4492                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4493                 }
4494                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4495                         eb = path->nodes[0];
4496                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
4497                                                       group, reloc_inode);
4498                         BUG_ON(ret);
4499                 }
4500                 btrfs_release_path(reloc_root, path);
4501         } else {
4502                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
4503                                        lowest_level);
4504                 BUG_ON(ret);
4505         }
4506
4507         /*
4508          * replace tree blocks in the fs tree with tree blocks in
4509          * the reloc tree.
4510          */
4511         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
4512         BUG_ON(ret < 0);
4513
4514         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4515                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4516                                         0, 0);
4517                 BUG_ON(ret);
4518                 extent_buffer_get(path->nodes[0]);
4519                 eb = path->nodes[0];
4520                 btrfs_release_path(reloc_root, path);
4521                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
4522                 BUG_ON(ret);
4523                 free_extent_buffer(eb);
4524         }
4525
4526         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4527         path->lowest_level = 0;
4528         return 0;
4529 }
4530
4531 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
4532                                         struct btrfs_root *root,
4533                                         struct btrfs_path *path,
4534                                         struct btrfs_key *first_key,
4535                                         struct btrfs_ref_path *ref_path)
4536 {
4537         int ret;
4538
4539         ret = relocate_one_path(trans, root, path, first_key,
4540                                 ref_path, NULL, NULL);
4541         BUG_ON(ret);
4542
4543         if (root == root->fs_info->extent_root)
4544                 btrfs_extent_post_op(trans, root);
4545
4546         return 0;
4547 }
4548
4549 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
4550                                     struct btrfs_root *extent_root,
4551                                     struct btrfs_path *path,
4552                                     struct btrfs_key *extent_key)
4553 {
4554         int ret;
4555
4556         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
4557         if (ret)
4558                 goto out;
4559         ret = btrfs_del_item(trans, extent_root, path);
4560 out:
4561         btrfs_release_path(extent_root, path);
4562         return ret;
4563 }
4564
4565 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
4566                                                 struct btrfs_ref_path *ref_path)
4567 {
4568         struct btrfs_key root_key;
4569
4570         root_key.objectid = ref_path->root_objectid;
4571         root_key.type = BTRFS_ROOT_ITEM_KEY;
4572         if (is_cowonly_root(ref_path->root_objectid))
4573                 root_key.offset = 0;
4574         else
4575                 root_key.offset = (u64)-1;
4576
4577         return btrfs_read_fs_root_no_name(fs_info, &root_key);
4578 }
4579
4580 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
4581                                         struct btrfs_path *path,
4582                                         struct btrfs_key *extent_key,
4583                                         struct btrfs_block_group_cache *group,
4584                                         struct inode *reloc_inode, int pass)
4585 {
4586         struct btrfs_trans_handle *trans;
4587         struct btrfs_root *found_root;
4588         struct btrfs_ref_path *ref_path = NULL;
4589         struct disk_extent *new_extents = NULL;
4590         int nr_extents = 0;
4591         int loops;
4592         int ret;
4593         int level;
4594         struct btrfs_key first_key;
4595         u64 prev_block = 0;
4596
4597
4598         trans = btrfs_start_transaction(extent_root, 1);
4599         BUG_ON(!trans);
4600
4601         if (extent_key->objectid == 0) {
4602                 ret = del_extent_zero(trans, extent_root, path, extent_key);
4603                 goto out;
4604         }
4605
4606         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
4607         if (!ref_path) {
4608                ret = -ENOMEM;
4609                goto out;
4610         }
4611
4612         for (loops = 0; ; loops++) {
4613                 if (loops == 0) {
4614                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
4615                                                    extent_key->objectid);
4616                 } else {
4617                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
4618                 }
4619                 if (ret < 0)
4620                         goto out;
4621                 if (ret > 0)
4622                         break;
4623
4624                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4625                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
4626                         continue;
4627
4628                 found_root = read_ref_root(extent_root->fs_info, ref_path);
4629                 BUG_ON(!found_root);
4630                 /*
4631                  * for reference counted tree, only process reference paths
4632                  * rooted at the latest committed root.
