fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include "hash.h"
  23 #include "crc32c.h"
  24 #include "ctree.h"
  25 #include "disk-io.h"
  26 #include "print-tree.h"
  27 #include "transaction.h"
  28 #include "volumes.h"
  29 #include "locking.h"
  30 #include "ref-cache.h"
  31
  32 #define PENDING_EXTENT_INSERT 0
  33 #define PENDING_EXTENT_DELETE 1
  34 #define PENDING_BACKREF_UPDATE 2
  35
  36 struct pending_extent_op {
  37         int type;
  38         u64 bytenr;
  39         u64 num_bytes;
  40         u64 parent;
  41         u64 orig_parent;
  42         u64 generation;
  43         u64 orig_generation;
  44         int level;
  45 };
  46
  47 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  48                                  btrfs_root *extent_root, int all);
  49 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  50                                btrfs_root *extent_root, int all);
  51 static struct btrfs_block_group_cache *
  52 __btrfs_find_block_group(struct btrfs_root *root,
  53                          struct btrfs_block_group_cache *hint,
  54                          u64 search_start, int data, int owner);
  55
  56 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  57 {
  58         return (cache->flags & bits) == bits;
  59 }
  60
  61 /*
  62  * this adds the block group to the fs_info rb tree for the block group
  63  * cache
  64  */
  65 int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  66                                 struct btrfs_block_group_cache *block_group)
  67 {
  68         struct rb_node **p;
  69         struct rb_node *parent = NULL;
  70         struct btrfs_block_group_cache *cache;
  71
  72         spin_lock(&info->block_group_cache_lock);
  73         p = &info->block_group_cache_tree.rb_node;
  74
  75         while (*p) {
  76                 parent = *p;
  77                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  78                                  cache_node);
  79                 if (block_group->key.objectid < cache->key.objectid) {
  80                         p = &(*p)->rb_left;
  81                 } else if (block_group->key.objectid > cache->key.objectid) {
  82                         p = &(*p)->rb_right;
  83                 } else {
  84                         spin_unlock(&info->block_group_cache_lock);
  85                         return -EEXIST;
  86                 }
  87         }
  88
  89         rb_link_node(&block_group->cache_node, parent, p);
  90         rb_insert_color(&block_group->cache_node,
  91                         &info->block_group_cache_tree);
  92         spin_unlock(&info->block_group_cache_lock);
  93
  94         return 0;
  95 }
  96
  97 /*
  98  * This will return the block group at or after bytenr if contains is 0, else
  99  * it will return the block group that contains the bytenr
 100  */
 101 static struct btrfs_block_group_cache *
 102 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 103                               int contains)
 104 {
 105         struct btrfs_block_group_cache *cache, *ret = NULL;
 106         struct rb_node *n;
 107         u64 end, start;
 108
 109         spin_lock(&info->block_group_cache_lock);
 110         n = info->block_group_cache_tree.rb_node;
 111
 112         while (n) {
 113                 cache = rb_entry(n, struct btrfs_block_group_cache,
 114                                  cache_node);
 115                 end = cache->key.objectid + cache->key.offset - 1;
 116                 start = cache->key.objectid;
 117
 118                 if (bytenr < start) {
 119                         if (!contains && (!ret || start < ret->key.objectid))
 120                                 ret = cache;
 121                         n = n->rb_left;
 122                 } else if (bytenr > start) {
 123                         if (contains && bytenr <= end) {
 124                                 ret = cache;
 125                                 break;
 126                         }
 127                         n = n->rb_right;
 128                 } else {
 129                         ret = cache;
 130                         break;
 131                 }
 132         }
 133         spin_unlock(&info->block_group_cache_lock);
 134
 135         return ret;
 136 }
 137
 138 /*
 139  * this is only called by cache_block_group, since we could have freed extents
 140  * we need to check the pinned_extents for any extents that can't be used yet
 141  * since their free space will be released as soon as the transaction commits.
 142  */
 143 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 144                               struct btrfs_fs_info *info, u64 start, u64 end)
 145 {
 146         u64 extent_start, extent_end, size;
 147         int ret;
 148
 149         mutex_lock(&info->pinned_mutex);
 150         while (start < end) {
 151                 ret = find_first_extent_bit(&info->pinned_extents, start,
 152                                             &extent_start, &extent_end,
 153                                             EXTENT_DIRTY);
 154                 if (ret)
 155                         break;
 156
 157                 if (extent_start == start) {
 158                         start = extent_end + 1;
 159                 } else if (extent_start > start && extent_start < end) {
 160                         size = extent_start - start;
 161                         ret = btrfs_add_free_space_lock(block_group, start,
 162                                                         size);
 163                         BUG_ON(ret);
 164                         start = extent_end + 1;
 165                 } else {
 166                         break;
 167                 }
 168         }
 169
 170         if (start < end) {
 171                 size = end - start;
 172                 ret = btrfs_add_free_space_lock(block_group, start, size);
 173                 BUG_ON(ret);
 174         }
 175         mutex_unlock(&info->pinned_mutex);
 176
 177         return 0;
 178 }
 179
 180 static int cache_block_group(struct btrfs_root *root,
 181                              struct btrfs_block_group_cache *block_group)
 182 {
 183         struct btrfs_path *path;
 184         int ret = 0;
 185         struct btrfs_key key;
 186         struct extent_buffer *leaf;
 187         int slot;
 188         u64 last = 0;
 189         u64 first_free;
 190         int found = 0;
 191
 192         if (!block_group)
 193                 return 0;
 194
 195         root = root->fs_info->extent_root;
 196
 197         if (block_group->cached)
 198                 return 0;
 199
 200         path = btrfs_alloc_path();
 201         if (!path)
 202                 return -ENOMEM;
 203
 204         path->reada = 2;
 205         /*
 206          * we get into deadlocks with paths held by callers of this function.
 207          * since the alloc_mutex is protecting things right now, just
 208          * skip the locking here
 209          */
 210         path->skip_locking = 1;
 211         first_free = max_t(u64, block_group->key.objectid,
 212                            BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
 213         key.objectid = block_group->key.objectid;
 214         key.offset = 0;
 215         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 216         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 217         if (ret < 0)
 218                 goto err;
 219         ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
 220         if (ret < 0)
 221                 goto err;
 222         if (ret == 0) {
 223                 leaf = path->nodes[0];
 224                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 225                 if (key.objectid + key.offset > first_free)
 226                         first_free = key.objectid + key.offset;
 227         }
 228         while(1) {
 229                 leaf = path->nodes[0];
 230                 slot = path->slots[0];
 231                 if (slot >= btrfs_header_nritems(leaf)) {
 232                         ret = btrfs_next_leaf(root, path);
 233                         if (ret < 0)
 234                                 goto err;
 235                         if (ret == 0)
 236                                 continue;
 237                         else
 238                                 break;
 239                 }
 240                 btrfs_item_key_to_cpu(leaf, &key, slot);
 241                 if (key.objectid < block_group->key.objectid)
 242                         goto next;
 243
 244                 if (key.objectid >= block_group->key.objectid +
 245                     block_group->key.offset)
 246                         break;
 247
 248                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 249                         if (!found) {
 250                                 last = first_free;
 251                                 found = 1;
 252                         }
 253
 254                         add_new_free_space(block_group, root->fs_info, last,
 255                                            key.objectid);
 256
 257                         last = key.objectid + key.offset;
 258                 }
 259 next:
 260                 path->slots[0]++;
 261         }
 262
 263         if (!found)
 264                 last = first_free;
 265
 266         add_new_free_space(block_group, root->fs_info, last,
 267                            block_group->key.objectid +
 268                            block_group->key.offset);
 269
 270         block_group->cached = 1;
 271         ret = 0;
 272 err:
 273         btrfs_free_path(path);
 274         return ret;
 275 }
 276
 277 /*
 278  * return the block group that starts at or after bytenr
 279  */
 280 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 281                                                        btrfs_fs_info *info,
 282                                                          u64 bytenr)
 283 {
 284         struct btrfs_block_group_cache *cache;
 285
 286         cache = block_group_cache_tree_search(info, bytenr, 0);
 287
 288         return cache;
 289 }
 290
 291 /*
 292  * return the block group that contains teh given bytenr
 293  */
 294 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 295                                                          btrfs_fs_info *info,
 296                                                          u64 bytenr)
 297 {
 298         struct btrfs_block_group_cache *cache;
 299
 300         cache = block_group_cache_tree_search(info, bytenr, 1);
 301
 302         return cache;
 303 }
 304
 305 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 306                                                   u64 flags)
 307 {
 308         struct list_head *head = &info->space_info;
 309         struct list_head *cur;
 310         struct btrfs_space_info *found;
 311         list_for_each(cur, head) {
 312                 found = list_entry(cur, struct btrfs_space_info, list);
 313                 if (found->flags == flags)
 314                         return found;
 315         }
 316         return NULL;
 317 }
 318
 319 static u64 div_factor(u64 num, int factor)
 320 {
 321         if (factor == 10)
 322                 return num;
 323         num *= factor;
 324         do_div(num, 10);
 325         return num;
 326 }
 327
 328 static struct btrfs_block_group_cache *
 329 __btrfs_find_block_group(struct btrfs_root *root,
 330                          struct btrfs_block_group_cache *hint,
 331                          u64 search_start, int data, int owner)
 332 {
 333         struct btrfs_block_group_cache *cache;
 334         struct btrfs_block_group_cache *found_group = NULL;
 335         struct btrfs_fs_info *info = root->fs_info;
 336         u64 used;
 337         u64 last = 0;
 338         u64 free_check;
 339         int full_search = 0;
 340         int factor = 10;
 341         int wrapped = 0;
 342
 343         if (data & BTRFS_BLOCK_GROUP_METADATA)
 344                 factor = 9;
 345
 346         if (search_start) {
 347                 struct btrfs_block_group_cache *shint;
 348                 shint = btrfs_lookup_first_block_group(info, search_start);
 349                 if (shint && block_group_bits(shint, data) && !shint->ro) {
 350                         spin_lock(&shint->lock);
 351                         used = btrfs_block_group_used(&shint->item);
 352                         if (used + shint->pinned + shint->reserved <
 353                             div_factor(shint->key.offset, factor)) {
 354                                 spin_unlock(&shint->lock);
 355                                 return shint;
 356                         }
 357                         spin_unlock(&shint->lock);
 358                 }
 359         }
 360         if (hint && !hint->ro && block_group_bits(hint, data)) {
 361                 spin_lock(&hint->lock);
 362                 used = btrfs_block_group_used(&hint->item);
 363                 if (used + hint->pinned + hint->reserved <
 364                     div_factor(hint->key.offset, factor)) {
 365                         spin_unlock(&hint->lock);
 366                         return hint;
 367                 }
 368                 spin_unlock(&hint->lock);
 369                 last = hint->key.objectid + hint->key.offset;
 370         } else {
 371                 if (hint)
 372                         last = max(hint->key.objectid, search_start);
 373                 else
 374                         last = search_start;
 375         }
 376 again:
 377         while (1) {
 378                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 379                 if (!cache)
 380                         break;
 381
 382                 spin_lock(&cache->lock);
 383                 last = cache->key.objectid + cache->key.offset;
 384                 used = btrfs_block_group_used(&cache->item);
 385
 386                 if (!cache->ro && block_group_bits(cache, data)) {
 387                         free_check = div_factor(cache->key.offset, factor);
 388                         if (used + cache->pinned + cache->reserved <
 389                             free_check) {
 390                                 found_group = cache;
 391                                 spin_unlock(&cache->lock);
 392                                 goto found;
 393                         }
 394                 }
 395                 spin_unlock(&cache->lock);
 396                 cond_resched();
 397         }
 398         if (!wrapped) {
 399                 last = search_start;
 400                 wrapped = 1;
 401                 goto again;
 402         }
 403         if (!full_search && factor < 10) {
 404                 last = search_start;
 405                 full_search = 1;
 406                 factor = 10;
 407                 goto again;
 408         }
 409 found:
 410         return found_group;
 411 }
 412
 413 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
 414                                                  struct btrfs_block_group_cache
 415                                                  *hint, u64 search_start,
 416                                                  int data, int owner)
 417 {
 418
 419         struct btrfs_block_group_cache *ret;
 420         ret = __btrfs_find_block_group(root, hint, search_start, data, owner);
 421         return ret;
 422 }
 423
 424 /* simple helper to search for an existing extent at a given offset */
 425 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 426 {
 427         int ret;
 428         struct btrfs_key key;
 429         struct btrfs_path *path;
 430
 431         path = btrfs_alloc_path();
 432         BUG_ON(!path);
 433         key.objectid = start;
 434         key.offset = len;
 435         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 436         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 437                                 0, 0);
 438         btrfs_free_path(path);
 439         return ret;
 440 }
 441
 442 /*
 443  * Back reference rules.  Back refs have three main goals:
 444  *
 445  * 1) differentiate between all holders of references to an extent so that
 446  *    when a reference is dropped we can make sure it was a valid reference
 447  *    before freeing the extent.
 448  *
 449  * 2) Provide enough information to quickly find the holders of an extent
 450  *    if we notice a given block is corrupted or bad.
 451  *
 452  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 453  *    maintenance.  This is actually the same as #2, but with a slightly
 454  *    different use case.
 455  *
 456  * File extents can be referenced by:
 457  *
 458  * - multiple snapshots, subvolumes, or different generations in one subvol
 459  * - different files inside a single subvolume
 460  * - different offsets inside a file (bookend extents in file.c)
 461  *
 462  * The extent ref structure has fields for:
 463  *
 464  * - Objectid of the subvolume root
 465  * - Generation number of the tree holding the reference
 466  * - objectid of the file holding the reference
 467  * - number of references holding by parent node (alway 1 for tree blocks)
 468  *
 469  * Btree leaf may hold multiple references to a file extent. In most cases,
 470  * these references are from same file and the corresponding offsets inside
 471  * the file are close together.
 472  *
 473  * When a file extent is allocated the fields are filled in:
 474  *     (root_key.objectid, trans->transid, inode objectid, 1)
 475  *
 476  * When a leaf is cow'd new references are added for every file extent found
 477  * in the leaf.  It looks similar to the create case, but trans->transid will
 478  * be different when the block is cow'd.
 479  *
 480  *     (root_key.objectid, trans->transid, inode objectid,
 481  *      number of references in the leaf)
 482  *
 483  * When a file extent is removed either during snapshot deletion or
 484  * file truncation, we find the corresponding back reference and check
 485  * the following fields:
 486  *
 487  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 488  *      inode objectid)
 489  *
 490  * Btree extents can be referenced by:
 491  *
 492  * - Different subvolumes
 493  * - Different generations of the same subvolume
 494  *
 495  * When a tree block is created, back references are inserted:
 496  *
 497  * (root->root_key.objectid, trans->transid, level, 1)
 498  *
 499  * When a tree block is cow'd, new back references are added for all the
 500  * blocks it points to. If the tree block isn't in reference counted root,
 501  * the old back references are removed. These new back references are of
 502  * the form (trans->transid will have increased since creation):
 503  *
 504  * (root->root_key.objectid, trans->transid, level, 1)
 505  *
 506  * When a backref is in deleting, the following fields are checked:
 507  *
 508  * if backref was for a tree root:
 509  *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
 510  * else
 511  *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
 512  *
 513  * Back Reference Key composing:
 514  *
 515  * The key objectid corresponds to the first byte in the extent, the key
 516  * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
 517  * byte of parent extent. If a extent is tree root, the key offset is set
 518  * to the key objectid.
