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