4633                  */
4634                 if (found_root->ref_cows &&
4635                     ref_path->root_generation != found_root->root_key.offset)
4636                         continue;
4637
4638                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4639                         if (pass == 0) {
4640                                 /*
4641                                  * copy data extents to new locations
4642                                  */
4643                                 u64 group_start = group->key.objectid;
4644                                 ret = relocate_data_extent(reloc_inode,
4645                                                            extent_key,
4646                                                            group_start);
4647                                 if (ret < 0)
4648                                         goto out;
4649                                 break;
4650                         }
4651                         level = 0;
4652                 } else {
4653                         level = ref_path->owner_objectid;
4654                 }
4655
4656                 if (prev_block != ref_path->nodes[level]) {
4657                         struct extent_buffer *eb;
4658                         u64 block_start = ref_path->nodes[level];
4659                         u64 block_size = btrfs_level_size(found_root, level);
4660
4661                         eb = read_tree_block(found_root, block_start,
4662                                              block_size, 0);
4663                         btrfs_tree_lock(eb);
4664                         BUG_ON(level != btrfs_header_level(eb));
4665
4666                         if (level == 0)
4667                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
4668                         else
4669                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
4670
4671                         btrfs_tree_unlock(eb);
4672                         free_extent_buffer(eb);
4673                         prev_block = block_start;
4674                 }
4675
4676                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4677                     pass >= 2) {
4678                         /*
4679                          * use fallback method to process the remaining
4680                          * references.
4681                          */
4682                         if (!new_extents) {
4683                                 u64 group_start = group->key.objectid;
4684                                 new_extents = kmalloc(sizeof(*new_extents),
4685                                                       GFP_NOFS);
4686                                 nr_extents = 1;
4687                                 ret = get_new_locations(reloc_inode,
4688                                                         extent_key,
4689                                                         group_start, 1,
4690                                                         &new_extents,
4691                                                         &nr_extents);
4692                                 if (ret)
4693                                         goto out;
4694                         }
4695                         btrfs_record_root_in_trans(found_root);
4696                         ret = replace_one_extent(trans, found_root,
4697                                                 path, extent_key,
4698                                                 &first_key, ref_path,
4699                                                 new_extents, nr_extents);
4700                         if (ret < 0)
4701                                 goto out;
4702                         continue;
4703                 }
4704
4705                 btrfs_record_root_in_trans(found_root);
4706                 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4707                         ret = relocate_tree_block(trans, found_root, path,
4708                                                   &first_key, ref_path);
4709                 } else {
4710                         /*
4711                          * try to update data extent references while
4712                          * keeping metadata shared between snapshots.
4713                          */
4714                         ret = relocate_one_path(trans, found_root, path,
4715                                                 &first_key, ref_path,
4716                                                 group, reloc_inode);
4717                 }
4718                 if (ret < 0)
4719                         goto out;
4720         }
4721         ret = 0;
4722 out:
4723         btrfs_end_transaction(trans, extent_root);
4724         kfree(new_extents);
4725         kfree(ref_path);
4726         return ret;
4727 }
4728
4729 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
4730 {
4731         u64 num_devices;
4732         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
4733                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
4734
4735         num_devices = root->fs_info->fs_devices->num_devices;