 519  */
 520
 521 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 522                                           struct btrfs_root *root,
 523                                           struct btrfs_path *path,
 524                                           u64 bytenr, u64 parent,
 525                                           u64 ref_root, u64 ref_generation,
 526                                           u64 owner_objectid, int del)
 527 {
 528         struct btrfs_key key;
 529         struct btrfs_extent_ref *ref;
 530         struct extent_buffer *leaf;
 531         u64 ref_objectid;
 532         int ret;
 533
 534         key.objectid = bytenr;
 535         key.type = BTRFS_EXTENT_REF_KEY;
 536         key.offset = parent;
 537
 538         ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
 539         if (ret < 0)
 540                 goto out;
 541         if (ret > 0) {
 542                 ret = -ENOENT;
 543                 goto out;
 544         }
 545
 546         leaf = path->nodes[0];
 547         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 548         ref_objectid = btrfs_ref_objectid(leaf, ref);
 549         if (btrfs_ref_root(leaf, ref) != ref_root ||
 550             btrfs_ref_generation(leaf, ref) != ref_generation ||
 551             (ref_objectid != owner_objectid &&
 552              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 553                 ret = -EIO;
 554                 WARN_ON(1);
 555                 goto out;
 556         }
 557         ret = 0;
 558 out:
 559         return ret;
 560 }
 561
 562 static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
 563                                           struct btrfs_root *root,
 564                                           struct btrfs_path *path,
 565                                           u64 bytenr, u64 parent,
 566                                           u64 ref_root, u64 ref_generation,
 567                                           u64 owner_objectid)
 568 {
 569         struct btrfs_key key;
 570         struct extent_buffer *leaf;
 571         struct btrfs_extent_ref *ref;
 572         u32 num_refs;
 573         int ret;
 574
 575         key.objectid = bytenr;
 576         key.type = BTRFS_EXTENT_REF_KEY;
 577         key.offset = parent;
 578
 579         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
 580         if (ret == 0) {
 581                 leaf = path->nodes[0];
 582                 ref = btrfs_item_ptr(leaf, path->slots[0],
 583                                      struct btrfs_extent_ref);
 584                 btrfs_set_ref_root(leaf, ref, ref_root);
 585                 btrfs_set_ref_generation(leaf, ref, ref_generation);
 586                 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
 587                 btrfs_set_ref_num_refs(leaf, ref, 1);
 588         } else if (ret == -EEXIST) {
 589                 u64 existing_owner;
 590                 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
 591                 leaf = path->nodes[0];
 592                 ref = btrfs_item_ptr(leaf, path->slots[0],
 593                                      struct btrfs_extent_ref);
 594                 if (btrfs_ref_root(leaf, ref) != ref_root ||
 595                     btrfs_ref_generation(leaf, ref) != ref_generation) {
 596                         ret = -EIO;
 597                         WARN_ON(1);
 598                         goto out;
 599                 }
 600
 601                 num_refs = btrfs_ref_num_refs(leaf, ref);
 602                 BUG_ON(num_refs == 0);
 603                 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
 604
 605                 existing_owner = btrfs_ref_objectid(leaf, ref);
 606                 if (existing_owner != owner_objectid &&
 607                     existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
 608                         btrfs_set_ref_objectid(leaf, ref,
 609                                         BTRFS_MULTIPLE_OBJECTIDS);
 610                 }
 611                 ret = 0;
 612         } else {
 613                 goto out;
 614         }
 615         btrfs_mark_buffer_dirty(path->nodes[0]);
 616 out:
 617         btrfs_release_path(root, path);
 618         return ret;
 619 }
 620
 621 static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
 622                                           struct btrfs_root *root,
 623                                           struct btrfs_path *path)
 624 {
 625         struct extent_buffer *leaf;
 626         struct btrfs_extent_ref *ref;
 627         u32 num_refs;
 628         int ret = 0;
 629
 630         leaf = path->nodes[0];
 631         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 632         num_refs = btrfs_ref_num_refs(leaf, ref);
 633         BUG_ON(num_refs == 0);
 634         num_refs -= 1;
 635         if (num_refs == 0) {
 636                 ret = btrfs_del_item(trans, root, path);
 637         } else {
 638                 btrfs_set_ref_num_refs(leaf, ref, num_refs);
 639                 btrfs_mark_buffer_dirty(leaf);
 640         }
 641         btrfs_release_path(root, path);
 642         return ret;
 643 }
 644
 645 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
 646                                      struct btrfs_root *root, u64 bytenr,
 647                                      u64 orig_parent, u64 parent,
 648                                      u64 orig_root, u64 ref_root,
 649                                      u64 orig_generation, u64 ref_generation,
 650                                      u64 owner_objectid)
 651 {
 652         int ret;
 653         struct btrfs_root *extent_root = root->fs_info->extent_root;
 654         struct btrfs_path *path;
 655
 656         if (root == root->fs_info->extent_root) {
 657                 struct pending_extent_op *extent_op;
 658                 u64 num_bytes;
 659
 660                 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
 661                 num_bytes = btrfs_level_size(root, (int)owner_objectid);
 662                 mutex_lock(&root->fs_info->extent_ins_mutex);
 663                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
 664                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
 665                         u64 priv;
 666                         ret = get_state_private(&root->fs_info->extent_ins,
 667                                                 bytenr, &priv);
 668                         BUG_ON(ret);
 669                         extent_op = (struct pending_extent_op *)
 670                                                         (unsigned long)priv;
 671                         BUG_ON(extent_op->parent != orig_parent);
 672                         BUG_ON(extent_op->generation != orig_generation);
 673
 674                         extent_op->parent = parent;
 675                         extent_op->generation = ref_generation;
 676                 } else {
 677                         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
 678                         BUG_ON(!extent_op);
 679
 680                         extent_op->type = PENDING_BACKREF_UPDATE;
 681                         extent_op->bytenr = bytenr;
 682                         extent_op->num_bytes = num_bytes;
 683                         extent_op->parent = parent;
 684                         extent_op->orig_parent = orig_parent;
 685                         extent_op->generation = ref_generation;
 686                         extent_op->orig_generation = orig_generation;
 687                         extent_op->level = (int)owner_objectid;
 688
 689                         set_extent_bits(&root->fs_info->extent_ins,
 690                                         bytenr, bytenr + num_bytes - 1,
 691                                         EXTENT_WRITEBACK, GFP_NOFS);
 692                         set_state_private(&root->fs_info->extent_ins,
 693                                           bytenr, (unsigned long)extent_op);
 694                 }
 695                 mutex_unlock(&root->fs_info->extent_ins_mutex);
 696                 return 0;
 697         }
 698
 699         path = btrfs_alloc_path();
 700         if (!path)
 701                 return -ENOMEM;
 702         ret = lookup_extent_backref(trans, extent_root, path,
 703                                     bytenr, orig_parent, orig_root,
 704                                     orig_generation, owner_objectid, 1);
 705         if (ret)
 706                 goto out;
 707         ret = remove_extent_backref(trans, extent_root, path);
 708         if (ret)
 709                 goto out;
 710         ret = insert_extent_backref(trans, extent_root, path, bytenr,
 711                                     parent, ref_root, ref_generation,
 712                                     owner_objectid);
 713         BUG_ON(ret);
 714         finish_current_insert(trans, extent_root, 0);
 715         del_pending_extents(trans, extent_root, 0);
 716 out:
 717         btrfs_free_path(path);
 718         return ret;
 719 }
 720
 721 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
 722                             struct btrfs_root *root, u64 bytenr,
 723                             u64 orig_parent, u64 parent,
 724                             u64 ref_root, u64 ref_generation,
 725                             u64 owner_objectid)
 726 {
 727         int ret;
 728         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
 729             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
 730                 return 0;
 731         ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
 732                                         parent, ref_root, ref_root,
 733                                         ref_generation, ref_generation,
 734                                         owner_objectid);
 735         return ret;
 736 }
 737
 738 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 739                                   struct btrfs_root *root, u64 bytenr,
 740                                   u64 orig_parent, u64 parent,
 741                                   u64 orig_root, u64 ref_root,
 742                                   u64 orig_generation, u64 ref_generation,
 743                                   u64 owner_objectid)
 744 {
 745         struct btrfs_path *path;
 746         int ret;
 747         struct btrfs_key key;
 748         struct extent_buffer *l;
 749         struct btrfs_extent_item *item;
 750         u32 refs;
 751
 752         path = btrfs_alloc_path();
 753         if (!path)
 754                 return -ENOMEM;
 755
 756         path->reada = 1;
 757         key.objectid = bytenr;
 758         key.type = BTRFS_EXTENT_ITEM_KEY;
 759         key.offset = (u64)-1;
 760
 761         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 762                                 0, 1);
 763         if (ret < 0)
 764                 return ret;
 765         BUG_ON(ret == 0 || path->slots[0] == 0);
 766
 767         path->slots[0]--;
 768         l = path->nodes[0];
 769
 770         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
 771         if (key.objectid != bytenr) {
 772                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
 773                 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
 774                 BUG();
 775         }
 776         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
 777
 778         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 779         refs = btrfs_extent_refs(l, item);
 780         btrfs_set_extent_refs(l, item, refs + 1);
 781         btrfs_mark_buffer_dirty(path->nodes[0]);
 782
 783         btrfs_release_path(root->fs_info->extent_root, path);
 784
 785         path->reada = 1;
 786         ret = insert_extent_backref(trans, root->fs_info->extent_root,
 787                                     path, bytenr, parent,
 788                                     ref_root, ref_generation,
 789                                     owner_objectid);
 790         BUG_ON(ret);
 791         finish_current_insert(trans, root->fs_info->extent_root, 0);
 792         del_pending_extents(trans, root->fs_info->extent_root, 0);
 793
 794         btrfs_free_path(path);
 795         return 0;
 796 }
 797
 798 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 799                          struct btrfs_root *root,
 800                          u64 bytenr, u64 num_bytes, u64 parent,
 801                          u64 ref_root, u64 ref_generation,
 802                          u64 owner_objectid)
 803 {
 804         int ret;
 805         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
 806             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
 807                 return 0;
 808         ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
 809                                      0, ref_root, 0, ref_generation,
 810                                      owner_objectid);
 811         return ret;
 812 }
 813
 814 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
 815                          struct btrfs_root *root)
 816 {
 817         finish_current_insert(trans, root->fs_info->extent_root, 1);
 818         del_pending_extents(trans, root->fs_info->extent_root, 1);
 819         return 0;
 820 }
 821
 822 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
 823                             struct btrfs_root *root, u64 bytenr,
 824                             u64 num_bytes, u32 *refs)
 825 {
 826         struct btrfs_path *path;
 827         int ret;
 828         struct btrfs_key key;
 829         struct extent_buffer *l;
 830         struct btrfs_extent_item *item;
 831
 832         WARN_ON(num_bytes < root->sectorsize);
 833         path = btrfs_alloc_path();
 834         path->reada = 1;
 835         key.objectid = bytenr;
 836         key.offset = num_bytes;
 837         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 838         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
 839                                 0, 0);
 840         if (ret < 0)
 841                 goto out;
 842         if (ret != 0) {
 843                 btrfs_print_leaf(root, path->nodes[0]);
 844                 printk("failed to find block number %Lu\n", bytenr);
 845                 BUG();
 846         }
 847         l = path->nodes[0];
 848         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
 849         *refs = btrfs_extent_refs(l, item);
 850 out:
 851         btrfs_free_path(path);
 852         return 0;
 853 }
 854
 855 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
 856                           struct btrfs_root *root, u64 bytenr)
 857 {
 858         struct btrfs_root *extent_root = root->fs_info->extent_root;
 859         struct btrfs_path *path;
 860         struct extent_buffer *leaf;
 861         struct btrfs_extent_ref *ref_item;
 862         struct btrfs_key key;
 863         struct btrfs_key found_key;
 864         u64 ref_root;
 865         u64 last_snapshot;
 866         u32 nritems;
 867         int ret;
 868
 869         key.objectid = bytenr;
 870         key.offset = (u64)-1;
 871         key.type = BTRFS_EXTENT_ITEM_KEY;
 872
 873         path = btrfs_alloc_path();
 874         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
 875         if (ret < 0)
 876                 goto out;
 877         BUG_ON(ret == 0);
 878
 879         ret = -ENOENT;
 880         if (path->slots[0] == 0)
 881                 goto out;
 882
 883         path->slots[0]--;
 884         leaf = path->nodes[0];
 885         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 886
 887         if (found_key.objectid != bytenr ||
 888             found_key.type != BTRFS_EXTENT_ITEM_KEY)
 889                 goto out;
 890
 891         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
 892         while (1) {
 893                 leaf = path->nodes[0];
 894                 nritems = btrfs_header_nritems(leaf);
 895                 if (path->slots[0] >= nritems) {
 896                         ret = btrfs_next_leaf(extent_root, path);
 897                         if (ret < 0)
 898                                 goto out;
 899                         if (ret == 0)
 900                                 continue;
 901                         break;
 902                 }
 903                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 904                 if (found_key.objectid != bytenr)
 905                         break;
 906
 907                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
 908                         path->slots[0]++;
 909                         continue;
 910                 }
 911
 912                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
 913                                           struct btrfs_extent_ref);
 914                 ref_root = btrfs_ref_root(leaf, ref_item);
 915                 if (ref_root != root->root_key.objectid &&
 916                     ref_root != BTRFS_TREE_LOG_OBJECTID) {
 917                         ret = 1;
 918                         goto out;
 919                 }
 920                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
 921                         ret = 1;
 922                         goto out;
 923                 }
 924
 925                 path->slots[0]++;
 926         }
 927         ret = 0;
 928 out:
 929         btrfs_free_path(path);
 930         return ret;
 931 }
 932
 933 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 934                     struct extent_buffer *buf, u32 nr_extents)
 935 {
 936         struct btrfs_key key;
 937         struct btrfs_file_extent_item *fi;
 938         u64 root_gen;
 939         u32 nritems;
 940         int i;
 941         int level;
 942         int ret = 0;
 943         int shared = 0;
 944
 945         if (!root->ref_cows)
 946                 return 0;
 947
 948         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
 949                 shared = 0;
 950                 root_gen = root->root_key.offset;
 951         } else {
 952                 shared = 1;
 953                 root_gen = trans->transid - 1;
 954         }
 955
 956         level = btrfs_header_level(buf);
 957         nritems = btrfs_header_nritems(buf);
 958
 959         if (level == 0) {
 960                 struct btrfs_leaf_ref *ref;
 961                 struct btrfs_extent_info *info;
 962
 963                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
 964                 if (!ref) {
 965                         ret = -ENOMEM;
 966                         goto out;
 967                 }
 968
 969                 ref->root_gen = root_gen;
 970                 ref->bytenr = buf->start;
 971                 ref->owner = btrfs_header_owner(buf);
 972                 ref->generation = btrfs_header_generation(buf);
 973                 ref->nritems = nr_extents;
 974                 info = ref->extents;
 975
 976                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
 977                         u64 disk_bytenr;
 978                         btrfs_item_key_to_cpu(buf, &key, i);
 979                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
 980                                 continue;
 981                         fi = btrfs_item_ptr(buf, i,
 982                                             struct btrfs_file_extent_item);
 983                         if (btrfs_file_extent_type(buf, fi) ==
 984                             BTRFS_FILE_EXTENT_INLINE)
 985                                 continue;
 986                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
 987                         if (disk_bytenr == 0)
 988                                 continue;
 989
 990                         info->bytenr = disk_bytenr;
 991                         info->num_bytes =
 992                                 btrfs_file_extent_disk_num_bytes(buf, fi);
 993                         info->objectid = key.objectid;
 994                         info->offset = key.offset;
 995                         info++;
 996                 }
 997
 998                 ret = btrfs_add_leaf_ref(root, ref, shared);
 999                 if (ret == -EEXIST && shared) {
1000                         struct btrfs_leaf_ref *old;
1001                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1002                         BUG_ON(!old);
1003                         btrfs_remove_leaf_ref(root, old);
1004                         btrfs_free_leaf_ref(root, old);
1005                         ret = btrfs_add_leaf_ref(root, ref, shared);
1006                 }
1007                 WARN_ON(ret);
1008                 btrfs_free_leaf_ref(root, ref);
1009         }
1010 out:
1011         return ret;
1012 }
1013
1014 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1015                   struct extent_buffer *orig_buf, struct extent_buffer *buf,
1016                   u32 *nr_extents)
1017 {
1018         u64 bytenr;
1019         u64 ref_root;
1020         u64 orig_root;
1021         u64 ref_generation;
1022         u64 orig_generation;
1023         u32 nritems;
1024         u32 nr_file_extents = 0;
1025         struct btrfs_key key;
1026         struct btrfs_file_extent_item *fi;
1027         int i;
1028         int level;
1029         int ret = 0;
1030         int faili = 0;
1031         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1032                             u64, u64, u64, u64, u64, u64, u64, u64);
1033
1034         ref_root = btrfs_header_owner(buf);
1035         ref_generation = btrfs_header_generation(buf);
1036         orig_root = btrfs_header_owner(orig_buf);
1037         orig_generation = btrfs_header_generation(orig_buf);
1038
1039         nritems = btrfs_header_nritems(buf);
1040         level = btrfs_header_level(buf);
1041
1042         if (root->ref_cows) {
1043                 process_func = __btrfs_inc_extent_ref;
1044         } else {
1045                 if (level == 0 &&
1046                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1047                         goto out;
1048                 if (level != 0 &&
1049                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1050                         goto out;
1051                 process_func = __btrfs_update_extent_ref;
1052         }
1053
1054         for (i = 0; i < nritems; i++) {
1055                 cond_resched();
1056                 if (level == 0) {
1057                         btrfs_item_key_to_cpu(buf, &key, i);
1058                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1059                                 continue;
1060                         fi = btrfs_item_ptr(buf, i,
1061                                             struct btrfs_file_extent_item);
1062                         if (btrfs_file_extent_type(buf, fi) ==
1063                             BTRFS_FILE_EXTENT_INLINE)
1064                                 continue;
1065                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1066                         if (bytenr == 0)
1067                                 continue;
1068
1069                         nr_file_extents++;
1070
1071                         ret = process_func(trans, root, bytenr,
1072                                            orig_buf->start, buf->start,
1073                                            orig_root, ref_root,
1074                                            orig_generation, ref_generation,
1075                                            key.objectid);
1076
1077                         if (ret) {
1078                                 faili = i;
1079                                 WARN_ON(1);
1080                                 goto fail;
1081                         }
1082                 } else {
1083                         bytenr = btrfs_node_blockptr(buf, i);
1084                         ret = process_func(trans, root, bytenr,
1085                                            orig_buf->start, buf->start,
1086                                            orig_root, ref_root,
1087                                            orig_generation, ref_generation,
1088                                            level - 1);
1089                         if (ret) {
1090                                 faili = i;
1091                                 WARN_ON(1);
1092                                 goto fail;
1093                         }
1094                 }
1095         }
1096 out:
1097         if (nr_extents) {
1098                 if (level == 0)
1099                         *nr_extents = nr_file_extents;
1100                 else
1101                         *nr_extents = nritems;
1102         }
1103         return 0;
1104 fail:
1105         WARN_ON(1);
1106         return ret;
1107 }
1108
1109 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1110                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1111                      struct extent_buffer *buf, int start_slot, int nr)
1112
1113 {
1114         u64 bytenr;
1115         u64 ref_root;
1116         u64 orig_root;
1117         u64 ref_generation;
1118         u64 orig_generation;
1119         struct btrfs_key key;
1120         struct btrfs_file_extent_item *fi;
1121         int i;
1122         int ret;
1123         int slot;
1124         int level;
1125
1126         BUG_ON(start_slot < 0);
1127         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1128