4736         if (num_devices == 1) {
4737                 stripped |= BTRFS_BLOCK_GROUP_DUP;
4738                 stripped = flags & ~stripped;
4739
4740                 /* turn raid0 into single device chunks */
4741                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
4742                         return stripped;
4743
4744                 /* turn mirroring into duplication */
4745                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
4746                              BTRFS_BLOCK_GROUP_RAID10))
4747                         return stripped | BTRFS_BLOCK_GROUP_DUP;
4748                 return flags;
4749         } else {
4750                 /* they already had raid on here, just return */
4751                 if (flags & stripped)
4752                         return flags;
4753
4754                 stripped |= BTRFS_BLOCK_GROUP_DUP;
4755                 stripped = flags & ~stripped;
4756
4757                 /* switch duplicated blocks with raid1 */
4758                 if (flags & BTRFS_BLOCK_GROUP_DUP)
4759                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
4760
4761                 /* turn single device chunks into raid0 */
4762                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
4763         }
4764         return flags;
4765 }
4766
4767 int __alloc_chunk_for_shrink(struct btrfs_root *root,
4768                      struct btrfs_block_group_cache *shrink_block_group,
4769                      int force)
4770 {
4771         struct btrfs_trans_handle *trans;
4772         u64 new_alloc_flags;
4773         u64 calc;
4774
4775         spin_lock(&shrink_block_group->lock);
4776         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
4777                 spin_unlock(&shrink_block_group->lock);
4778
4779                 trans = btrfs_start_transaction(root, 1);
4780                 spin_lock(&shrink_block_group->lock);
4781
4782                 new_alloc_flags = update_block_group_flags(root,
4783                                                    shrink_block_group->flags);
4784                 if (new_alloc_flags != shrink_block_group->flags) {
4785                         calc =
4786                              btrfs_block_group_used(&shrink_block_group->item);
4787                 } else {
4788                         calc = shrink_block_group->key.offset;
4789                 }
4790                 spin_unlock(&shrink_block_group->lock);
4791
4792                 do_chunk_alloc(trans, root->fs_info->extent_root,
4793                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
4794
4795                 btrfs_end_transaction(trans, root);
4796         } else
4797                 spin_unlock(&shrink_block_group->lock);
4798         return 0;
4799 }
4800
4801 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4802                                  struct btrfs_root *root,
4803                                  u64 objectid, u64 size)
4804 {
4805         struct btrfs_path *path;
4806         struct btrfs_inode_item *item;
4807         struct extent_buffer *leaf;
4808         int ret;
4809
4810         path = btrfs_alloc_path();
4811         if (!path)
4812                 return -ENOMEM;
4813
4814         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4815         if (ret)
4816                 goto out;
4817
4818         leaf = path->nodes[0];
4819         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4820         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4821         btrfs_set_inode_generation(leaf, item, 1);
4822         btrfs_set_inode_size(leaf, item, size);
4823         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4824         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM |
4825                                           BTRFS_INODE_NOCOMPRESS);
4826         btrfs_mark_buffer_dirty(leaf);
4827         btrfs_release_path(root, path);
4828 out:
4829         btrfs_free_path(path);
4830         return ret;
4831 }
4832
4833 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
4834                                         struct btrfs_block_group_cache *group)
4835 {
4836         struct inode *inode = NULL;
4837         struct btrfs_trans_handle *trans;
4838         struct btrfs_root *root;
4839         struct btrfs_key root_key;
4840         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
4841         int err = 0;
4842
4843         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4844         root_key.type = BTRFS_ROOT_ITEM_KEY;
4845         root_key.offset = (u64)-1;
4846         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
4847         if (IS_ERR(root))
4848                 return ERR_CAST(root);
4849
4850         trans = btrfs_start_transaction(root, 1);
4851         BUG_ON(!trans);
4852
4853         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
4854         if (err)
4855                 goto out;
4856
4857         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
4858         BUG_ON(err);
4859
4860         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
4861                                        group->key.offset, 0, group->key.offset,
4862                                        0, 0, 0);
4863         BUG_ON(err);
4864
4865         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
4866         if (inode->i_state & I_NEW) {
4867                 BTRFS_I(inode)->root = root;
4868                 BTRFS_I(inode)->location.objectid = objectid;
4869                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
4870                 BTRFS_I(inode)->location.offset = 0;
4871                 btrfs_read_locked_inode(inode);
4872                 unlock_new_inode(inode);
4873                 BUG_ON(is_bad_inode(inode));
4874         } else {
4875                 BUG_ON(1);
4876         }
4877
4878         err = btrfs_orphan_add(trans, inode);
4879 out:
4880         btrfs_end_transaction(trans, root);
4881         if (err) {
4882                 if (inode)
4883                         iput(inode);
4884                 inode = ERR_PTR(err);
4885         }
4886         return inode;
4887 }
4888
4889 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
4890 {
4891         struct btrfs_trans_handle *trans;
4892         struct btrfs_path *path;
4893         struct btrfs_fs_info *info = root->fs_info;
4894         struct extent_buffer *leaf;
4895         struct inode *reloc_inode;
4896         struct btrfs_block_group_cache *block_group;
4897         struct btrfs_key key;
4898         u64 skipped;
4899         u64 cur_byte;
4900         u64 total_found;
4901         u32 nritems;
4902         int ret;
4903         int progress;
4904         int pass = 0;
4905
4906         root = root->fs_info->extent_root;
4907
4908         block_group = btrfs_lookup_block_group(info, group_start);
4909         BUG_ON(!block_group);
4910
4911         printk("btrfs relocating block group %llu flags %llu\n",
4912                (unsigned long long)block_group->key.objectid,
4913                (unsigned long long)block_group->flags);
4914
4915         path = btrfs_alloc_path();
4916         BUG_ON(!path);
4917
4918         reloc_inode = create_reloc_inode(info, block_group);
4919         BUG_ON(IS_ERR(reloc_inode));
4920
4921         __alloc_chunk_for_shrink(root, block_group, 1);
4922         block_group->ro = 1;
4923         block_group->space_info->total_bytes -= block_group->key.offset;
4924
4925         btrfs_start_delalloc_inodes(info->tree_root);
4926         btrfs_wait_ordered_extents(info->tree_root, 0);
4927 again:
4928         skipped = 0;
4929         total_found = 0;
4930         progress = 0;
4931         key.objectid = block_group->key.objectid;
4932         key.offset = 0;
4933         key.type = 0;
4934         cur_byte = key.objectid;
4935
4936         trans = btrfs_start_transaction(info->tree_root, 1);
4937         btrfs_commit_transaction(trans, info->tree_root);
4938
4939         mutex_lock(&root->fs_info->cleaner_mutex);
4940         btrfs_clean_old_snapshots(info->tree_root);
4941         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
4942         mutex_unlock(&root->fs_info->cleaner_mutex);
4943
4944         while(1) {
4945                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4946                 if (ret < 0)
4947                         goto out;
4948 next:
4949                 leaf = path->nodes[0];
4950                 nritems = btrfs_header_nritems(leaf);
4951                 if (path->slots[0] >= nritems) {
4952                         ret = btrfs_next_leaf(root, path);
4953                         if (ret < 0)
4954                                 goto out;
4955                         if (ret == 1) {
4956                                 ret = 0;
4957                                 break;
4958                         }
4959                         leaf = path->nodes[0];
4960                         nritems = btrfs_header_nritems(leaf);
4961                 }
4962
4963                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4964
4965                 if (key.objectid >= block_group->key.objectid +
4966                     block_group->key.offset)
4967                         break;
4968
4969                 if (progress && need_resched()) {
4970                         btrfs_release_path(root, path);
4971                         cond_resched();
4972                         progress = 0;
4973                         continue;
4974                 }
4975                 progress = 1;
4976
4977                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
4978                     key.objectid + key.offset <= cur_byte) {
4979                         path->slots[0]++;
4980                         goto next;
4981                 }
4982
4983                 total_found++;
4984                 cur_byte = key.objectid + key.offset;
4985                 btrfs_release_path(root, path);
4986
4987                 __alloc_chunk_for_shrink(root, block_group, 0);
4988                 ret = relocate_one_extent(root, path, &key, block_group,
4989                                           reloc_inode, pass);
4990                 BUG_ON(ret < 0);