1129         ref_root = btrfs_header_owner(buf);
1130         ref_generation = btrfs_header_generation(buf);
1131         orig_root = btrfs_header_owner(orig_buf);
1132         orig_generation = btrfs_header_generation(orig_buf);
1133         level = btrfs_header_level(buf);
1134
1135         if (!root->ref_cows) {
1136                 if (level == 0 &&
1137                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1138                         return 0;
1139                 if (level != 0 &&
1140                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1141                         return 0;
1142         }
1143
1144         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1145                 cond_resched();
1146                 if (level == 0) {
1147                         btrfs_item_key_to_cpu(buf, &key, slot);
1148                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1149                                 continue;
1150                         fi = btrfs_item_ptr(buf, slot,
1151                                             struct btrfs_file_extent_item);
1152                         if (btrfs_file_extent_type(buf, fi) ==
1153                             BTRFS_FILE_EXTENT_INLINE)
1154                                 continue;
1155                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1156                         if (bytenr == 0)
1157                                 continue;
1158                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1159                                             orig_buf->start, buf->start,
1160                                             orig_root, ref_root,
1161                                             orig_generation, ref_generation,
1162                                             key.objectid);
1163                         if (ret)
1164                                 goto fail;
1165                 } else {
1166                         bytenr = btrfs_node_blockptr(buf, slot);
1167                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1168                                             orig_buf->start, buf->start,
1169                                             orig_root, ref_root,
1170                                             orig_generation, ref_generation,
1171                                             level - 1);
1172                         if (ret)
1173                                 goto fail;
1174                 }
1175         }
1176         return 0;
1177 fail:
1178         WARN_ON(1);
1179         return -1;
1180 }
1181
1182 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1183                                  struct btrfs_root *root,
1184                                  struct btrfs_path *path,
1185                                  struct btrfs_block_group_cache *cache)
1186 {
1187         int ret;
1188         int pending_ret;
1189         struct btrfs_root *extent_root = root->fs_info->extent_root;
1190         unsigned long bi;
1191         struct extent_buffer *leaf;
1192
1193         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1194         if (ret < 0)
1195                 goto fail;
1196         BUG_ON(ret);
1197
1198         leaf = path->nodes[0];
1199         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1200         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1201         btrfs_mark_buffer_dirty(leaf);
1202         btrfs_release_path(extent_root, path);
1203 fail:
1204         finish_current_insert(trans, extent_root, 0);
1205         pending_ret = del_pending_extents(trans, extent_root, 0);
1206         if (ret)
1207                 return ret;
1208         if (pending_ret)
1209                 return pending_ret;
1210         return 0;
1211
1212 }
1213
1214 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1215                                    struct btrfs_root *root)
1216 {
1217         struct btrfs_block_group_cache *cache, *entry;
1218         struct rb_node *n;
1219         int err = 0;
1220         int werr = 0;
1221         struct btrfs_path *path;
1222         u64 last = 0;
1223
1224         path = btrfs_alloc_path();
1225         if (!path)
1226                 return -ENOMEM;
1227
1228         while(1) {
1229                 cache = NULL;
1230                 spin_lock(&root->fs_info->block_group_cache_lock);
1231                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1232                      n; n = rb_next(n)) {
1233                         entry = rb_entry(n, struct btrfs_block_group_cache,
1234                                          cache_node);
1235                         if (entry->dirty) {
1236                                 cache = entry;
1237                                 break;
1238                         }
1239                 }
1240                 spin_unlock(&root->fs_info->block_group_cache_lock);
1241
1242                 if (!cache)
1243                         break;
1244
1245                 cache->dirty = 0;
1246                 last += cache->key.offset;
1247
1248                 err = write_one_cache_group(trans, root,
1249                                             path, cache);
1250                 /*
1251                  * if we fail to write the cache group, we want
1252                  * to keep it marked dirty in hopes that a later
1253                  * write will work
1254                  */
1255                 if (err) {
1256                         werr = err;
1257                         continue;
1258                 }
1259         }
1260         btrfs_free_path(path);
1261         return werr;
1262 }
1263
1264 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1265                              u64 total_bytes, u64 bytes_used,
1266                              struct btrfs_space_info **space_info)
1267 {
1268         struct btrfs_space_info *found;
1269
1270         found = __find_space_info(info, flags);
1271         if (found) {
1272                 spin_lock(&found->lock);
1273                 found->total_bytes += total_bytes;
1274                 found->bytes_used += bytes_used;
1275                 found->full = 0;
1276                 spin_unlock(&found->lock);
1277                 *space_info = found;
1278                 return 0;
1279         }
1280         found = kmalloc(sizeof(*found), GFP_NOFS);
1281         if (!found)
1282                 return -ENOMEM;
1283
1284         list_add(&found->list, &info->space_info);
1285         INIT_LIST_HEAD(&found->block_groups);
1286         init_rwsem(&found->groups_sem);
1287         spin_lock_init(&found->lock);
1288         found->flags = flags;
1289         found->total_bytes = total_bytes;
1290         found->bytes_used = bytes_used;
1291         found->bytes_pinned = 0;
1292         found->bytes_reserved = 0;
1293         found->full = 0;
1294         found->force_alloc = 0;
1295         *space_info = found;
1296         return 0;
1297 }
1298
1299 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1300 {
1301         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1302                                    BTRFS_BLOCK_GROUP_RAID1 |
1303                                    BTRFS_BLOCK_GROUP_RAID10 |
1304                                    BTRFS_BLOCK_GROUP_DUP);
1305         if (extra_flags) {
1306                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1307                         fs_info->avail_data_alloc_bits |= extra_flags;
1308                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1309                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1310                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1311                         fs_info->avail_system_alloc_bits |= extra_flags;
1312         }
1313 }
1314
1315 static u64 reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1316 {
1317         u64 num_devices = root->fs_info->fs_devices->num_devices;
1318
1319         if (num_devices == 1)
1320                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1321         if (num_devices < 4)
1322                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1323
1324         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1325             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1326                       BTRFS_BLOCK_GROUP_RAID10))) {
1327                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1328         }
1329
1330         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1331             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1332                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1333         }
1334
1335         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1336             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1337              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1338              (flags & BTRFS_BLOCK_GROUP_DUP)))
1339                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1340         return flags;
1341 }
1342
1343 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1344                           struct btrfs_root *extent_root, u64 alloc_bytes,
1345                           u64 flags, int force)
1346 {
1347         struct btrfs_space_info *space_info;
1348         u64 thresh;
1349         u64 start;
1350         u64 num_bytes;
1351         int ret = 0, waited = 0;
1352
1353         flags = reduce_alloc_profile(extent_root, flags);
1354
1355         space_info = __find_space_info(extent_root->fs_info, flags);
1356         if (!space_info) {
1357                 ret = update_space_info(extent_root->fs_info, flags,
1358                                         0, 0, &space_info);
1359                 BUG_ON(ret);
1360         }
1361         BUG_ON(!space_info);
1362
1363         spin_lock(&space_info->lock);
1364         if (space_info->force_alloc) {
1365                 force = 1;
1366                 space_info->force_alloc = 0;
1367         }
1368         if (space_info->full) {
1369                 spin_unlock(&space_info->lock);
1370                 goto out;
1371         }
1372
1373         thresh = div_factor(space_info->total_bytes, 6);
1374         if (!force &&
1375            (space_info->bytes_used + space_info->bytes_pinned +
1376             space_info->bytes_reserved + alloc_bytes) < thresh) {
1377                 spin_unlock(&space_info->lock);
1378                 goto out;
1379         }
1380
1381         spin_unlock(&space_info->lock);
1382
1383         ret = mutex_trylock(&extent_root->fs_info->chunk_mutex);
1384         if (!ret && !force) {
1385                 goto out;
1386         } else if (!ret) {
1387                 mutex_lock(&extent_root->fs_info->chunk_mutex);
1388                 waited = 1;
1389         }
1390
1391         if (waited) {
1392                 spin_lock(&space_info->lock);
1393                 if (space_info->full) {
1394                         spin_unlock(&space_info->lock);
1395                         goto out_unlock;
1396                 }
1397                 spin_unlock(&space_info->lock);
1398         }
1399
1400         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1401         if (ret) {
1402 printk("space info full %Lu\n", flags);
1403                 space_info->full = 1;
1404                 goto out_unlock;
1405         }
1406
1407         ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1408                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1409         BUG_ON(ret);
1410 out_unlock:
1411         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1412 out:
1413         return ret;
1414 }
1415
1416 static int update_block_group(struct btrfs_trans_handle *trans,
1417                               struct btrfs_root *root,
1418                               u64 bytenr, u64 num_bytes, int alloc,
1419                               int mark_free)
1420 {
1421         struct btrfs_block_group_cache *cache;
1422         struct btrfs_fs_info *info = root->fs_info;
1423         u64 total = num_bytes;
1424         u64 old_val;
1425         u64 byte_in_group;
1426
1427         while(total) {
1428                 cache = btrfs_lookup_block_group(info, bytenr);
1429                 if (!cache) {
1430                         return -1;
1431                 }
1432                 byte_in_group = bytenr - cache->key.objectid;
1433                 WARN_ON(byte_in_group > cache->key.offset);
1434
1435                 spin_lock(&cache->space_info->lock);
1436                 spin_lock(&cache->lock);
1437                 cache->dirty = 1;
1438                 old_val = btrfs_block_group_used(&cache->item);
1439                 num_bytes = min(total, cache->key.offset - byte_in_group);
1440                 if (alloc) {
1441                         old_val += num_bytes;
1442                         cache->space_info->bytes_used += num_bytes;
1443                         btrfs_set_block_group_used(&cache->item, old_val);
1444                         spin_unlock(&cache->lock);
1445                         spin_unlock(&cache->space_info->lock);
1446                 } else {
1447                         old_val -= num_bytes;
1448                         cache->space_info->bytes_used -= num_bytes;
1449                         btrfs_set_block_group_used(&cache->item, old_val);
1450                         spin_unlock(&cache->lock);
1451                         spin_unlock(&cache->space_info->lock);
1452                         if (mark_free) {
1453                                 int ret;
1454                                 ret = btrfs_add_free_space(cache, bytenr,
1455                                                            num_bytes);
1456                                 if (ret)
1457                                         return -1;
1458                         }
1459                 }
1460                 total -= num_bytes;
1461                 bytenr += num_bytes;
1462         }
1463         return 0;
1464 }
1465
1466 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1467 {
1468         struct btrfs_block_group_cache *cache;
1469
1470         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1471         if (!cache)
1472                 return 0;
1473
1474         return cache->key.objectid;
1475 }
1476
1477 int btrfs_update_pinned_extents(struct btrfs_root *root,
1478                                 u64 bytenr, u64 num, int pin)
1479 {
1480         u64 len;
1481         struct btrfs_block_group_cache *cache;
1482         struct btrfs_fs_info *fs_info = root->fs_info;
1483
1484         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
1485         if (pin) {
1486                 set_extent_dirty(&fs_info->pinned_extents,
1487                                 bytenr, bytenr + num - 1, GFP_NOFS);
1488         } else {
1489                 clear_extent_dirty(&fs_info->pinned_extents,
1490                                 bytenr, bytenr + num - 1, GFP_NOFS);
1491         }
1492         while (num > 0) {
1493                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1494                 BUG_ON(!cache);
1495                 len = min(num, cache->key.offset -
1496                           (bytenr - cache->key.objectid));
1497                 if (pin) {
1498                         spin_lock(&cache->space_info->lock);
1499                         spin_lock(&cache->lock);
1500                         cache->pinned += len;
1501                         cache->space_info->bytes_pinned += len;
1502                         spin_unlock(&cache->lock);
1503                         spin_unlock(&cache->space_info->lock);
1504                         fs_info->total_pinned += len;
1505                 } else {
1506                         spin_lock(&cache->space_info->lock);
1507                         spin_lock(&cache->lock);
1508                         cache->pinned -= len;
1509                         cache->space_info->bytes_pinned -= len;
1510                         spin_unlock(&cache->lock);
1511                         spin_unlock(&cache->space_info->lock);
1512                         fs_info->total_pinned -= len;
1513                 }
1514                 bytenr += len;
1515                 num -= len;
1516         }
1517         return 0;
1518 }
1519
1520 static int update_reserved_extents(struct btrfs_root *root,
1521                                    u64 bytenr, u64 num, int reserve)
1522 {
1523         u64 len;
1524         struct btrfs_block_group_cache *cache;
1525         struct btrfs_fs_info *fs_info = root->fs_info;
1526
1527         while (num > 0) {
1528                 cache = btrfs_lookup_block_group(fs_info, bytenr);
1529                 BUG_ON(!cache);
1530                 len = min(num, cache->key.offset -
1531                           (bytenr - cache->key.objectid));
1532
1533                 spin_lock(&cache->space_info->lock);
1534                 spin_lock(&cache->lock);
1535                 if (reserve) {
1536                         cache->reserved += len;
1537                         cache->space_info->bytes_reserved += len;
1538                 } else {
1539                         cache->reserved -= len;
1540                         cache->space_info->bytes_reserved -= len;
1541                 }
1542                 spin_unlock(&cache->lock);
1543                 spin_unlock(&cache->space_info->lock);
1544                 bytenr += len;
1545                 num -= len;
1546         }
1547         return 0;
1548 }
1549
1550 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1551 {
1552         u64 last = 0;
1553         u64 start;
1554         u64 end;
1555         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1556         int ret;
1557
1558         mutex_lock(&root->fs_info->pinned_mutex);
1559         while(1) {
1560                 ret = find_first_extent_bit(pinned_extents, last,
1561                                             &start, &end, EXTENT_DIRTY);
1562                 if (ret)
1563                         break;
1564                 set_extent_dirty(copy, start, end, GFP_NOFS);
1565                 last = end + 1;
1566         }
1567         mutex_unlock(&root->fs_info->pinned_mutex);
1568         return 0;
1569 }
1570
1571 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1572                                struct btrfs_root *root,
1573                                struct extent_io_tree *unpin)
1574 {
1575         u64 start;
1576         u64 end;
1577         int ret;
1578         struct btrfs_block_group_cache *cache;
1579
1580         mutex_lock(&root->fs_info->pinned_mutex);
1581         while(1) {
1582                 ret = find_first_extent_bit(unpin, 0, &start, &end,
1583                                             EXTENT_DIRTY);
1584                 if (ret)
1585                         break;
1586                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
1587                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1588                 cache = btrfs_lookup_block_group(root->fs_info, start);
1589                 if (cache->cached)
1590                         btrfs_add_free_space(cache, start, end - start + 1);
1591                 if (need_resched()) {
1592                         mutex_unlock(&root->fs_info->pinned_mutex);
1593                         cond_resched();
1594                         mutex_lock(&root->fs_info->pinned_mutex);
1595                 }
1596         }
1597         mutex_unlock(&root->fs_info->pinned_mutex);
1598         return 0;
1599 }
1600
1601 static int finish_current_insert(struct btrfs_trans_handle *trans,
1602                                  struct btrfs_root *extent_root, int all)
1603 {
1604         u64 start;
1605         u64 end;
1606         u64 priv;
1607         u64 search = 0;
1608         struct btrfs_fs_info *info = extent_root->fs_info;
1609         struct btrfs_path *path;
1610         struct btrfs_extent_ref *ref;
1611         struct pending_extent_op *extent_op;
1612         struct btrfs_key key;
1613         struct btrfs_extent_item extent_item;
1614         int ret;
1615         int err = 0;
1616
1617         btrfs_set_stack_extent_refs(&extent_item, 1);
1618         path = btrfs_alloc_path();
1619
1620         while(1) {
1621                 mutex_lock(&info->extent_ins_mutex);
1622                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
1623                                             &end, EXTENT_WRITEBACK);
1624                 if (ret) {
1625                         mutex_unlock(&info->extent_ins_mutex);
1626                         if (search && all) {
1627                                 search = 0;
1628                                 continue;
1629                         }
1630                         break;
1631                 }
1632
1633                 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
1634                 if (!ret) {
1635                         search = end + 1;
1636                         mutex_unlock(&info->extent_ins_mutex);
1637                         cond_resched();
1638                         continue;
1639                 }
1640                 BUG_ON(ret < 0);
1641
1642                 ret = get_state_private(&info->extent_ins, start, &priv);
1643                 BUG_ON(ret);
1644                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1645
1646                 mutex_unlock(&info->extent_ins_mutex);
1647
1648                 if (extent_op->type == PENDING_EXTENT_INSERT) {
1649                         key.objectid = start;
1650                         key.offset = end + 1 - start;
1651                         key.type = BTRFS_EXTENT_ITEM_KEY;
1652                         err = btrfs_insert_item(trans, extent_root, &key,
1653                                         &extent_item, sizeof(extent_item));
1654                         BUG_ON(err);
1655
1656                         mutex_lock(&info->extent_ins_mutex);
1657                         clear_extent_bits(&info->extent_ins, start, end,
1658                                           EXTENT_WRITEBACK, GFP_NOFS);
1659                         mutex_unlock(&info->extent_ins_mutex);
1660
1661                         err = insert_extent_backref(trans, extent_root, path,
1662                                                 start, extent_op->parent,
1663                                                 extent_root->root_key.objectid,
1664                                                 extent_op->generation,
1665                                                 extent_op->level);
1666                         BUG_ON(err);
1667                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
1668                         err = lookup_extent_backref(trans, extent_root, path,
1669                                                 start, extent_op->orig_parent,
1670                                                 extent_root->root_key.objectid,
1671                                                 extent_op->orig_generation,
1672                                                 extent_op->level, 0);
1673                         BUG_ON(err);
1674
1675                         mutex_lock(&info->extent_ins_mutex);
1676                         clear_extent_bits(&info->extent_ins, start, end,
1677                                           EXTENT_WRITEBACK, GFP_NOFS);
1678                         mutex_unlock(&info->extent_ins_mutex);
1679
1680                         key.objectid = start;
1681                         key.offset = extent_op->parent;
1682                         key.type = BTRFS_EXTENT_REF_KEY;
1683                         err = btrfs_set_item_key_safe(trans, extent_root, path,
1684                                                       &key);
1685                         BUG_ON(err);
1686                         ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
1687                                              struct btrfs_extent_ref);
1688                         btrfs_set_ref_generation(path->nodes[0], ref,
1689                                                  extent_op->generation);
1690                         btrfs_mark_buffer_dirty(path->nodes[0]);
1691                         btrfs_release_path(extent_root, path);
1692                 } else {
1693                         BUG_ON(1);
1694                 }
1695                 kfree(extent_op);
1696                 unlock_extent(&info->extent_ins, start, end, GFP_NOFS);
1697                 if (all)
1698                         search = 0;
1699                 else
1700                         search = end + 1;
1701
1702                 cond_resched();
1703         }
1704         btrfs_free_path(path);
1705         return 0;
1706 }
1707
1708 static int pin_down_bytes(struct btrfs_trans_handle *trans,
1709                           struct btrfs_root *root,
1710                           u64 bytenr, u64 num_bytes, int is_data)
1711 {
1712         int err = 0;
1713         struct extent_buffer *buf;
1714
1715         if (is_data)
1716                 goto pinit;
1717
1718         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1719         if (!buf)
1720                 goto pinit;
1721
1722         /* we can reuse a block if it hasn't been written
1723          * and it is from this transaction.  We can't
1724          * reuse anything from the tree log root because
1725          * it has tiny sub-transactions.
1726          */
1727         if (btrfs_buffer_uptodate(buf, 0) &&
1728             btrfs_try_tree_lock(buf)) {
1729                 u64 header_owner = btrfs_header_owner(buf);
1730                 u64 header_transid = btrfs_header_generation(buf);
1731                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
1732                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
1733                     header_transid == trans->transid &&
1734                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
1735                         clean_tree_block(NULL, root, buf);
1736                         btrfs_tree_unlock(buf);
1737                         free_extent_buffer(buf);
1738                         return 1;
1739                 }
1740                 btrfs_tree_unlock(buf);
1741         }
1742         free_extent_buffer(buf);
1743 pinit:
1744         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
1745
1746         BUG_ON(err < 0);
1747         return 0;
1748 }
1749
1750 /*
1751  * remove an extent from the root, returns 0 on success
1752  */
1753 static int __free_extent(struct btrfs_trans_handle *trans,
1754                          struct btrfs_root *root,
1755                          u64 bytenr, u64 num_bytes, u64 parent,
1756                          u64 root_objectid, u64 ref_generation,
1757                          u64 owner_objectid, int pin, int mark_free)
1758 {
1759         struct btrfs_path *path;
1760         struct btrfs_key key;
1761         struct btrfs_fs_info *info = root->fs_info;
1762         struct btrfs_root *extent_root = info->extent_root;
1763         struct extent_buffer *leaf;
1764         int ret;
1765         int extent_slot = 0;
1766         int found_extent = 0;
1767         int num_to_del = 1;
1768         struct btrfs_extent_item *ei;
1769         u32 refs;
1770
1771         key.objectid = bytenr;
1772         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1773         key.offset = num_bytes;
1774         path = btrfs_alloc_path();
1775         if (!path)
1776                 return -ENOMEM;
1777
1778         path->reada = 1;
1779         ret = lookup_extent_backref(trans, extent_root, path,
1780                                     bytenr, parent, root_objectid,
1781                                     ref_generation, owner_objectid, 1);
1782         if (ret == 0) {
1783                 struct btrfs_key found_key;
1784                 extent_slot = path->slots[0];
1785                 while(extent_slot > 0) {
1786                         extent_slot--;
1787                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1788                                               extent_slot);
1789                         if (found_key.objectid != bytenr)
1790                                 break;
1791                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1792                             found_key.offset == num_bytes) {
1793                                 found_extent = 1;
1794                                 break;
1795                         }
1796                         if (path->slots[0] - extent_slot > 5)
1797                                 break;
1798                 }
1799                 if (!found_extent) {
1800                         ret = remove_extent_backref(trans, extent_root, path);
1801                         BUG_ON(ret);
1802                         btrfs_release_path(extent_root, path);
1803                         ret = btrfs_search_slot(trans, extent_root,
1804                                                 &key, path, -1, 1);
1805                         BUG_ON(ret);
1806                         extent_slot = path->slots[0];
1807                 }
1808         } else {
1809                 btrfs_print_leaf(extent_root, path->nodes[0]);
1810                 WARN_ON(1);
1811                 printk("Unable to find ref byte nr %Lu root %Lu "
1812                        "gen %Lu owner %Lu\n", bytenr,
1813                        root_objectid, ref_generation, owner_objectid);
1814         }
1815
1816         leaf = path->nodes[0];
1817         ei = btrfs_item_ptr(leaf, extent_slot,
1818                             struct btrfs_extent_item);
1819         refs = btrfs_extent_refs(leaf, ei);
1820         BUG_ON(refs == 0);
1821         refs -= 1;
1822         btrfs_set_extent_refs(leaf, ei, refs);
1823
1824         btrfs_mark_buffer_dirty(leaf);
1825
1826         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1827                 struct btrfs_extent_ref *ref;
1828                 ref = btrfs_item_ptr(leaf, path->slots[0],
1829                                      struct btrfs_extent_ref);
1830                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
1831                 /* if the back ref and the extent are next to each other
1832                  * they get deleted below in one shot
1833                  */
1834                 path->slots[0] = extent_slot;
1835                 num_to_del = 2;
1836         } else if (found_extent) {
1837                 /* otherwise delete the extent back ref */
1838                 ret = remove_extent_backref(trans, extent_root, path);
1839                 BUG_ON(ret);
1840                 /* if refs are 0, we need to setup the path for deletion */
1841                 if (refs == 0) {
1842                         btrfs_release_path(extent_root, path);
1843                         ret = btrfs_search_slot(trans, extent_root, &key, path,
1844                                                 -1, 1);
1845                         BUG_ON(ret);
1846                 }
1847         }
1848
1849         if (refs == 0) {
1850                 u64 super_used;
1851                 u64 root_used;
1852 #ifdef BIO_RW_DISCARD
1853                 u64 map_length = num_bytes;
1854                 struct btrfs_multi_bio *multi = NULL;
1855 #endif
1856
1857                 if (pin) {
1858                         mutex_lock(&root->fs_info->pinned_mutex);
1859                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
1860                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
1861                         mutex_unlock(&root->fs_info->pinned_mutex);
1862                         if (ret > 0)
1863                                 mark_free = 1;
1864                         BUG_ON(ret < 0);
1865                 }
1866
1867                 /* block accounting for super block */
1868                 spin_lock_irq(&info->delalloc_lock);
1869                 super_used = btrfs_super_bytes_used(&info->super_copy);
1870                 btrfs_set_super_bytes_used(&info->super_copy,
1871                                            super_used - num_bytes);
1872                 spin_unlock_irq(&info->delalloc_lock);
1873
1874                 /* block accounting for root item */
1875                 root_used = btrfs_root_used(&root->root_item);
1876                 btrfs_set_root_used(&root->root_item,
1877                                            root_used - num_bytes);
1878                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1879                                       num_to_del);
1880                 BUG_ON(ret);
1881                 btrfs_release_path(extent_root, path);
1882                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1883                                          mark_free);
1884                 BUG_ON(ret);
1885
1886 #ifdef BIO_RW_DISCARD
1887                 /* Tell the block device(s) that the sectors can be discarded */
1888                 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
1889                                       bytenr, &map_length, &multi, 0);
1890                 if (!ret) {
1891                         struct btrfs_bio_stripe *stripe = multi->stripes;
1892                         int i;
1893
1894                         if (map_length > num_bytes)
1895                                 map_length = num_bytes;
1896
1897                         for (i = 0; i < multi->num_stripes; i++, stripe++) {
1898                                 blkdev_issue_discard(stripe->dev->bdev,
1899                                                      stripe->physical >> 9,
1900                                                      map_length >> 9);
1901                         }
1902                         kfree(multi);
1903                 }
1904 #endif
1905         }
1906         btrfs_free_path(path);
1907         finish_current_insert(trans, extent_root, 0);
1908         return ret;
1909 }
1910
1911 /*
1912  * find all the blocks marked as pending in the radix tree and remove
1913  * them from the extent map
1914  */
1915 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1916                                btrfs_root *extent_root, int all)
1917 {
1918         int ret;
1919         int err = 0;
1920         u64 start;
1921         u64 end;
1922         u64 priv;
1923         u64 search = 0;
1924         struct extent_io_tree *pending_del;
1925         struct extent_io_tree *extent_ins;
1926         struct pending_extent_op *extent_op;
1927         struct btrfs_fs_info *info = extent_root->fs_info;
1928
1929         extent_ins = &extent_root->fs_info->extent_ins;
1930         pending_del = &extent_root->fs_info->pending_del;
1931
1932         while(1) {
1933                 mutex_lock(&info->extent_ins_mutex);
1934                 ret = find_first_extent_bit(pending_del, search, &start, &end,
1935                                             EXTENT_WRITEBACK);
1936                 if (ret) {
1937                         mutex_unlock(&info->extent_ins_mutex);
1938                         if (all && search) {
1939                                 search = 0;
1940                                 continue;
1941                         }
1942                         break;
1943                 }
1944
1945                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
1946                 if (!ret) {
1947                         search = end+1;
1948                         mutex_unlock(&info->extent_ins_mutex);
1949                         cond_resched();
1950                         continue;
1951                 }
1952                 BUG_ON(ret < 0);
1953
1954                 ret = get_state_private(pending_del, start, &priv);
1955                 BUG_ON(ret);
1956                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1957
1958                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
1959                                   GFP_NOFS);
1960                 if (!test_range_bit(extent_ins, start, end,
1961                                     EXTENT_WRITEBACK, 0)) {
1962                         mutex_unlock(&info->extent_ins_mutex);
1963 free_extent:
1964                         ret = __free_extent(trans, extent_root,
1965                                             start, end + 1 - start,
1966                                             extent_op->orig_parent,
1967                                             extent_root->root_key.objectid,
1968                                             extent_op->orig_generation,
1969                                             extent_op->level, 1, 0);
1970                         kfree(extent_op);
1971                 } else {
1972                         kfree(extent_op);
1973
1974                         ret = get_state_private(&info->extent_ins, start,
1975                                                 &priv);
1976                         BUG_ON(ret);
1977                         extent_op = (struct pending_extent_op *)
1978                                                 (unsigned long)priv;
1979
1980                         clear_extent_bits(&info->extent_ins, start, end,
1981                                           EXTENT_WRITEBACK, GFP_NOFS);
1982
1983                         mutex_unlock(&info->extent_ins_mutex);
1984
1985                         if (extent_op->type == PENDING_BACKREF_UPDATE)
1986                                 goto free_extent;
1987
1988                         mutex_lock(&extent_root->fs_info->pinned_mutex);
1989                         ret = pin_down_bytes(trans, extent_root, start,
1990                                              end + 1 - start, 0);
1991                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
1992
1993                         ret = update_block_group(trans, extent_root, start,
1994                                                 end + 1 - start, 0, ret > 0);
1995
1996                         BUG_ON(ret);
1997                         kfree(extent_op);
1998                 }
1999                 if (ret)
2000                         err = ret;
2001                 unlock_extent(extent_ins, start, end, GFP_NOFS);
2002
2003                 if (all)
2004                         search = 0;
2005                 else
2006                         search = end + 1;
2007                 cond_resched();
2008         }
2009         return err;
2010 }
2011
2012 /*
2013  * remove an extent from the root, returns 0 on success
2014  */
2015 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2016                                struct btrfs_root *root,
2017                                u64 bytenr, u64 num_bytes, u64 parent,
2018                                u64 root_objectid, u64 ref_generation,
2019                                u64 owner_objectid, int pin)
2020 {
2021         struct btrfs_root *extent_root = root->fs_info->extent_root;
2022         int pending_ret;
2023         int ret;
2024
2025         WARN_ON(num_bytes < root->sectorsize);
2026         if (root == extent_root) {
2027                 struct pending_extent_op *extent_op;
2028
2029                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2030                 BUG_ON(!extent_op);
2031
2032                 extent_op->type = PENDING_EXTENT_DELETE;
2033                 extent_op->bytenr = bytenr;
2034                 extent_op->num_bytes = num_bytes;
2035                 extent_op->parent = parent;
2036                 extent_op->orig_parent = parent;
2037                 extent_op->generation = ref_generation;
2038                 extent_op->orig_generation = ref_generation;
2039                 extent_op->level = (int)owner_objectid;
2040
2041                 mutex_lock(&root->fs_info->extent_ins_mutex);
2042                 set_extent_bits(&root->fs_info->pending_del,
2043                                 bytenr, bytenr + num_bytes - 1,
2044                                 EXTENT_WRITEBACK, GFP_NOFS);
2045                 set_state_private(&root->fs_info->pending_del,
2046                                   bytenr, (unsigned long)extent_op);
2047                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2048                 return 0;
2049         }
2050         /* if metadata always pin */
2051         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2052                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2053                         struct btrfs_block_group_cache *cache;
2054
2055                         /* btrfs_free_reserved_extent */
2056                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2057                         BUG_ON(!cache);
2058                         btrfs_add_free_space(cache, bytenr, num_bytes);
2059                         update_reserved_extents(root, bytenr, num_bytes, 0);
2060                         return 0;
2061                 }
2062                 pin = 1;
2063         }
2064
2065         /* if data pin when any transaction has committed this */
2066         if (ref_generation != trans->transid)
2067                 pin = 1;
2068
2069         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2070                             root_objectid, ref_generation,
2071                             owner_objectid, pin, pin == 0);
2072
2073         finish_current_insert(trans, root->fs_info->extent_root, 0);
2074         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2075         return ret ? ret : pending_ret;
2076 }
2077
2078 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2079                       struct btrfs_root *root,
2080                       u64 bytenr, u64 num_bytes, u64 parent,
2081                       u64 root_objectid, u64 ref_generation,
2082                       u64 owner_objectid, int pin)
2083 {
2084         int ret;
2085
2086         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2087                                   root_objectid, ref_generation,
2088                                   owner_objectid, pin);
2089         return ret;
2090 }
2091
2092 static u64 stripe_align(struct btrfs_root *root, u64 val)
2093 {
2094         u64 mask = ((u64)root->stripesize - 1);
2095         u64 ret = (val + mask) & ~mask;
2096         return ret;
2097 }
2098
2099 /*
2100  * walks the btree of allocated extents and find a hole of a given size.
2101  * The key ins is changed to record the hole:
2102  * ins->objectid == block start
2103  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2104  * ins->offset == number of blocks
2105  * Any available blocks before search_start are skipped.
2106  */
2107 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2108                                      struct btrfs_root *orig_root,
2109                                      u64 num_bytes, u64 empty_size,
2110                                      u64 search_start, u64 search_end,
2111                                      u64 hint_byte, struct btrfs_key *ins,
2112                                      u64 exclude_start, u64 exclude_nr,
2113                                      int data)
2114 {
2115         int ret = 0;
2116         struct btrfs_root * root = orig_root->fs_info->extent_root;
2117         u64 total_needed = num_bytes;
2118         u64 *last_ptr = NULL;
2119         struct btrfs_block_group_cache *block_group = NULL;
2120         int chunk_alloc_done = 0;
2121         int empty_cluster = 2 * 1024 * 1024;
2122         int allowed_chunk_alloc = 0;
2123         struct list_head *head = NULL, *cur = NULL;
2124         int loop = 0;
2125         struct btrfs_space_info *space_info;
2126
2127         WARN_ON(num_bytes < root->sectorsize);
2128         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2129         ins->objectid = 0;
2130         ins->offset = 0;
2131
2132         if (orig_root->ref_cows || empty_size)
2133                 allowed_chunk_alloc = 1;
2134
2135         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2136                 last_ptr = &root->fs_info->last_alloc;
2137                 empty_cluster = 256 * 1024;
2138         }
2139
2140         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2141                 last_ptr = &root->fs_info->last_data_alloc;
2142
2143         if (last_ptr) {
2144                 if (*last_ptr)
2145                         hint_byte = *last_ptr;
2146                 else
2147                         empty_size += empty_cluster;
2148         }
2149         search_start = max(search_start, first_logical_byte(root, 0));
2150         search_start = max(search_start, hint_byte);
2151         total_needed += empty_size;
2152
2153         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2154         if (!block_group)
2155                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2156                                                              search_start);
2157         space_info = __find_space_info(root->fs_info, data);
2158
2159         down_read(&space_info->groups_sem);
2160         while (1) {
2161                 struct btrfs_free_space *free_space;
2162                 /*
2163                  * the only way this happens if our hint points to a block
2164                  * group thats not of the proper type, while looping this
2165                  * should never happen
2166                  */
2167                 WARN_ON(!block_group);
2168                 mutex_lock(&block_group->alloc_mutex);
2169                 if (unlikely(!block_group_bits(block_group, data)))
2170                         goto new_group;
2171
2172                 ret = cache_block_group(root, block_group);
2173                 if (ret) {
2174                         mutex_unlock(&block_group->alloc_mutex);
2175                         break;
2176                 }
2177
2178                 if (block_group->ro)
2179                         goto new_group;
2180
2181                 free_space = btrfs_find_free_space(block_group, search_start,
2182                                                    total_needed);
2183                 if (free_space) {
2184                         u64 start = block_group->key.objectid;
2185                         u64 end = block_group->key.objectid +
2186                                 block_group->key.offset;
2187
2188                         search_start = stripe_align(root, free_space->offset);
2189
2190                         /* move on to the next group */
2191                         if (search_start + num_bytes >= search_end)
2192                                 goto new_group;
2193
2194                         /* move on to the next group */
2195                         if (search_start + num_bytes > end)
2196                                 goto new_group;
2197
2198                         if (exclude_nr > 0 &&
2199                             (search_start + num_bytes > exclude_start &&
2200                              search_start < exclude_start + exclude_nr)) {
2201                                 search_start = exclude_start + exclude_nr;
2202                                 /*
2203                                  * if search_start is still in this block group
2204                                  * then we just re-search this block group
2205                                  */
2206                                 if (search_start >= start &&
2207                                     search_start < end) {
2208                                         mutex_unlock(&block_group->alloc_mutex);
2209                                         continue;
2210                                 }
2211
2212                                 /* else we go to the next block group */
2213                                 goto new_group;
2214                         }
2215
2216                         ins->objectid = search_start;
2217                         ins->offset = num_bytes;
2218
2219                         btrfs_remove_free_space_lock(block_group, search_start,
2220                                                      num_bytes);
2221                         /* we are all good, lets return */
2222                         mutex_unlock(&block_group->alloc_mutex);
2223                         break;
2224                 }
2225 new_group:
2226                 mutex_unlock(&block_group->alloc_mutex);
2227                 /*
2228                  * Here's how this works.
2229                  * loop == 0: we were searching a block group via a hint
2230                  *              and didn't find anything, so we start at
2231                  *              the head of the block groups and keep searching
2232                  * loop == 1: we're searching through all of the block groups
2233                  *              if we hit the head again we have searched
2234                  *              all of the block groups for this space and we
2235                  *              need to try and allocate, if we cant error out.
2236                  * loop == 2: we allocated more space and are looping through
2237                  *              all of the block groups again.
2238                  */
2239                 if (loop == 0) {
2240                         head = &space_info->block_groups;
2241                         cur = head->next;
2242
2243                         if (last_ptr && *last_ptr) {
2244                                 total_needed += empty_cluster;
2245                                 *last_ptr = 0;
2246                         }
2247                         loop++;
2248                 } else if (loop == 1 && cur == head) {
2249                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2250                                 up_read(&space_info->groups_sem);
2251                                 ret = do_chunk_alloc(trans, root, num_bytes +
2252                                                      2 * 1024 * 1024, data, 1);
2253                                 if (ret < 0)
2254                                         break;
2255                                 down_read(&space_info->groups_sem);
2256                                 loop++;
2257                                 head = &space_info->block_groups;
2258                                 cur = head->next;
2259                                 chunk_alloc_done = 1;
2260                         } else if (!allowed_chunk_alloc) {
2261                                 space_info->force_alloc = 1;
2262                                 break;
2263                         } else {
2264                                 break;
2265                         }
2266                 } else if (cur == head) {
2267                         break;
2268                 }
2269
2270                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2271                                          list);
2272                 search_start = block_group->key.objectid;
2273                 cur = cur->next;
2274         }
2275
2276         /* we found what we needed */
2277         if (ins->objectid) {
2278                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2279                         trans->block_group = block_group;
2280
2281                 if (last_ptr)
2282                         *last_ptr = ins->objectid + ins->offset;
2283                 ret = 0;
2284         } else if (!ret) {
2285                 ret = -ENOSPC;
2286         }
2287
2288         up_read(&space_info->groups_sem);
2289         return ret;
2290 }
2291
2292 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
2293 {
2294         struct btrfs_block_group_cache *cache;
2295         struct list_head *l;
2296
2297         printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
2298                info->total_bytes - info->bytes_used - info->bytes_pinned -
2299                info->bytes_reserved, (info->full) ? "" : "not ");
2300
2301         down_read(&info->groups_sem);
2302         list_for_each(l, &info->block_groups) {
2303                 cache = list_entry(l, struct btrfs_block_group_cache, list);
2304                 spin_lock(&cache->lock);
2305                 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
2306                        "%Lu pinned %Lu reserved\n",
2307                        cache->key.objectid, cache->key.offset,
2308                        btrfs_block_group_used(&cache->item),
2309                        cache->pinned, cache->reserved);
2310                 btrfs_dump_free_space(cache, bytes);
2311                 spin_unlock(&cache->lock);
2312         }
2313         up_read(&info->groups_sem);
2314 }
2315
2316 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2317                                   struct btrfs_root *root,
2318                                   u64 num_bytes, u64 min_alloc_size,
2319                                   u64 empty_size, u64 hint_byte,
2320                                   u64 search_end, struct btrfs_key *ins,
2321                                   u64 data)
2322 {
2323         int ret;
2324         u64 search_start = 0;
2325         u64 alloc_profile;
2326         struct btrfs_fs_info *info = root->fs_info;
2327
2328         if (data) {
2329                 alloc_profile = info->avail_data_alloc_bits &
2330                                 info->data_alloc_profile;
2331                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2332         } else if (root == root->fs_info->chunk_root) {
2333                 alloc_profile = info->avail_system_alloc_bits &
2334                                 info->system_alloc_profile;
2335                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2336         } else {
2337                 alloc_profile = info->avail_metadata_alloc_bits &
2338                                 info->metadata_alloc_profile;
2339                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2340         }
2341 again:
2342         data = reduce_alloc_profile(root, data);
2343         /*
2344          * the only place that sets empty_size is btrfs_realloc_node, which
2345          * is not called recursively on allocations
2346          */
2347         if (empty_size || root->ref_cows) {
2348                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2349                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2350                                      2 * 1024 * 1024,
2351                                      BTRFS_BLOCK_GROUP_METADATA |
2352                                      (info->metadata_alloc_profile &
2353                                       info->avail_metadata_alloc_bits), 0);
2354                 }
2355                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2356                                      num_bytes + 2 * 1024 * 1024, data, 0);
2357         }
2358
2359         WARN_ON(num_bytes < root->sectorsize);
2360         ret = find_free_extent(trans, root, num_bytes, empty_size,
2361                                search_start, search_end, hint_byte, ins,
2362                                trans->alloc_exclude_start,
2363                                trans->alloc_exclude_nr, data);
2364
2365         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
2366                 num_bytes = num_bytes >> 1;
2367                 num_bytes = num_bytes & ~(root->sectorsize - 1);
2368                 num_bytes = max(num_bytes, min_alloc_size);
2369                 do_chunk_alloc(trans, root->fs_info->extent_root,
2370                                num_bytes, data, 1);
2371                 goto again;
2372         }
2373         if (ret) {
2374                 struct btrfs_space_info *sinfo;
2375
2376                 sinfo = __find_space_info(root->fs_info, data);
2377                 printk("allocation failed flags %Lu, wanted %Lu\n",
2378                        data, num_bytes);
2379                 dump_space_info(sinfo, num_bytes);
2380                 BUG();
2381         }
2382
2383         return ret;
2384 }
2385
2386 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
2387 {
2388         struct btrfs_block_group_cache *cache;
2389
2390         cache = btrfs_lookup_block_group(root->fs_info, start);
2391         if (!cache) {
2392                 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
2393                 return -ENOSPC;
2394         }
2395         btrfs_add_free_space(cache, start, len);
2396         update_reserved_extents(root, start, len, 0);
2397         return 0;
2398 }
2399
2400 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2401                                   struct btrfs_root *root,
2402                                   u64 num_bytes, u64 min_alloc_size,
2403                                   u64 empty_size, u64 hint_byte,
2404                                   u64 search_end, struct btrfs_key *ins,
2405                                   u64 data)
2406 {
2407         int ret;
2408         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
2409                                      empty_size, hint_byte, search_end, ins,
2410                                      data);
2411         update_reserved_extents(root, ins->objectid, ins->offset, 1);
2412         return ret;
2413 }
2414
2415 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2416                                          struct btrfs_root *root, u64 parent,
2417                                          u64 root_objectid, u64 ref_generation,
2418                                          u64 owner, struct btrfs_key *ins)
2419 {
2420         int ret;
2421         int pending_ret;
2422         u64 super_used;
2423         u64 root_used;
2424         u64 num_bytes = ins->offset;
2425         u32 sizes[2];
2426         struct btrfs_fs_info *info = root->fs_info;
2427         struct btrfs_root *extent_root = info->extent_root;
2428         struct btrfs_extent_item *extent_item;
2429         struct btrfs_extent_ref *ref;
2430         struct btrfs_path *path;
2431         struct btrfs_key keys[2];
2432
2433         if (parent == 0)
2434                 parent = ins->objectid;
2435
2436         /* block accounting for super block */
2437         spin_lock_irq(&info->delalloc_lock);
2438         super_used = btrfs_super_bytes_used(&info->super_copy);
2439         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
2440         spin_unlock_irq(&info->delalloc_lock);
2441
2442         /* block accounting for root item */
2443         root_used = btrfs_root_used(&root->root_item);
2444         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
2445
2446         if (root == extent_root) {
2447                 struct pending_extent_op *extent_op;
2448
2449                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2450                 BUG_ON(!extent_op);
2451
2452                 extent_op->type = PENDING_EXTENT_INSERT;
2453                 extent_op->bytenr = ins->objectid;
2454                 extent_op->num_bytes = ins->offset;
2455                 extent_op->parent = parent;
2456                 extent_op->orig_parent = 0;
2457                 extent_op->generation = ref_generation;
2458                 extent_op->orig_generation = 0;
2459                 extent_op->level = (int)owner;
2460
2461                 mutex_lock(&root->fs_info->extent_ins_mutex);
2462                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2463                                 ins->objectid + ins->offset - 1,
2464                                 EXTENT_WRITEBACK, GFP_NOFS);
2465                 set_state_private(&root->fs_info->extent_ins,
2466                                   ins->objectid, (unsigned long)extent_op);
2467                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2468                 goto update_block;
2469         }
2470
2471         memcpy(&keys[0], ins, sizeof(*ins));
2472         keys[1].objectid = ins->objectid;
2473         keys[1].type = BTRFS_EXTENT_REF_KEY;
2474         keys[1].offset = parent;
2475         sizes[0] = sizeof(*extent_item);
2476         sizes[1] = sizeof(*ref);
2477
2478         path = btrfs_alloc_path();
2479         BUG_ON(!path);
2480
2481         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
2482                                        sizes, 2);
2483         BUG_ON(ret);
2484
2485         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2486                                      struct btrfs_extent_item);
2487         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
2488         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2489                              struct btrfs_extent_ref);
2490
2491         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
2492         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
2493         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
2494         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
2495
2496         btrfs_mark_buffer_dirty(path->nodes[0]);
2497
2498         trans->alloc_exclude_start = 0;
2499         trans->alloc_exclude_nr = 0;
2500         btrfs_free_path(path);
2501         finish_current_insert(trans, extent_root, 0);
2502         pending_ret = del_pending_extents(trans, extent_root, 0);
2503
2504         if (ret)
2505                 goto out;
2506         if (pending_ret) {
2507                 ret = pending_ret;
2508                 goto out;
2509         }
2510
2511 update_block:
2512         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
2513         if (ret) {
2514                 printk("update block group failed for %Lu %Lu\n",
2515                        ins->objectid, ins->offset);
2516                 BUG();
2517         }
2518 out:
2519         return ret;
2520 }
2521
2522 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
2523                                 struct btrfs_root *root, u64 parent,
2524                                 u64 root_objectid, u64 ref_generation,
2525                                 u64 owner, struct btrfs_key *ins)
2526 {
2527         int ret;
2528
2529         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
2530                 return 0;
2531         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2532                                             ref_generation, owner, ins);
2533         update_reserved_extents(root, ins->objectid, ins->offset, 0);
2534         return ret;
2535 }
2536
2537 /*
2538  * this is used by the tree logging recovery code.  It records that
2539  * an extent has been allocated and makes sure to clear the free
2540  * space cache bits as well
2541  */
2542 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
2543                                 struct btrfs_root *root, u64 parent,
2544                                 u64 root_objectid, u64 ref_generation,
2545                                 u64 owner, struct btrfs_key *ins)
2546 {
2547         int ret;
2548         struct btrfs_block_group_cache *block_group;
2549
2550         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
2551         mutex_lock(&block_group->alloc_mutex);
2552         cache_block_group(root, block_group);
2553
2554         ret = btrfs_remove_free_space_lock(block_group, ins->objectid,
2555                                            ins->offset);
2556         mutex_unlock(&block_group->alloc_mutex);
2557         BUG_ON(ret);
2558         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
2559                                             ref_generation, owner, ins);
2560         return ret;
2561 }
2562
2563 /*
2564  * finds a free extent and does all the dirty work required for allocation
2565  * returns the key for the extent through ins, and a tree buffer for
2566  * the first block of the extent through buf.