4991                 if (ret > 0)
4992                         skipped++;
4993
4994                 key.objectid = cur_byte;
4995                 key.type = 0;
4996                 key.offset = 0;
4997         }
4998
4999         btrfs_release_path(root, path);
5000
5001         if (pass == 0) {
5002                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5003                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5004                 WARN_ON(reloc_inode->i_mapping->nrpages);
5005         }
5006
5007         if (total_found > 0) {
5008                 printk("btrfs found %llu extents in pass %d\n",
5009                        (unsigned long long)total_found, pass);
5010                 pass++;
5011                 if (total_found == skipped && pass > 2) {
5012                         iput(reloc_inode);
5013                         reloc_inode = create_reloc_inode(info, block_group);
5014                         pass = 0;
5015                 }
5016                 goto again;
5017         }
5018
5019         /* delete reloc_inode */
5020         iput(reloc_inode);
5021
5022         /* unpin extents in this range */
5023         trans = btrfs_start_transaction(info->tree_root, 1);
5024         btrfs_commit_transaction(trans, info->tree_root);
5025
5026         spin_lock(&block_group->lock);
5027         WARN_ON(block_group->pinned > 0);
5028         WARN_ON(block_group->reserved > 0);
5029         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5030         spin_unlock(&block_group->lock);
5031         ret = 0;
5032 out:
5033         btrfs_free_path(path);
5034         return ret;
5035 }
5036
5037 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
5038                            struct btrfs_key *key)
5039 {
5040         int ret = 0;
5041         struct btrfs_key found_key;
5042         struct extent_buffer *leaf;
5043         int slot;
5044
5045         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5046         if (ret < 0)
5047                 goto out;
5048
5049         while(1) {
5050                 slot = path->slots[0];
5051                 leaf = path->nodes[0];
5052                 if (slot >= btrfs_header_nritems(leaf)) {
5053                         ret = btrfs_next_leaf(root, path);
5054                         if (ret == 0)
5055                                 continue;
5056                         if (ret < 0)
5057                                 goto out;
5058                         break;
5059                 }
5060                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5061
5062                 if (found_key.objectid >= key->objectid &&
5063                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5064                         ret = 0;
5065                         goto out;
5066                 }
5067                 path->slots[0]++;
5068         }
5069         ret = -ENOENT;
5070 out:
5071         return ret;
5072 }
5073
5074 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5075 {
5076         struct btrfs_block_group_cache *block_group;
5077         struct rb_node *n;
5078
5079         spin_lock(&info->block_group_cache_lock);
5080         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5081                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5082                                        cache_node);
5083                 rb_erase(&block_group->cache_node,
5084                          &info->block_group_cache_tree);
5085                 spin_unlock(&info->block_group_cache_lock);
5086
5087                 btrfs_remove_free_space_cache(block_group);
5088                 down_write(&block_group->space_info->groups_sem);
5089                 list_del(&block_group->list);
5090                 up_write(&block_group->space_info->groups_sem);
5091                 kfree(block_group);
5092
5093                 spin_lock(&info->block_group_cache_lock);
5094         }
5095         spin_unlock(&info->block_group_cache_lock);
5096         return 0;
5097 }
5098
5099 int btrfs_read_block_groups(struct btrfs_root *root)
5100 {
5101         struct btrfs_path *path;
5102         int ret;
5103         struct btrfs_block_group_cache *cache;
5104         struct btrfs_fs_info *info = root->fs_info;
5105         struct btrfs_space_info *space_info;
5106         struct btrfs_key key;
5107         struct btrfs_key found_key;
5108         struct extent_buffer *leaf;
5109
5110         root = info->extent_root;
5111         key.objectid = 0;
5112         key.offset = 0;
5113         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5114         path = btrfs_alloc_path();
5115         if (!path)
5116                 return -ENOMEM;
5117
5118         while(1) {
5119                 ret = find_first_block_group(root, path, &key);
5120                 if (ret > 0) {
5121                         ret = 0;
5122                         goto error;
5123                 }
5124                 if (ret != 0)
5125                         goto error;
5126
5127                 leaf = path->nodes[0];