2567  *
2568  * returns 0 if everything worked, non-zero otherwise.
2569  */
2570 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
2571                        struct btrfs_root *root,
2572                        u64 num_bytes, u64 parent, u64 min_alloc_size,
2573                        u64 root_objectid, u64 ref_generation,
2574                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
2575                        u64 search_end, struct btrfs_key *ins, u64 data)
2576 {
2577         int ret;
2578
2579         ret = __btrfs_reserve_extent(trans, root, num_bytes,
2580                                      min_alloc_size, empty_size, hint_byte,
2581                                      search_end, ins, data);
2582         BUG_ON(ret);
2583         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
2584                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
2585                                         root_objectid, ref_generation,
2586                                         owner_objectid, ins);
2587                 BUG_ON(ret);
2588
2589         } else {
2590                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
2591         }
2592         return ret;
2593 }
2594
2595 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
2596                                             struct btrfs_root *root,
2597                                             u64 bytenr, u32 blocksize)
2598 {
2599         struct extent_buffer *buf;
2600
2601         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
2602         if (!buf)
2603                 return ERR_PTR(-ENOMEM);
2604         btrfs_set_header_generation(buf, trans->transid);
2605         btrfs_tree_lock(buf);
2606         clean_tree_block(trans, root, buf);
2607         btrfs_set_buffer_uptodate(buf);
2608         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2609                 set_extent_dirty(&root->dirty_log_pages, buf->start,
2610                          buf->start + buf->len - 1, GFP_NOFS);
2611         } else {
2612                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
2613                          buf->start + buf->len - 1, GFP_NOFS);
2614         }
2615         trans->blocks_used++;
2616         return buf;
2617 }
2618
2619 /*
2620  * helper function to allocate a block for a given tree
2621  * returns the tree buffer or NULL.
2622  */
2623 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2624                                              struct btrfs_root *root,
2625                                              u32 blocksize, u64 parent,
2626                                              u64 root_objectid,
2627                                              u64 ref_generation,
2628                                              int level,
2629                                              u64 hint,
2630                                              u64 empty_size)
2631 {
2632         struct btrfs_key ins;
2633         int ret;
2634         struct extent_buffer *buf;
2635
2636         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
2637                                  root_objectid, ref_generation, level,
2638                                  empty_size, hint, (u64)-1, &ins, 0);
2639         if (ret) {
2640                 BUG_ON(ret > 0);
2641                 return ERR_PTR(ret);
2642         }
2643
2644         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
2645         return buf;
2646 }
2647
2648 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
2649                         struct btrfs_root *root, struct extent_buffer *leaf)
2650 {
2651         u64 leaf_owner;
2652         u64 leaf_generation;
2653         struct btrfs_key key;
2654         struct btrfs_file_extent_item *fi;
2655         int i;
2656         int nritems;
2657         int ret;
2658
2659         BUG_ON(!btrfs_is_leaf(leaf));
2660         nritems = btrfs_header_nritems(leaf);
2661         leaf_owner = btrfs_header_owner(leaf);
2662         leaf_generation = btrfs_header_generation(leaf);
2663
2664         for (i = 0; i < nritems; i++) {
2665                 u64 disk_bytenr;
2666                 cond_resched();
2667
2668                 btrfs_item_key_to_cpu(leaf, &key, i);
2669                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2670                         continue;
2671                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2672                 if (btrfs_file_extent_type(leaf, fi) ==
2673                     BTRFS_FILE_EXTENT_INLINE)
2674                         continue;
2675                 /*
2676                  * FIXME make sure to insert a trans record that
2677                  * repeats the snapshot del on crash
2678                  */
2679                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2680                 if (disk_bytenr == 0)
2681                         continue;
2682
2683                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
2684                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2685                                 leaf->start, leaf_owner, leaf_generation,
2686                                 key.objectid, 0);
2687                 BUG_ON(ret);
2688
2689                 atomic_inc(&root->fs_info->throttle_gen);
2690                 wake_up(&root->fs_info->transaction_throttle);
2691                 cond_resched();
2692         }
2693         return 0;
2694 }
2695
2696 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
2697                                         struct btrfs_root *root,
2698                                         struct btrfs_leaf_ref *ref)
2699 {
2700         int i;
2701         int ret;
2702         struct btrfs_extent_info *info = ref->extents;
2703
2704         for (i = 0; i < ref->nritems; i++) {
2705                 ret = __btrfs_free_extent(trans, root, info->bytenr,
2706                                           info->num_bytes, ref->bytenr,
2707                                           ref->owner, ref->generation,
2708                                           info->objectid, 0);
2709
2710                 atomic_inc(&root->fs_info->throttle_gen);
2711                 wake_up(&root->fs_info->transaction_throttle);
2712                 cond_resched();
2713
2714                 BUG_ON(ret);
2715                 info++;
2716         }
2717
2718         return 0;
2719 }
2720
2721 int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
2722                               u32 *refs)
2723 {
2724         int ret;
2725
2726         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
2727         BUG_ON(ret);
2728
2729 #if 0 // some debugging code in case we see problems here
2730         /* if the refs count is one, it won't get increased again.  But
2731          * if the ref count is > 1, someone may be decreasing it at
2732          * the same time we are.
2733          */
2734         if (*refs != 1) {
2735                 struct extent_buffer *eb = NULL;
2736                 eb = btrfs_find_create_tree_block(root, start, len);
2737                 if (eb)
2738                         btrfs_tree_lock(eb);
2739
2740                 mutex_lock(&root->fs_info->alloc_mutex);
2741                 ret = lookup_extent_ref(NULL, root, start, len, refs);
2742                 BUG_ON(ret);
2743                 mutex_unlock(&root->fs_info->alloc_mutex);
2744
2745                 if (eb) {
2746                         btrfs_tree_unlock(eb);
2747                         free_extent_buffer(eb);
2748                 }
2749                 if (*refs == 1) {
2750                         printk("block %llu went down to one during drop_snap\n",
2751                                (unsigned long long)start);
2752                 }
2753
2754         }
2755 #endif
2756
2757         cond_resched();
2758         return ret;
2759 }
2760
2761 /*
2762  * helper function for drop_snapshot, this walks down the tree dropping ref
2763  * counts as it goes.
2764  */
2765 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2766                                    struct btrfs_root *root,
2767                                    struct btrfs_path *path, int *level)
2768 {
2769         u64 root_owner;
2770         u64 root_gen;
2771         u64 bytenr;
2772         u64 ptr_gen;
2773         struct extent_buffer *next;
2774         struct extent_buffer *cur;
2775         struct extent_buffer *parent;
2776         struct btrfs_leaf_ref *ref;
2777         u32 blocksize;
2778         int ret;
2779         u32 refs;
2780
2781         WARN_ON(*level < 0);
2782         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2783         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
2784                                 path->nodes[*level]->len, &refs);
2785         BUG_ON(ret);
2786         if (refs > 1)
2787                 goto out;
2788
2789         /*
2790          * walk down to the last node level and free all the leaves
2791          */
2792         while(*level >= 0) {
2793                 WARN_ON(*level < 0);
2794                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2795                 cur = path->nodes[*level];
2796
2797                 if (btrfs_header_level(cur) != *level)
2798                         WARN_ON(1);
2799
2800                 if (path->slots[*level] >=
2801                     btrfs_header_nritems(cur))
2802                         break;
2803                 if (*level == 0) {
2804                         ret = btrfs_drop_leaf_ref(trans, root, cur);
2805                         BUG_ON(ret);
2806                         break;
2807                 }
2808                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2809                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2810                 blocksize = btrfs_level_size(root, *level - 1);
2811
2812                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
2813                 BUG_ON(ret);
2814                 if (refs != 1) {
2815                         parent = path->nodes[*level];
2816                         root_owner = btrfs_header_owner(parent);
2817                         root_gen = btrfs_header_generation(parent);
2818                         path->slots[*level]++;
2819
2820                         ret = __btrfs_free_extent(trans, root, bytenr,
2821                                                 blocksize, parent->start,
2822                                                 root_owner, root_gen,
2823                                                 *level - 1, 1);
2824                         BUG_ON(ret);
2825
2826                         atomic_inc(&root->fs_info->throttle_gen);
2827                         wake_up(&root->fs_info->transaction_throttle);
2828                         cond_resched();
2829
2830                         continue;
2831                 }
2832                 /*
2833                  * at this point, we have a single ref, and since the
2834                  * only place referencing this extent is a dead root
2835                  * the reference count should never go higher.
2836                  * So, we don't need to check it again
2837                  */
2838                 if (*level == 1) {
2839                         ref = btrfs_lookup_leaf_ref(root, bytenr);
2840                         if (ref && ref->generation != ptr_gen) {
2841                                 btrfs_free_leaf_ref(root, ref);
2842                                 ref = NULL;
2843                         }
2844                         if (ref) {
2845                                 ret = cache_drop_leaf_ref(trans, root, ref);
2846                                 BUG_ON(ret);
2847                                 btrfs_remove_leaf_ref(root, ref);
2848                                 btrfs_free_leaf_ref(root, ref);
2849                                 *level = 0;
2850                                 break;
2851                         }
2852                         if (printk_ratelimit()) {
2853                                 printk("leaf ref miss for bytenr %llu\n",
2854                                        (unsigned long long)bytenr);
2855                         }
2856                 }
2857                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2858                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2859                         free_extent_buffer(next);
2860
2861                         next = read_tree_block(root, bytenr, blocksize,
2862                                                ptr_gen);
2863                         cond_resched();
2864 #if 0
2865                         /*
2866                          * this is a debugging check and can go away
2867                          * the ref should never go all the way down to 1
2868                          * at this point
2869                          */
2870                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
2871                                                 &refs);
2872                         BUG_ON(ret);
2873                         WARN_ON(refs != 1);
2874 #endif
2875                 }
2876                 WARN_ON(*level <= 0);
2877                 if (path->nodes[*level-1])
2878                         free_extent_buffer(path->nodes[*level-1]);
2879                 path->nodes[*level-1] = next;
2880                 *level = btrfs_header_level(next);
2881                 path->slots[*level] = 0;
2882                 cond_resched();
2883         }
2884 out:
2885         WARN_ON(*level < 0);
2886         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2887
2888         if (path->nodes[*level] == root->node) {
2889                 parent = path->nodes[*level];
2890                 bytenr = path->nodes[*level]->start;
2891         } else {
2892                 parent = path->nodes[*level + 1];
2893                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
2894         }
2895
2896         blocksize = btrfs_level_size(root, *level);
2897         root_owner = btrfs_header_owner(parent);
2898         root_gen = btrfs_header_generation(parent);
2899
2900         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
2901                                   parent->start, root_owner, root_gen,
2902                                   *level, 1);
2903         free_extent_buffer(path->nodes[*level]);
2904         path->nodes[*level] = NULL;
2905         *level += 1;
2906         BUG_ON(ret);
2907
2908         cond_resched();
2909         return 0;
2910 }
2911
2912 /*
2913  * helper function for drop_subtree, this function is similar to
2914  * walk_down_tree. The main difference is that it checks reference
2915  * counts while tree blocks are locked.
2916  */
2917 static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
2918                                       struct btrfs_root *root,
2919                                       struct btrfs_path *path, int *level)
2920 {
2921         struct extent_buffer *next;
2922         struct extent_buffer *cur;
2923         struct extent_buffer *parent;
2924         u64 bytenr;
2925         u64 ptr_gen;
2926         u32 blocksize;
2927         u32 refs;
2928         int ret;
2929
2930         cur = path->nodes[*level];
2931         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
2932                                       &refs);
2933         BUG_ON(ret);
2934         if (refs > 1)
2935                 goto out;
2936
2937         while (*level >= 0) {
2938                 cur = path->nodes[*level];
2939                 if (*level == 0) {
2940                         ret = btrfs_drop_leaf_ref(trans, root, cur);
2941                         BUG_ON(ret);
2942                         clean_tree_block(trans, root, cur);
2943                         break;
2944                 }
2945                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
2946                         clean_tree_block(trans, root, cur);
2947                         break;
2948                 }
2949
2950                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2951                 blocksize = btrfs_level_size(root, *level - 1);
2952                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2953
2954                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
2955                 btrfs_tree_lock(next);
2956
2957                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2958                                               &refs);
2959                 BUG_ON(ret);
2960                 if (refs > 1) {
2961                         parent = path->nodes[*level];
2962                         ret = btrfs_free_extent(trans, root, bytenr,
2963                                         blocksize, parent->start,
2964                                         btrfs_header_owner(parent),
2965                                         btrfs_header_generation(parent),
2966                                         *level - 1, 1);
2967                         BUG_ON(ret);
2968                         path->slots[*level]++;
2969                         btrfs_tree_unlock(next);
2970                         free_extent_buffer(next);
2971                         continue;
2972                 }
2973
2974                 *level = btrfs_header_level(next);
2975                 path->nodes[*level] = next;
2976                 path->slots[*level] = 0;
2977                 path->locks[*level] = 1;
2978                 cond_resched();
2979         }
2980 out:
2981         parent = path->nodes[*level + 1];
2982         bytenr = path->nodes[*level]->start;
2983         blocksize = path->nodes[*level]->len;
2984
2985         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2986                         parent->start, btrfs_header_owner(parent),
2987                         btrfs_header_generation(parent), *level, 1);
2988         BUG_ON(ret);
2989
2990         if (path->locks[*level]) {
2991                 btrfs_tree_unlock(path->nodes[*level]);
2992                 path->locks[*level] = 0;
2993         }
2994         free_extent_buffer(path->nodes[*level]);
2995         path->nodes[*level] = NULL;
2996         *level += 1;
2997         cond_resched();
2998         return 0;
2999 }
3000
3001 /*
3002  * helper for dropping snapshots.  This walks back up the tree in the path
3003  * to find the first node higher up where we haven't yet gone through
3004  * all the slots
3005  */
3006 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3007                                  struct btrfs_root *root,
3008                                  struct btrfs_path *path,
3009                                  int *level, int max_level)
3010 {
3011         u64 root_owner;
3012         u64 root_gen;
3013         struct btrfs_root_item *root_item = &root->root_item;
3014         int i;
3015         int slot;
3016         int ret;
3017
3018         for (i = *level; i < max_level && path->nodes[i]; i++) {
3019                 slot = path->slots[i];
3020                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3021                         struct extent_buffer *node;
3022                         struct btrfs_disk_key disk_key;
3023                         node = path->nodes[i];
3024                         path->slots[i]++;
3025                         *level = i;
3026                         WARN_ON(*level == 0);
3027                         btrfs_node_key(node, &disk_key, path->slots[i]);
3028                         memcpy(&root_item->drop_progress,
3029                                &disk_key, sizeof(disk_key));
3030                         root_item->drop_level = i;
3031                         return 0;
3032                 } else {
3033                         struct extent_buffer *parent;
3034                         if (path->nodes[*level] == root->node)
3035                                 parent = path->nodes[*level];
3036                         else
3037                                 parent = path->nodes[*level + 1];
3038
3039                         root_owner = btrfs_header_owner(parent);
3040                         root_gen = btrfs_header_generation(parent);
3041
3042                         clean_tree_block(trans, root, path->nodes[*level]);
3043                         ret = btrfs_free_extent(trans, root,
3044                                                 path->nodes[*level]->start,
3045                                                 path->nodes[*level]->len,
3046                                                 parent->start, root_owner,
3047                                                 root_gen, *level, 1);
3048                         BUG_ON(ret);
3049                         if (path->locks[*level]) {
3050                                 btrfs_tree_unlock(path->nodes[*level]);
3051                                 path->locks[*level] = 0;
3052                         }
3053                         free_extent_buffer(path->nodes[*level]);
3054                         path->nodes[*level] = NULL;
3055                         *level = i + 1;
3056                 }
3057         }
3058         return 1;
3059 }
3060
3061 /*
3062  * drop the reference count on the tree rooted at 'snap'.  This traverses
3063  * the tree freeing any blocks that have a ref count of zero after being
3064  * decremented.
3065  */
3066 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3067                         *root)
3068 {
3069         int ret = 0;
3070         int wret;
3071         int level;
3072         struct btrfs_path *path;
3073         int i;
3074         int orig_level;
3075         struct btrfs_root_item *root_item = &root->root_item;
3076
3077         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3078         path = btrfs_alloc_path();
3079         BUG_ON(!path);
3080
3081         level = btrfs_header_level(root->node);
3082         orig_level = level;
3083         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3084                 path->nodes[level] = root->node;
3085                 extent_buffer_get(root->node);
3086                 path->slots[level] = 0;
3087         } else {
3088                 struct btrfs_key key;
3089                 struct btrfs_disk_key found_key;
3090                 struct extent_buffer *node;
3091
3092                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3093                 level = root_item->drop_level;
3094                 path->lowest_level = level;
3095                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3096                 if (wret < 0) {
3097                         ret = wret;
3098                         goto out;
3099                 }
3100                 node = path->nodes[level];
3101                 btrfs_node_key(node, &found_key, path->slots[level]);
3102                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3103                                sizeof(found_key)));
3104                 /*
3105                  * unlock our path, this is safe because only this
3106                  * function is allowed to delete this snapshot
3107                  */
3108                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3109                         if (path->nodes[i] && path->locks[i]) {
3110                                 path->locks[i] = 0;
3111                                 btrfs_tree_unlock(path->nodes[i]);
3112                         }
3113                 }
3114         }
3115         while(1) {
3116                 wret = walk_down_tree(trans, root, path, &level);
3117                 if (wret > 0)
3118                         break;
3119                 if (wret < 0)
3120                         ret = wret;
3121
3122                 wret = walk_up_tree(trans, root, path, &level,
3123                                     BTRFS_MAX_LEVEL);
3124                 if (wret > 0)
3125                         break;
3126                 if (wret < 0)
3127                         ret = wret;
3128                 if (trans->transaction->in_commit) {
3129                         ret = -EAGAIN;
3130                         break;
3131                 }
3132                 atomic_inc(&root->fs_info->throttle_gen);
3133                 wake_up(&root->fs_info->transaction_throttle);
3134         }
3135         for (i = 0; i <= orig_level; i++) {
3136                 if (path->nodes[i]) {
3137                         free_extent_buffer(path->nodes[i]);
3138                         path->nodes[i] = NULL;
3139                 }
3140         }
3141 out:
3142         btrfs_free_path(path);
3143         return ret;
3144 }
3145
3146 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3147                         struct btrfs_root *root,
3148                         struct extent_buffer *node,
3149                         struct extent_buffer *parent)
3150 {
3151         struct btrfs_path *path;
3152         int level;
3153         int parent_level;
3154         int ret = 0;
3155         int wret;
3156
3157         path = btrfs_alloc_path();
3158         BUG_ON(!path);
3159
3160         BUG_ON(!btrfs_tree_locked(parent));
3161         parent_level = btrfs_header_level(parent);
3162         extent_buffer_get(parent);
3163         path->nodes[parent_level] = parent;
3164         path->slots[parent_level] = btrfs_header_nritems(parent);
3165
3166         BUG_ON(!btrfs_tree_locked(node));
3167         level = btrfs_header_level(node);
3168         extent_buffer_get(node);
3169         path->nodes[level] = node;
3170         path->slots[level] = 0;
3171
3172         while (1) {
3173                 wret = walk_down_subtree(trans, root, path, &level);
3174                 if (wret < 0)
3175                         ret = wret;
3176                 if (wret != 0)
3177                         break;
3178
3179                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3180                 if (wret < 0)
3181                         ret = wret;
3182                 if (wret != 0)
3183                         break;
3184         }
3185
3186         btrfs_free_path(path);
3187         return ret;
3188 }
3189
3190 static unsigned long calc_ra(unsigned long start, unsigned long last,
3191                              unsigned long nr)
3192 {
3193         return min(last, start + nr - 1);
3194 }
3195
3196 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3197                                          u64 len)
3198 {
3199         u64 page_start;
3200         u64 page_end;
3201         unsigned long first_index;
3202         unsigned long last_index;
3203         unsigned long i;
3204         struct page *page;
3205         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3206         struct file_ra_state *ra;
3207         struct btrfs_ordered_extent *ordered;
3208         unsigned int total_read = 0;
3209         unsigned int total_dirty = 0;
3210         int ret = 0;
3211
3212         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3213
3214         mutex_lock(&inode->i_mutex);
3215         first_index = start >> PAGE_CACHE_SHIFT;
3216         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3217
3218         /* make sure the dirty trick played by the caller work */
3219         ret = invalidate_inode_pages2_range(inode->i_mapping,
3220                                             first_index, last_index);
3221         if (ret)
3222                 goto out_unlock;
3223
3224         file_ra_state_init(ra, inode->i_mapping);
3225
3226         for (i = first_index ; i <= last_index; i++) {
3227                 if (total_read % ra->ra_pages == 0) {
3228                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3229                                        calc_ra(i, last_index, ra->ra_pages));
3230                 }
3231                 total_read++;
3232 again:
3233                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3234                         BUG_ON(1);
3235                 page = grab_cache_page(inode->i_mapping, i);
3236                 if (!page) {
3237                         ret = -ENOMEM;
3238                         goto out_unlock;
3239                 }
3240                 if (!PageUptodate(page)) {
3241                         btrfs_readpage(NULL, page);
3242                         lock_page(page);
3243                         if (!PageUptodate(page)) {
3244                                 unlock_page(page);
3245                                 page_cache_release(page);
3246                                 ret = -EIO;
3247                                 goto out_unlock;
3248                         }
3249                 }
3250                 wait_on_page_writeback(page);
3251
3252                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3253                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3254                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3255
3256                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3257                 if (ordered) {
3258                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3259                         unlock_page(page);
3260                         page_cache_release(page);
3261                         btrfs_start_ordered_extent(inode, ordered, 1);
3262                         btrfs_put_ordered_extent(ordered);
3263                         goto again;
3264                 }
3265                 set_page_extent_mapped(page);
3266
3267                 btrfs_set_extent_delalloc(inode, page_start, page_end);
3268                 if (i == first_index)
3269                         set_extent_bits(io_tree, page_start, page_end,
3270                                         EXTENT_BOUNDARY, GFP_NOFS);
3271
3272                 set_page_dirty(page);
3273                 total_dirty++;
3274
3275                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3276                 unlock_page(page);
3277                 page_cache_release(page);
3278         }
3279
3280 out_unlock:
3281         kfree(ra);
3282         mutex_unlock(&inode->i_mutex);
3283         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
3284         return ret;
3285 }
3286
3287 static int noinline relocate_data_extent(struct inode *reloc_inode,
3288                                          struct btrfs_key *extent_key,
3289                                          u64 offset)
3290 {
3291         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3292         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
3293         struct extent_map *em;
3294         u64 start = extent_key->objectid - offset;
3295         u64 end = start + extent_key->offset - 1;
3296
3297         em = alloc_extent_map(GFP_NOFS);
3298         BUG_ON(!em || IS_ERR(em));
3299
3300         em->start = start;
3301         em->len = extent_key->offset;
3302         em->block_len = extent_key->offset;
3303         em->block_start = extent_key->objectid;
3304         em->bdev = root->fs_info->fs_devices->latest_bdev;
3305         set_bit(EXTENT_FLAG_PINNED, &em->flags);
3306
3307         /* setup extent map to cheat btrfs_readpage */
3308         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3309         while (1) {
3310                 int ret;
3311                 spin_lock(&em_tree->lock);
3312                 ret = add_extent_mapping(em_tree, em);
3313                 spin_unlock(&em_tree->lock);
3314                 if (ret != -EEXIST) {
3315                         free_extent_map(em);
3316                         break;
3317                 }
3318                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
3319         }
3320         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
3321
3322         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
3323 }
3324
3325 struct btrfs_ref_path {
3326         u64 extent_start;
3327         u64 nodes[BTRFS_MAX_LEVEL];
3328         u64 root_objectid;
3329         u64 root_generation;
3330         u64 owner_objectid;
3331         u32 num_refs;
3332         int lowest_level;
3333         int current_level;
3334         int shared_level;
3335
3336         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
3337         u64 new_nodes[BTRFS_MAX_LEVEL];
3338 };
3339
3340 struct disk_extent {
3341         u64 ram_bytes;
3342         u64 disk_bytenr;
3343         u64 disk_num_bytes;
3344         u64 offset;
3345         u64 num_bytes;
3346         u8 compression;
3347         u8 encryption;
3348         u16 other_encoding;
3349 };
3350
3351 static int is_cowonly_root(u64 root_objectid)
3352 {
3353         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
3354             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
3355             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
3356             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
3357             root_objectid == BTRFS_TREE_LOG_OBJECTID)
3358                 return 1;
3359         return 0;
3360 }
3361
3362 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
3363                                     struct btrfs_root *extent_root,
3364                                     struct btrfs_ref_path *ref_path,
3365                                     int first_time)
3366 {
3367         struct extent_buffer *leaf;
3368         struct btrfs_path *path;
3369         struct btrfs_extent_ref *ref;
3370         struct btrfs_key key;
3371         struct btrfs_key found_key;
3372         u64 bytenr;
3373         u32 nritems;
3374         int level;
3375         int ret = 1;
3376
3377         path = btrfs_alloc_path();
3378         if (!path)
3379                 return -ENOMEM;
3380
3381         if (first_time) {
3382                 ref_path->lowest_level = -1;
3383                 ref_path->current_level = -1;
3384                 ref_path->shared_level = -1;
3385                 goto walk_up;
3386         }
3387 walk_down:
3388         level = ref_path->current_level - 1;
3389         while (level >= -1) {
3390                 u64 parent;
3391                 if (level < ref_path->lowest_level)
3392                         break;
3393
3394                 if (level >= 0) {
3395                         bytenr = ref_path->nodes[level];
3396                 } else {
3397                         bytenr = ref_path->extent_start;
3398                 }
3399                 BUG_ON(bytenr == 0);
3400
3401                 parent = ref_path->nodes[level + 1];
3402                 ref_path->nodes[level + 1] = 0;
3403                 ref_path->current_level = level;
3404                 BUG_ON(parent == 0);
3405
3406                 key.objectid = bytenr;
3407                 key.offset = parent + 1;
3408                 key.type = BTRFS_EXTENT_REF_KEY;
3409
3410                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3411                 if (ret < 0)
3412                         goto out;
3413                 BUG_ON(ret == 0);
3414
3415                 leaf = path->nodes[0];
3416                 nritems = btrfs_header_nritems(leaf);
3417                 if (path->slots[0] >= nritems) {
3418                         ret = btrfs_next_leaf(extent_root, path);
3419                         if (ret < 0)
3420                                 goto out;
3421                         if (ret > 0)
3422                                 goto next;
3423                         leaf = path->nodes[0];
3424                 }
3425
3426                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3427                 if (found_key.objectid == bytenr &&
3428                     found_key.type == BTRFS_EXTENT_REF_KEY) {
3429                         if (level < ref_path->shared_level)
3430                                 ref_path->shared_level = level;
3431                         goto found;
3432                 }
3433 next:
3434                 level--;
3435                 btrfs_release_path(extent_root, path);
3436                 cond_resched();
3437         }
3438         /* reached lowest level */
3439         ret = 1;
3440         goto out;
3441 walk_up:
3442         level = ref_path->current_level;
3443         while (level < BTRFS_MAX_LEVEL - 1) {
3444                 u64 ref_objectid;
3445                 if (level >= 0) {
3446                         bytenr = ref_path->nodes[level];
3447                 } else {
3448                         bytenr = ref_path->extent_start;
3449                 }
3450                 BUG_ON(bytenr == 0);
3451
3452                 key.objectid = bytenr;
3453                 key.offset = 0;
3454                 key.type = BTRFS_EXTENT_REF_KEY;
3455
3456                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3457                 if (ret < 0)
3458                         goto out;
3459
3460                 leaf = path->nodes[0];
3461                 nritems = btrfs_header_nritems(leaf);
3462                 if (path->slots[0] >= nritems) {
3463                         ret = btrfs_next_leaf(extent_root, path);
3464                         if (ret < 0)
3465                                 goto out;
3466                         if (ret > 0) {
3467                                 /* the extent was freed by someone */
3468                                 if (ref_path->lowest_level == level)
3469                                         goto out;
3470                                 btrfs_release_path(extent_root, path);
3471                                 goto walk_down;
3472                         }
3473                         leaf = path->nodes[0];
3474                 }
3475
3476                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3477                 if (found_key.objectid != bytenr ||
3478                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
3479                         /* the extent was freed by someone */
3480                         if (ref_path->lowest_level == level) {
3481                                 ret = 1;
3482                                 goto out;
3483                         }
3484                         btrfs_release_path(extent_root, path);
3485                         goto walk_down;
3486                 }
3487 found:
3488                 ref = btrfs_item_ptr(leaf, path->slots[0],
3489                                 struct btrfs_extent_ref);
3490                 ref_objectid = btrfs_ref_objectid(leaf, ref);
3491                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3492                         if (first_time) {
3493                                 level = (int)ref_objectid;
3494                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
3495                                 ref_path->lowest_level = level;
3496                                 ref_path->current_level = level;
3497                                 ref_path->nodes[level] = bytenr;
3498                         } else {
3499                                 WARN_ON(ref_objectid != level);
3500                         }
3501                 } else {
3502                         WARN_ON(level != -1);
3503                 }
3504                 first_time = 0;
3505
3506                 if (ref_path->lowest_level == level) {
3507                         ref_path->owner_objectid = ref_objectid;
3508                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
3509                 }
3510
3511                 /*
3512                  * the block is tree root or the block isn't in reference
3513                  * counted tree.
3514                  */
3515                 if (found_key.objectid == found_key.offset ||
3516                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
3517                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3518                         ref_path->root_generation =
3519                                 btrfs_ref_generation(leaf, ref);
3520                         if (level < 0) {
3521                                 /* special reference from the tree log */
3522                                 ref_path->nodes[0] = found_key.offset;
3523                                 ref_path->current_level = 0;
3524                         }
3525                         ret = 0;
3526                         goto out;
3527                 }
3528
3529                 level++;
3530                 BUG_ON(ref_path->nodes[level] != 0);
3531                 ref_path->nodes[level] = found_key.offset;
3532                 ref_path->current_level = level;
3533
3534                 /*
3535                  * the reference was created in the running transaction,
3536                  * no need to continue walking up.
3537                  */
3538                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
3539                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
3540                         ref_path->root_generation =
3541                                 btrfs_ref_generation(leaf, ref);
3542                         ret = 0;
3543                         goto out;
3544                 }
3545
3546                 btrfs_release_path(extent_root, path);
3547                 cond_resched();
3548         }
3549         /* reached max tree level, but no tree root found. */
3550         BUG();
3551 out:
3552         btrfs_free_path(path);
3553         return ret;
3554 }
3555
3556 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
3557                                 struct btrfs_root *extent_root,
3558                                 struct btrfs_ref_path *ref_path,
3559                                 u64 extent_start)
3560 {
3561         memset(ref_path, 0, sizeof(*ref_path));
3562         ref_path->extent_start = extent_start;
3563
3564         return __next_ref_path(trans, extent_root, ref_path, 1);
3565 }
3566
3567 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
3568                                struct btrfs_root *extent_root,
3569                                struct btrfs_ref_path *ref_path)
3570 {
3571         return __next_ref_path(trans, extent_root, ref_path, 0);
3572 }
3573
3574 static int noinline get_new_locations(struct inode *reloc_inode,
3575                                       struct btrfs_key *extent_key,
3576                                       u64 offset, int no_fragment,
3577                                       struct disk_extent **extents,
3578                                       int *nr_extents)
3579 {
3580         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
3581         struct btrfs_path *path;
3582         struct btrfs_file_extent_item *fi;
3583         struct extent_buffer *leaf;
3584         struct disk_extent *exts = *extents;
3585         struct btrfs_key found_key;
3586         u64 cur_pos;
3587         u64 last_byte;
3588         u32 nritems;
3589         int nr = 0;
3590         int max = *nr_extents;
3591         int ret;
3592
3593         WARN_ON(!no_fragment && *extents);
3594         if (!exts) {
3595                 max = 1;
3596                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
3597                 if (!exts)
3598                         return -ENOMEM;
3599         }
3600
3601         path = btrfs_alloc_path();
3602         BUG_ON(!path);
3603
3604         cur_pos = extent_key->objectid - offset;
3605         last_byte = extent_key->objectid + extent_key->offset;
3606         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
3607                                        cur_pos, 0);
3608         if (ret < 0)
3609                 goto out;
3610         if (ret > 0) {
3611                 ret = -ENOENT;
3612                 goto out;
3613         }
3614
3615         while (1) {
3616                 leaf = path->nodes[0];
3617                 nritems = btrfs_header_nritems(leaf);
3618                 if (path->slots[0] >= nritems) {
3619                         ret = btrfs_next_leaf(root, path);
3620                         if (ret < 0)
3621                                 goto out;
3622                         if (ret > 0)
3623                                 break;
3624                         leaf = path->nodes[0];
3625                 }
3626
3627                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3628                 if (found_key.offset != cur_pos ||
3629                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
3630                     found_key.objectid != reloc_inode->i_ino)
3631                         break;
3632
3633                 fi = btrfs_item_ptr(leaf, path->slots[0],
3634                                     struct btrfs_file_extent_item);
3635                 if (btrfs_file_extent_type(leaf, fi) !=
3636                     BTRFS_FILE_EXTENT_REG ||
3637                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
3638                         break;
3639
3640                 if (nr == max) {
3641                         struct disk_extent *old = exts;
3642                         max *= 2;
3643                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
3644                         memcpy(exts, old, sizeof(*exts) * nr);
3645                         if (old != *extents)
3646                                 kfree(old);
3647                 }
3648
3649                 exts[nr].disk_bytenr =
3650                         btrfs_file_extent_disk_bytenr(leaf, fi);
3651                 exts[nr].disk_num_bytes =
3652                         btrfs_file_extent_disk_num_bytes(leaf, fi);
3653                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
3654                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3655                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
3656                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
3657                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
3658                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
3659                                                                            fi);
3660                 BUG_ON(exts[nr].offset > 0);
3661                 BUG_ON(exts[nr].compression || exts[nr].encryption);
3662                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
3663
3664                 cur_pos += exts[nr].num_bytes;
3665                 nr++;
3666
3667                 if (cur_pos + offset >= last_byte)
3668                         break;
3669
3670                 if (no_fragment) {
3671                         ret = 1;
3672                         goto out;
3673                 }
3674                 path->slots[0]++;
3675         }
3676
3677         WARN_ON(cur_pos + offset > last_byte);
3678         if (cur_pos + offset < last_byte) {
3679                 ret = -ENOENT;
3680                 goto out;
3681         }
3682         ret = 0;
3683 out:
3684         btrfs_free_path(path);
3685         if (ret) {
3686                 if (exts != *extents)
3687                         kfree(exts);
3688         } else {
3689                 *extents = exts;
3690                 *nr_extents = nr;
3691         }
3692         return ret;
3693 }
3694
3695 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
3696                                         struct btrfs_root *root,
3697                                         struct btrfs_path *path,
3698                                         struct btrfs_key *extent_key,
3699                                         struct btrfs_key *leaf_key,
3700                                         struct btrfs_ref_path *ref_path,
3701                                         struct disk_extent *new_extents,
3702                                         int nr_extents)
3703 {
3704         struct extent_buffer *leaf;
3705         struct btrfs_file_extent_item *fi;
3706         struct inode *inode = NULL;
3707         struct btrfs_key key;
3708         u64 lock_start = 0;
3709         u64 lock_end = 0;
3710         u64 num_bytes;
3711         u64 ext_offset;
3712         u64 first_pos;
3713         u32 nritems;
3714         int nr_scaned = 0;
3715         int extent_locked = 0;
3716         int extent_type;
3717         int ret;
3718
3719         memcpy(&key, leaf_key, sizeof(key));
3720         first_pos = INT_LIMIT(loff_t) - extent_key->offset;
3721         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3722                 if (key.objectid < ref_path->owner_objectid ||
3723                     (key.objectid == ref_path->owner_objectid &&
3724                      key.type < BTRFS_EXTENT_DATA_KEY)) {
3725                         key.objectid = ref_path->owner_objectid;
3726                         key.type = BTRFS_EXTENT_DATA_KEY;
3727                         key.offset = 0;
3728                 }
3729         }
3730
3731         while (1) {
3732                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3733                 if (ret < 0)
3734                         goto out;
3735
3736                 leaf = path->nodes[0];
3737                 nritems = btrfs_header_nritems(leaf);
3738 next:
3739                 if (extent_locked && ret > 0) {
3740                         /*
3741                          * the file extent item was modified by someone
3742                          * before the extent got locked.