5128                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5129                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5130                 if (!cache) {
5131                         ret = -ENOMEM;
5132                         break;
5133                 }
5134
5135                 spin_lock_init(&cache->lock);
5136                 mutex_init(&cache->alloc_mutex);
5137                 INIT_LIST_HEAD(&cache->list);
5138                 read_extent_buffer(leaf, &cache->item,
5139                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5140                                    sizeof(cache->item));
5141                 memcpy(&cache->key, &found_key, sizeof(found_key));
5142
5143                 key.objectid = found_key.objectid + found_key.offset;
5144                 btrfs_release_path(root, path);
5145                 cache->flags = btrfs_block_group_flags(&cache->item);
5146
5147                 ret = update_space_info(info, cache->flags, found_key.offset,
5148                                         btrfs_block_group_used(&cache->item),
5149                                         &space_info);
5150                 BUG_ON(ret);
5151                 cache->space_info = space_info;
5152                 down_write(&space_info->groups_sem);
5153                 list_add_tail(&cache->list, &space_info->block_groups);
5154                 up_write(&space_info->groups_sem);
5155
5156                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5157                 BUG_ON(ret);
5158
5159                 set_avail_alloc_bits(root->fs_info, cache->flags);
5160         }
5161         ret = 0;
5162 error:
5163         btrfs_free_path(path);
5164         return ret;
5165 }
5166
5167 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5168                            struct btrfs_root *root, u64 bytes_used,
5169                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5170                            u64 size)
5171 {
5172         int ret;
5173         struct btrfs_root *extent_root;
5174         struct btrfs_block_group_cache *cache;
5175
5176         extent_root = root->fs_info->extent_root;
5177
5178         root->fs_info->last_trans_new_blockgroup = trans->transid;
5179
5180         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5181         if (!cache)
5182                 return -ENOMEM;
5183
5184         cache->key.objectid = chunk_offset;
5185         cache->key.offset = size;
5186         spin_lock_init(&cache->lock);
5187         mutex_init(&cache->alloc_mutex);
5188         INIT_LIST_HEAD(&cache->list);
5189         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5190
5191         btrfs_set_block_group_used(&cache->item, bytes_used);
5192         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5193         cache->flags = type;
5194         btrfs_set_block_group_flags(&cache->item, type);
5195
5196         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5197                                 &cache->space_info);
5198         BUG_ON(ret);
5199         down_write(&cache->space_info->groups_sem);
5200         list_add_tail(&cache->list, &cache->space_info->block_groups);
5201         up_write(&cache->space_info->groups_sem);
5202
5203         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5204         BUG_ON(ret);
5205
5206         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5207                                 sizeof(cache->item));
5208         BUG_ON(ret);
5209
5210         finish_current_insert(trans, extent_root, 0);
5211         ret = del_pending_extents(trans, extent_root, 0);
5212         BUG_ON(ret);
5213         set_avail_alloc_bits(extent_root->fs_info, type);
5214
5215         return 0;
5216 }
5217
5218 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5219                              struct btrfs_root *root, u64 group_start)
5220 {
5221         struct btrfs_path *path;
5222         struct btrfs_block_group_cache *block_group;
5223         struct btrfs_key key;
5224         int ret;
5225
5226         root = root->fs_info->extent_root;
5227
5228         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5229         BUG_ON(!block_group);
5230
5231         memcpy(&key, &block_group->key, sizeof(key));
5232
5233         path = btrfs_alloc_path();
5234         BUG_ON(!path);
5235
5236         btrfs_remove_free_space_cache(block_group);
5237         rb_erase(&block_group->cache_node,
5238                  &root->fs_info->block_group_cache_tree);
5239         down_write(&block_group->space_info->groups_sem);
5240         list_del(&block_group->list);
5241         up_write(&block_group->space_info->groups_sem);
5242
5243         /*
5244         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5245         kfree(shrink_block_group);
5246         */
5247
5248         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5249         if (ret > 0)
5250                 ret = -EIO;
5251         if (ret < 0)
5252                 goto out;
5253
5254         ret = btrfs_del_item(trans, root, path);
5255 out:
5256         btrfs_free_path(path);
5257         return ret;
5258 }