3743                          */
3744                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3745                                       lock_end, GFP_NOFS);
3746                         extent_locked = 0;
3747                 }
3748
3749                 if (path->slots[0] >= nritems) {
3750                         if (++nr_scaned > 2)
3751                                 break;
3752
3753                         BUG_ON(extent_locked);
3754                         ret = btrfs_next_leaf(root, path);
3755                         if (ret < 0)
3756                                 goto out;
3757                         if (ret > 0)
3758                                 break;
3759                         leaf = path->nodes[0];
3760                         nritems = btrfs_header_nritems(leaf);
3761                 }
3762
3763                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3764
3765                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
3766                         if ((key.objectid > ref_path->owner_objectid) ||
3767                             (key.objectid == ref_path->owner_objectid &&
3768                              key.type > BTRFS_EXTENT_DATA_KEY) ||
3769                             (key.offset >= first_pos + extent_key->offset))
3770                                 break;
3771                 }
3772
3773                 if (inode && key.objectid != inode->i_ino) {
3774                         BUG_ON(extent_locked);
3775                         btrfs_release_path(root, path);
3776                         mutex_unlock(&inode->i_mutex);
3777                         iput(inode);
3778                         inode = NULL;
3779                         continue;
3780                 }
3781
3782                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
3783                         path->slots[0]++;
3784                         ret = 1;
3785                         goto next;
3786                 }
3787                 fi = btrfs_item_ptr(leaf, path->slots[0],
3788                                     struct btrfs_file_extent_item);
3789                 extent_type = btrfs_file_extent_type(leaf, fi);
3790                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
3791                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
3792                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3793                      extent_key->objectid)) {
3794                         path->slots[0]++;
3795                         ret = 1;
3796                         goto next;
3797                 }
3798
3799                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
3800                 ext_offset = btrfs_file_extent_offset(leaf, fi);
3801
3802                 if (first_pos > key.offset - ext_offset)
3803                         first_pos = key.offset - ext_offset;
3804
3805                 if (!extent_locked) {
3806                         lock_start = key.offset;
3807                         lock_end = lock_start + num_bytes - 1;
3808                 } else {
3809                         if (lock_start > key.offset ||
3810                             lock_end + 1 < key.offset + num_bytes) {
3811                                 unlock_extent(&BTRFS_I(inode)->io_tree,
3812                                               lock_start, lock_end, GFP_NOFS);
3813                                 extent_locked = 0;
3814                         }
3815                 }
3816
3817                 if (!inode) {
3818                         btrfs_release_path(root, path);
3819
3820                         inode = btrfs_iget_locked(root->fs_info->sb,
3821                                                   key.objectid, root);
3822                         if (inode->i_state & I_NEW) {
3823                                 BTRFS_I(inode)->root = root;
3824                                 BTRFS_I(inode)->location.objectid =
3825                                         key.objectid;
3826                                 BTRFS_I(inode)->location.type =
3827                                         BTRFS_INODE_ITEM_KEY;
3828                                 BTRFS_I(inode)->location.offset = 0;
3829                                 btrfs_read_locked_inode(inode);
3830                                 unlock_new_inode(inode);
3831                         }
3832                         /*
3833                          * some code call btrfs_commit_transaction while
3834                          * holding the i_mutex, so we can't use mutex_lock
3835                          * here.
3836                          */
3837                         if (is_bad_inode(inode) ||
3838                             !mutex_trylock(&inode->i_mutex)) {
3839                                 iput(inode);
3840                                 inode = NULL;
3841                                 key.offset = (u64)-1;
3842                                 goto skip;
3843                         }
3844                 }
3845
3846                 if (!extent_locked) {
3847                         struct btrfs_ordered_extent *ordered;
3848
3849                         btrfs_release_path(root, path);
3850
3851                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3852                                     lock_end, GFP_NOFS);
3853                         ordered = btrfs_lookup_first_ordered_extent(inode,
3854                                                                     lock_end);
3855                         if (ordered &&
3856                             ordered->file_offset <= lock_end &&
3857                             ordered->file_offset + ordered->len > lock_start) {
3858                                 unlock_extent(&BTRFS_I(inode)->io_tree,
3859                                               lock_start, lock_end, GFP_NOFS);
3860                                 btrfs_start_ordered_extent(inode, ordered, 1);
3861                                 btrfs_put_ordered_extent(ordered);
3862                                 key.offset += num_bytes;
3863                                 goto skip;
3864                         }
3865                         if (ordered)
3866                                 btrfs_put_ordered_extent(ordered);
3867
3868                         extent_locked = 1;
3869                         continue;
3870                 }
3871
3872                 if (nr_extents == 1) {
3873                         /* update extent pointer in place */
3874                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
3875                                                 new_extents[0].disk_bytenr);
3876                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3877                                                 new_extents[0].disk_num_bytes);
3878                         btrfs_mark_buffer_dirty(leaf);
3879
3880                         btrfs_drop_extent_cache(inode, key.offset,
3881                                                 key.offset + num_bytes - 1, 0);
3882
3883                         ret = btrfs_inc_extent_ref(trans, root,
3884                                                 new_extents[0].disk_bytenr,
3885                                                 new_extents[0].disk_num_bytes,
3886                                                 leaf->start,
3887                                                 root->root_key.objectid,
3888                                                 trans->transid,
3889                                                 key.objectid);
3890                         BUG_ON(ret);
3891
3892                         ret = btrfs_free_extent(trans, root,
3893                                                 extent_key->objectid,
3894                                                 extent_key->offset,
3895                                                 leaf->start,
3896                                                 btrfs_header_owner(leaf),
3897                                                 btrfs_header_generation(leaf),
3898                                                 key.objectid, 0);
3899                         BUG_ON(ret);
3900
3901                         btrfs_release_path(root, path);
3902                         key.offset += num_bytes;
3903                 } else {
3904                         BUG_ON(1);
3905 #if 0
3906                         u64 alloc_hint;
3907                         u64 extent_len;
3908                         int i;
3909                         /*
3910                          * drop old extent pointer at first, then insert the
3911                          * new pointers one bye one
3912                          */
3913                         btrfs_release_path(root, path);
3914                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
3915                                                  key.offset + num_bytes,
3916                                                  key.offset, &alloc_hint);
3917                         BUG_ON(ret);
3918
3919                         for (i = 0; i < nr_extents; i++) {
3920                                 if (ext_offset >= new_extents[i].num_bytes) {
3921                                         ext_offset -= new_extents[i].num_bytes;
3922                                         continue;
3923                                 }
3924                                 extent_len = min(new_extents[i].num_bytes -
3925                                                  ext_offset, num_bytes);
3926
3927                                 ret = btrfs_insert_empty_item(trans, root,
3928                                                               path, &key,
3929                                                               sizeof(*fi));
3930                                 BUG_ON(ret);
3931
3932                                 leaf = path->nodes[0];
3933                                 fi = btrfs_item_ptr(leaf, path->slots[0],
3934                                                 struct btrfs_file_extent_item);
3935                                 btrfs_set_file_extent_generation(leaf, fi,
3936                                                         trans->transid);
3937                                 btrfs_set_file_extent_type(leaf, fi,
3938                                                         BTRFS_FILE_EXTENT_REG);
3939                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
3940                                                 new_extents[i].disk_bytenr);
3941                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
3942                                                 new_extents[i].disk_num_bytes);
3943                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
3944                                                 new_extents[i].ram_bytes);
3945
3946                                 btrfs_set_file_extent_compression(leaf, fi,
3947                                                 new_extents[i].compression);
3948                                 btrfs_set_file_extent_encryption(leaf, fi,
3949                                                 new_extents[i].encryption);
3950                                 btrfs_set_file_extent_other_encoding(leaf, fi,
3951                                                 new_extents[i].other_encoding);
3952
3953                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3954                                                         extent_len);
3955                                 ext_offset += new_extents[i].offset;
3956                                 btrfs_set_file_extent_offset(leaf, fi,
3957                                                         ext_offset);
3958                                 btrfs_mark_buffer_dirty(leaf);
3959
3960                                 btrfs_drop_extent_cache(inode, key.offset,
3961                                                 key.offset + extent_len - 1, 0);
3962
3963                                 ret = btrfs_inc_extent_ref(trans, root,
3964                                                 new_extents[i].disk_bytenr,
3965                                                 new_extents[i].disk_num_bytes,
3966                                                 leaf->start,
3967                                                 root->root_key.objectid,
3968                                                 trans->transid, key.objectid);
3969                                 BUG_ON(ret);
3970                                 btrfs_release_path(root, path);
3971
3972                                 inode_add_bytes(inode, extent_len);
3973
3974                                 ext_offset = 0;
3975                                 num_bytes -= extent_len;
3976                                 key.offset += extent_len;
3977
3978                                 if (num_bytes == 0)
3979                                         break;
3980                         }
3981                         BUG_ON(i >= nr_extents);
3982 #endif
3983                 }
3984
3985                 if (extent_locked) {
3986                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
3987                                       lock_end, GFP_NOFS);
3988                         extent_locked = 0;
3989                 }
3990 skip:
3991                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
3992                     key.offset >= first_pos + extent_key->offset)
3993                         break;
3994
3995                 cond_resched();
3996         }
3997         ret = 0;
3998 out:
3999         btrfs_release_path(root, path);
4000         if (inode) {
4001                 mutex_unlock(&inode->i_mutex);
4002                 if (extent_locked) {
4003                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4004                                       lock_end, GFP_NOFS);
4005                 }
4006                 iput(inode);
4007         }
4008         return ret;
4009 }
4010
4011 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4012                                struct btrfs_root *root,
4013                                struct extent_buffer *buf, u64 orig_start)
4014 {
4015         int level;
4016         int ret;
4017
4018         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4019         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4020
4021         level = btrfs_header_level(buf);
4022         if (level == 0) {
4023                 struct btrfs_leaf_ref *ref;
4024                 struct btrfs_leaf_ref *orig_ref;
4025
4026                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4027                 if (!orig_ref)
4028                         return -ENOENT;
4029
4030                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4031                 if (!ref) {
4032                         btrfs_free_leaf_ref(root, orig_ref);
4033                         return -ENOMEM;
4034                 }
4035
4036                 ref->nritems = orig_ref->nritems;
4037                 memcpy(ref->extents, orig_ref->extents,
4038                         sizeof(ref->extents[0]) * ref->nritems);
4039
4040                 btrfs_free_leaf_ref(root, orig_ref);
4041
4042                 ref->root_gen = trans->transid;
4043                 ref->bytenr = buf->start;
4044                 ref->owner = btrfs_header_owner(buf);
4045                 ref->generation = btrfs_header_generation(buf);
4046                 ret = btrfs_add_leaf_ref(root, ref, 0);
4047                 WARN_ON(ret);
4048                 btrfs_free_leaf_ref(root, ref);
4049         }
4050         return 0;
4051 }
4052
4053 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4054                                         struct extent_buffer *leaf,
4055                                         struct btrfs_block_group_cache *group,
4056                                         struct btrfs_root *target_root)
4057 {
4058         struct btrfs_key key;
4059         struct inode *inode = NULL;
4060         struct btrfs_file_extent_item *fi;
4061         u64 num_bytes;
4062         u64 skip_objectid = 0;
4063         u32 nritems;
4064         u32 i;
4065
4066         nritems = btrfs_header_nritems(leaf);
4067         for (i = 0; i < nritems; i++) {
4068                 btrfs_item_key_to_cpu(leaf, &key, i);
4069                 if (key.objectid == skip_objectid ||
4070                     key.type != BTRFS_EXTENT_DATA_KEY)
4071                         continue;
4072                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4073                 if (btrfs_file_extent_type(leaf, fi) ==
4074                     BTRFS_FILE_EXTENT_INLINE)
4075                         continue;
4076                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4077                         continue;
4078                 if (!inode || inode->i_ino != key.objectid) {
4079                         iput(inode);
4080                         inode = btrfs_ilookup(target_root->fs_info->sb,
4081                                               key.objectid, target_root, 1);
4082                 }
4083                 if (!inode) {
4084                         skip_objectid = key.objectid;
4085                         continue;
4086                 }
4087                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4088
4089                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4090                             key.offset + num_bytes - 1, GFP_NOFS);
4091                 btrfs_drop_extent_cache(inode, key.offset,
4092                                         key.offset + num_bytes - 1, 1);
4093                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4094                               key.offset + num_bytes - 1, GFP_NOFS);
4095                 cond_resched();
4096         }
4097         iput(inode);
4098         return 0;
4099 }
4100
4101 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4102                                         struct btrfs_root *root,
4103                                         struct extent_buffer *leaf,
4104                                         struct btrfs_block_group_cache *group,
4105                                         struct inode *reloc_inode)
4106 {
4107         struct btrfs_key key;
4108         struct btrfs_key extent_key;
4109         struct btrfs_file_extent_item *fi;
4110         struct btrfs_leaf_ref *ref;
4111         struct disk_extent *new_extent;
4112         u64 bytenr;
4113         u64 num_bytes;
4114         u32 nritems;
4115         u32 i;
4116         int ext_index;
4117         int nr_extent;
4118         int ret;
4119
4120         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4121         BUG_ON(!new_extent);
4122
4123         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4124         BUG_ON(!ref);
4125
4126         ext_index = -1;
4127         nritems = btrfs_header_nritems(leaf);
4128         for (i = 0; i < nritems; i++) {
4129                 btrfs_item_key_to_cpu(leaf, &key, i);
4130                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4131                         continue;
4132                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4133                 if (btrfs_file_extent_type(leaf, fi) ==
4134                     BTRFS_FILE_EXTENT_INLINE)
4135                         continue;
4136                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4137                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4138                 if (bytenr == 0)
4139                         continue;
4140
4141                 ext_index++;
4142                 if (bytenr >= group->key.objectid + group->key.offset ||
4143                     bytenr + num_bytes <= group->key.objectid)
4144                         continue;
4145
4146                 extent_key.objectid = bytenr;
4147                 extent_key.offset = num_bytes;
4148                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4149                 nr_extent = 1;
4150                 ret = get_new_locations(reloc_inode, &extent_key,
4151                                         group->key.objectid, 1,
4152                                         &new_extent, &nr_extent);
4153                 if (ret > 0)
4154                         continue;
4155                 BUG_ON(ret < 0);
4156
4157                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4158                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4159                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4160                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4161
4162                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4163                                                 new_extent->disk_bytenr);
4164                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4165                                                 new_extent->disk_num_bytes);
4166                 btrfs_mark_buffer_dirty(leaf);
4167
4168                 ret = btrfs_inc_extent_ref(trans, root,
4169                                         new_extent->disk_bytenr,
4170                                         new_extent->disk_num_bytes,
4171                                         leaf->start,
4172                                         root->root_key.objectid,
4173                                         trans->transid, key.objectid);
4174                 BUG_ON(ret);
4175                 ret = btrfs_free_extent(trans, root,
4176                                         bytenr, num_bytes, leaf->start,
4177                                         btrfs_header_owner(leaf),
4178                                         btrfs_header_generation(leaf),
4179                                         key.objectid, 0);
4180                 BUG_ON(ret);
4181                 cond_resched();
4182         }
4183         kfree(new_extent);
4184         BUG_ON(ext_index + 1 != ref->nritems);
4185         btrfs_free_leaf_ref(root, ref);
4186         return 0;
4187 }
4188
4189 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4190                           struct btrfs_root *root)
4191 {
4192         struct btrfs_root *reloc_root;
4193         int ret;
4194
4195         if (root->reloc_root) {
4196                 reloc_root = root->reloc_root;
4197                 root->reloc_root = NULL;
4198                 list_add(&reloc_root->dead_list,
4199                          &root->fs_info->dead_reloc_roots);
4200
4201                 btrfs_set_root_bytenr(&reloc_root->root_item,
4202                                       reloc_root->node->start);
4203                 btrfs_set_root_level(&root->root_item,
4204                                      btrfs_header_level(reloc_root->node));
4205                 memset(&reloc_root->root_item.drop_progress, 0,
4206                         sizeof(struct btrfs_disk_key));
4207                 reloc_root->root_item.drop_level = 0;
4208
4209                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4210                                         &reloc_root->root_key,
4211                                         &reloc_root->root_item);
4212                 BUG_ON(ret);
4213         }
4214         return 0;
4215 }
4216
4217 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4218 {
4219         struct btrfs_trans_handle *trans;
4220         struct btrfs_root *reloc_root;
4221         struct btrfs_root *prev_root = NULL;
4222         struct list_head dead_roots;
4223         int ret;
4224         unsigned long nr;
4225
4226         INIT_LIST_HEAD(&dead_roots);
4227         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4228
4229         while (!list_empty(&dead_roots)) {
4230                 reloc_root = list_entry(dead_roots.prev,
4231                                         struct btrfs_root, dead_list);
4232                 list_del_init(&reloc_root->dead_list);
4233
4234                 BUG_ON(reloc_root->commit_root != NULL);
4235                 while (1) {
4236                         trans = btrfs_join_transaction(root, 1);
4237                         BUG_ON(!trans);
4238
4239                         mutex_lock(&root->fs_info->drop_mutex);
4240                         ret = btrfs_drop_snapshot(trans, reloc_root);
4241                         if (ret != -EAGAIN)
4242                                 break;
4243                         mutex_unlock(&root->fs_info->drop_mutex);
4244
4245                         nr = trans->blocks_used;
4246                         ret = btrfs_end_transaction(trans, root);
4247                         BUG_ON(ret);
4248                         btrfs_btree_balance_dirty(root, nr);
4249                 }
4250
4251                 free_extent_buffer(reloc_root->node);
4252
4253                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4254                                      &reloc_root->root_key);
4255                 BUG_ON(ret);
4256                 mutex_unlock(&root->fs_info->drop_mutex);
4257
4258                 nr = trans->blocks_used;
4259                 ret = btrfs_end_transaction(trans, root);
4260                 BUG_ON(ret);
4261                 btrfs_btree_balance_dirty(root, nr);
4262
4263                 kfree(prev_root);
4264                 prev_root = reloc_root;
4265         }
4266         if (prev_root) {
4267                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4268                 kfree(prev_root);
4269         }
4270         return 0;
4271 }
4272
4273 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4274 {
4275         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4276         return 0;
4277 }
4278
4279 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
4280 {
4281         struct btrfs_root *reloc_root;
4282         struct btrfs_trans_handle *trans;
4283         struct btrfs_key location;
4284         int found;
4285         int ret;
4286
4287         mutex_lock(&root->fs_info->tree_reloc_mutex);
4288         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
4289         BUG_ON(ret);
4290         found = !list_empty(&root->fs_info->dead_reloc_roots);
4291         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4292
4293         if (found) {
4294                 trans = btrfs_start_transaction(root, 1);
4295                 BUG_ON(!trans);
4296                 ret = btrfs_commit_transaction(trans, root);
4297                 BUG_ON(ret);
4298         }
4299
4300         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4301         location.offset = (u64)-1;
4302         location.type = BTRFS_ROOT_ITEM_KEY;
4303
4304         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
4305         BUG_ON(!reloc_root);
4306         btrfs_orphan_cleanup(reloc_root);
4307         return 0;
4308 }
4309
4310 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
4311                                     struct btrfs_root *root)
4312 {
4313         struct btrfs_root *reloc_root;
4314         struct extent_buffer *eb;
4315         struct btrfs_root_item *root_item;
4316         struct btrfs_key root_key;
4317         int ret;
4318
4319         BUG_ON(!root->ref_cows);
4320         if (root->reloc_root)
4321                 return 0;
4322
4323         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
4324         BUG_ON(!root_item);
4325
4326         ret = btrfs_copy_root(trans, root, root->commit_root,
4327                               &eb, BTRFS_TREE_RELOC_OBJECTID);
4328         BUG_ON(ret);
4329
4330         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4331         root_key.offset = root->root_key.objectid;
4332         root_key.type = BTRFS_ROOT_ITEM_KEY;
4333
4334         memcpy(root_item, &root->root_item, sizeof(root_item));
4335         btrfs_set_root_refs(root_item, 0);
4336         btrfs_set_root_bytenr(root_item, eb->start);
4337         btrfs_set_root_level(root_item, btrfs_header_level(eb));
4338         btrfs_set_root_generation(root_item, trans->transid);
4339
4340         btrfs_tree_unlock(eb);
4341         free_extent_buffer(eb);
4342
4343         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
4344                                 &root_key, root_item);
4345         BUG_ON(ret);
4346         kfree(root_item);
4347
4348         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
4349                                                  &root_key);
4350         BUG_ON(!reloc_root);
4351         reloc_root->last_trans = trans->transid;
4352         reloc_root->commit_root = NULL;
4353         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
4354
4355         root->reloc_root = reloc_root;
4356         return 0;
4357 }
4358
4359 /*
4360  * Core function of space balance.
4361  *
4362  * The idea is using reloc trees to relocate tree blocks in reference
4363  * counted roots. There is one reloc tree for each subvol, and all
4364  * reloc trees share same root key objectid. Reloc trees are snapshots
4365  * of the latest committed roots of subvols (root->commit_root).
4366  *
4367  * To relocate a tree block referenced by a subvol, there are two steps.
4368  * COW the block through subvol's reloc tree, then update block pointer
4369  * in the subvol to point to the new block. Since all reloc trees share
4370  * same root key objectid, doing special handing for tree blocks owned
4371  * by them is easy. Once a tree block has been COWed in one reloc tree,
4372  * we can use the resulting new block directly when the same block is
4373  * required to COW again through other reloc trees. By this way, relocated
4374  * tree blocks are shared between reloc trees, so they are also shared
4375  * between subvols.
4376  */
4377 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
4378                                       struct btrfs_root *root,
4379                                       struct btrfs_path *path,
4380                                       struct btrfs_key *first_key,
4381                                       struct btrfs_ref_path *ref_path,
4382                                       struct btrfs_block_group_cache *group,
4383                                       struct inode *reloc_inode)
4384 {
4385         struct btrfs_root *reloc_root;
4386         struct extent_buffer *eb = NULL;
4387         struct btrfs_key *keys;
4388         u64 *nodes;
4389         int level;
4390         int shared_level;
4391         int lowest_level = 0;
4392         int ret;
4393
4394         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
4395                 lowest_level = ref_path->owner_objectid;
4396
4397         if (!root->ref_cows) {
4398                 path->lowest_level = lowest_level;
4399                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
4400                 BUG_ON(ret < 0);
4401                 path->lowest_level = 0;
4402                 btrfs_release_path(root, path);
4403                 return 0;
4404         }
4405
4406         mutex_lock(&root->fs_info->tree_reloc_mutex);
4407         ret = init_reloc_tree(trans, root);
4408         BUG_ON(ret);
4409         reloc_root = root->reloc_root;
4410
4411         shared_level = ref_path->shared_level;
4412         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
4413
4414         keys = ref_path->node_keys;
4415         nodes = ref_path->new_nodes;
4416         memset(&keys[shared_level + 1], 0,
4417                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
4418         memset(&nodes[shared_level + 1], 0,
4419                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
4420
4421         if (nodes[lowest_level] == 0) {
4422                 path->lowest_level = lowest_level;
4423                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4424                                         0, 1);
4425                 BUG_ON(ret);
4426                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
4427                         eb = path->nodes[level];
4428                         if (!eb || eb == reloc_root->node)
4429                                 break;
4430                         nodes[level] = eb->start;
4431                         if (level == 0)
4432                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
4433                         else
4434                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
4435                 }
4436                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4437                         eb = path->nodes[0];
4438                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
4439                                                       group, reloc_inode);
4440                         BUG_ON(ret);
4441                 }
4442                 btrfs_release_path(reloc_root, path);
4443         } else {
4444                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
4445                                        lowest_level);
4446                 BUG_ON(ret);
4447         }
4448
4449         /*
4450          * replace tree blocks in the fs tree with tree blocks in
4451          * the reloc tree.
4452          */
4453         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
4454         BUG_ON(ret < 0);
4455
4456         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4457                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
4458                                         0, 0);
4459                 BUG_ON(ret);
4460                 extent_buffer_get(path->nodes[0]);
4461                 eb = path->nodes[0];
4462                 btrfs_release_path(reloc_root, path);
4463                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
4464                 BUG_ON(ret);
4465                 free_extent_buffer(eb);
4466         }
4467
4468         mutex_unlock(&root->fs_info->tree_reloc_mutex);
4469         path->lowest_level = 0;
4470         return 0;
4471 }
4472
4473 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
4474                                         struct btrfs_root *root,
4475                                         struct btrfs_path *path,
4476                                         struct btrfs_key *first_key,
4477                                         struct btrfs_ref_path *ref_path)
4478 {
4479         int ret;
4480
4481         ret = relocate_one_path(trans, root, path, first_key,
4482                                 ref_path, NULL, NULL);
4483         BUG_ON(ret);
4484
4485         if (root == root->fs_info->extent_root)
4486                 btrfs_extent_post_op(trans, root);
4487
4488         return 0;
4489 }
4490
4491 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
4492                                     struct btrfs_root *extent_root,
4493                                     struct btrfs_path *path,
4494                                     struct btrfs_key *extent_key)
4495 {
4496         int ret;
4497
4498         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
4499         if (ret)
4500                 goto out;
4501         ret = btrfs_del_item(trans, extent_root, path);
4502 out:
4503         btrfs_release_path(extent_root, path);
4504         return ret;
4505 }
4506
4507 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
4508                                                 struct btrfs_ref_path *ref_path)
4509 {
4510         struct btrfs_key root_key;
4511
4512         root_key.objectid = ref_path->root_objectid;
4513         root_key.type = BTRFS_ROOT_ITEM_KEY;
4514         if (is_cowonly_root(ref_path->root_objectid))
4515                 root_key.offset = 0;
4516         else
4517                 root_key.offset = (u64)-1;
4518
4519         return btrfs_read_fs_root_no_name(fs_info, &root_key);
4520 }
4521
4522 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
4523                                         struct btrfs_path *path,
4524                                         struct btrfs_key *extent_key,
4525                                         struct btrfs_block_group_cache *group,
4526                                         struct inode *reloc_inode, int pass)
4527 {
4528         struct btrfs_trans_handle *trans;
4529         struct btrfs_root *found_root;
4530         struct btrfs_ref_path *ref_path = NULL;
4531         struct disk_extent *new_extents = NULL;
4532         int nr_extents = 0;
4533         int loops;
4534         int ret;
4535         int level;
4536         struct btrfs_key first_key;
4537         u64 prev_block = 0;
4538
4539
4540         trans = btrfs_start_transaction(extent_root, 1);
4541         BUG_ON(!trans);
4542
4543         if (extent_key->objectid == 0) {
4544                 ret = del_extent_zero(trans, extent_root, path, extent_key);
4545                 goto out;
4546         }
4547
4548         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
4549         if (!ref_path) {
4550                ret = -ENOMEM;
4551                goto out;
4552         }
4553
4554         for (loops = 0; ; loops++) {
4555                 if (loops == 0) {
4556                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
4557                                                    extent_key->objectid);
4558                 } else {
4559                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
4560                 }
4561                 if (ret < 0)
4562                         goto out;
4563                 if (ret > 0)
4564                         break;
4565
4566                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4567                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
4568                         continue;
4569
4570                 found_root = read_ref_root(extent_root->fs_info, ref_path);
4571                 BUG_ON(!found_root);
4572                 /*
4573                  * for reference counted tree, only process reference paths
4574                  * rooted at the latest committed root.
4575                  */
4576                 if (found_root->ref_cows &&
4577                     ref_path->root_generation != found_root->root_key.offset)
4578                         continue;
4579
4580                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
4581                         if (pass == 0) {
4582                                 /*
4583                                  * copy data extents to new locations
4584                                  */
4585                                 u64 group_start = group->key.objectid;
4586                                 ret = relocate_data_extent(reloc_inode,
4587                                                            extent_key,
4588                                                            group_start);
4589                                 if (ret < 0)
4590                                         goto out;
4591                                 break;
4592                         }
4593                         level = 0;
4594                 } else {
4595                         level = ref_path->owner_objectid;
4596                 }
4597
4598                 if (prev_block != ref_path->nodes[level]) {
4599                         struct extent_buffer *eb;
4600                         u64 block_start = ref_path->nodes[level];
4601                         u64 block_size = btrfs_level_size(found_root, level);
4602
4603                         eb = read_tree_block(found_root, block_start,
4604                                              block_size, 0);
4605                         btrfs_tree_lock(eb);
4606                         BUG_ON(level != btrfs_header_level(eb));
4607
4608                         if (level == 0)
4609                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
4610                         else
4611                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
4612
4613                         btrfs_tree_unlock(eb);
4614                         free_extent_buffer(eb);
4615                         prev_block = block_start;
4616                 }
4617
4618                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
4619                     pass >= 2) {
4620                         /*
4621                          * use fallback method to process the remaining
4622                          * references.
4623                          */
4624                         if (!new_extents) {
4625                                 u64 group_start = group->key.objectid;
4626                                 new_extents = kmalloc(sizeof(*new_extents),
4627                                                       GFP_NOFS);
4628                                 nr_extents = 1;
4629                                 ret = get_new_locations(reloc_inode,
4630                                                         extent_key,
4631                                                         group_start, 1,
4632                                                         &new_extents,
4633                                                         &nr_extents);
4634                                 if (ret)
4635                                         goto out;
4636                         }
4637                         btrfs_record_root_in_trans(found_root);
4638                         ret = replace_one_extent(trans, found_root,
4639                                                 path, extent_key,
4640                                                 &first_key, ref_path,
4641                                                 new_extents, nr_extents);
4642                         if (ret < 0)
4643                                 goto out;
4644                         continue;
4645                 }
4646
4647                 btrfs_record_root_in_trans(found_root);
4648                 if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4649                         ret = relocate_tree_block(trans, found_root, path,
4650                                                   &first_key, ref_path);
4651                 } else {
4652                         /*
4653                          * try to update data extent references while
4654                          * keeping metadata shared between snapshots.
4655                          */
4656                         ret = relocate_one_path(trans, found_root, path,
4657                                                 &first_key, ref_path,
4658                                                 group, reloc_inode);
4659                 }
4660                 if (ret < 0)
4661                         goto out;
4662         }
4663         ret = 0;
4664 out:
4665         btrfs_end_transaction(trans, extent_root);
4666         kfree(new_extents);
4667         kfree(ref_path);
4668         return ret;
4669 }
4670
4671 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
4672 {
4673         u64 num_devices;
4674         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
4675                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
4676
4677         num_devices = root->fs_info->fs_devices->num_devices;
4678         if (num_devices == 1) {
4679                 stripped |= BTRFS_BLOCK_GROUP_DUP;
4680                 stripped = flags & ~stripped;
4681
4682                 /* turn raid0 into single device chunks */
4683                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
4684                         return stripped;
4685
4686                 /* turn mirroring into duplication */
4687                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
4688                              BTRFS_BLOCK_GROUP_RAID10))
4689                         return stripped | BTRFS_BLOCK_GROUP_DUP;
4690                 return flags;
4691         } else {
4692                 /* they already had raid on here, just return */
4693                 if (flags & stripped)
4694                         return flags;
4695
4696                 stripped |= BTRFS_BLOCK_GROUP_DUP;
4697                 stripped = flags & ~stripped;
4698
4699                 /* switch duplicated blocks with raid1 */
4700                 if (flags & BTRFS_BLOCK_GROUP_DUP)
4701                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
4702
4703                 /* turn single device chunks into raid0 */
4704                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
4705         }
4706         return flags;
4707 }
4708
4709 int __alloc_chunk_for_shrink(struct btrfs_root *root,
4710                      struct btrfs_block_group_cache *shrink_block_group,
4711                      int force)
4712 {
4713         struct btrfs_trans_handle *trans;
4714         u64 new_alloc_flags;
4715         u64 calc;
4716
4717         spin_lock(&shrink_block_group->lock);
4718         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
4719                 spin_unlock(&shrink_block_group->lock);
4720
4721                 trans = btrfs_start_transaction(root, 1);
4722                 spin_lock(&shrink_block_group->lock);
4723
4724                 new_alloc_flags = update_block_group_flags(root,
4725                                                    shrink_block_group->flags);
4726                 if (new_alloc_flags != shrink_block_group->flags) {
4727                         calc =
4728                              btrfs_block_group_used(&shrink_block_group->item);
4729                 } else {
4730                         calc = shrink_block_group->key.offset;
4731                 }
4732                 spin_unlock(&shrink_block_group->lock);
4733
4734                 do_chunk_alloc(trans, root->fs_info->extent_root,
4735                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
4736
4737                 btrfs_end_transaction(trans, root);
4738         } else
4739                 spin_unlock(&shrink_block_group->lock);
4740         return 0;
4741 }
4742
4743 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4744                                  struct btrfs_root *root,
4745                                  u64 objectid, u64 size)
4746 {
4747         struct btrfs_path *path;
4748         struct btrfs_inode_item *item;
4749         struct extent_buffer *leaf;
4750         int ret;
4751
4752         path = btrfs_alloc_path();
4753         if (!path)
4754                 return -ENOMEM;
4755
4756         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4757         if (ret)
4758                 goto out;
4759
4760         leaf = path->nodes[0];
4761         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4762         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4763         btrfs_set_inode_generation(leaf, item, 1);
4764         btrfs_set_inode_size(leaf, item, size);
4765         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4766         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NODATASUM |
4767                                           BTRFS_INODE_NOCOMPRESS);
4768         btrfs_mark_buffer_dirty(leaf);
4769         btrfs_release_path(root, path);
4770 out:
4771         btrfs_free_path(path);
4772         return ret;
4773 }
4774
4775 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
4776                                         struct btrfs_block_group_cache *group)
4777 {
4778         struct inode *inode = NULL;
4779         struct btrfs_trans_handle *trans;
4780         struct btrfs_root *root;
4781         struct btrfs_key root_key;
4782         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
4783         int err = 0;
4784
4785         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
4786         root_key.type = BTRFS_ROOT_ITEM_KEY;
4787         root_key.offset = (u64)-1;
4788         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
4789         if (IS_ERR(root))
4790                 return ERR_CAST(root);
4791
4792         trans = btrfs_start_transaction(root, 1);
4793         BUG_ON(!trans);
4794
4795         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
4796         if (err)
4797                 goto out;
4798
4799         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
4800         BUG_ON(err);
4801
4802         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
4803                                        group->key.offset, 0, group->key.offset,
4804                                        0, 0, 0);
4805         BUG_ON(err);
4806
4807         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
4808         if (inode->i_state & I_NEW) {
4809                 BTRFS_I(inode)->root = root;
4810                 BTRFS_I(inode)->location.objectid = objectid;
4811                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
4812                 BTRFS_I(inode)->location.offset = 0;
4813                 btrfs_read_locked_inode(inode);
4814                 unlock_new_inode(inode);
4815                 BUG_ON(is_bad_inode(inode));
4816         } else {
4817                 BUG_ON(1);
4818         }
4819
4820         err = btrfs_orphan_add(trans, inode);
4821 out:
4822         btrfs_end_transaction(trans, root);
4823         if (err) {
4824                 if (inode)
4825                         iput(inode);
4826                 inode = ERR_PTR(err);
4827         }
4828         return inode;
4829 }
4830
4831 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
4832 {
4833         struct btrfs_trans_handle *trans;
4834         struct btrfs_path *path;
4835         struct btrfs_fs_info *info = root->fs_info;
4836         struct extent_buffer *leaf;
4837         struct inode *reloc_inode;
4838         struct btrfs_block_group_cache *block_group;
4839         struct btrfs_key key;
4840         u64 skipped;
4841         u64 cur_byte;
4842         u64 total_found;
4843         u32 nritems;
4844         int ret;
4845         int progress;
4846         int pass = 0;
4847
4848         root = root->fs_info->extent_root;
4849
4850         block_group = btrfs_lookup_block_group(info, group_start);
4851         BUG_ON(!block_group);
4852
4853         printk("btrfs relocating block group %llu flags %llu\n",
4854                (unsigned long long)block_group->key.objectid,
4855                (unsigned long long)block_group->flags);
4856
4857         path = btrfs_alloc_path();
4858         BUG_ON(!path);
4859
4860         reloc_inode = create_reloc_inode(info, block_group);
4861         BUG_ON(IS_ERR(reloc_inode));
4862
4863         __alloc_chunk_for_shrink(root, block_group, 1);
4864         block_group->ro = 1;
4865         block_group->space_info->total_bytes -= block_group->key.offset;
4866
4867         btrfs_start_delalloc_inodes(info->tree_root);
4868         btrfs_wait_ordered_extents(info->tree_root, 0);
4869 again:
4870         skipped = 0;
4871         total_found = 0;
4872         progress = 0;
4873         key.objectid = block_group->key.objectid;
4874         key.offset = 0;
4875         key.type = 0;
4876         cur_byte = key.objectid;
4877
4878         trans = btrfs_start_transaction(info->tree_root, 1);
4879         btrfs_commit_transaction(trans, info->tree_root);
4880
4881         mutex_lock(&root->fs_info->cleaner_mutex);
4882         btrfs_clean_old_snapshots(info->tree_root);
4883         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
4884         mutex_unlock(&root->fs_info->cleaner_mutex);
4885
4886         while(1) {
4887                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4888                 if (ret < 0)
4889                         goto out;
4890 next:
4891                 leaf = path->nodes[0];
4892                 nritems = btrfs_header_nritems(leaf);
4893                 if (path->slots[0] >= nritems) {
4894                         ret = btrfs_next_leaf(root, path);
4895                         if (ret < 0)
4896                                 goto out;
4897                         if (ret == 1) {
4898                                 ret = 0;
4899                                 break;
4900                         }
4901                         leaf = path->nodes[0];
4902                         nritems = btrfs_header_nritems(leaf);
4903                 }
4904
4905                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4906
4907                 if (key.objectid >= block_group->key.objectid +
4908                     block_group->key.offset)
4909                         break;
4910
4911                 if (progress && need_resched()) {
4912                         btrfs_release_path(root, path);
4913                         cond_resched();
4914                         progress = 0;
4915                         continue;
4916                 }
4917                 progress = 1;
4918
4919                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
4920                     key.objectid + key.offset <= cur_byte) {
4921                         path->slots[0]++;
4922                         goto next;
4923                 }
4924
4925                 total_found++;
4926                 cur_byte = key.objectid + key.offset;
4927                 btrfs_release_path(root, path);
4928
4929                 __alloc_chunk_for_shrink(root, block_group, 0);
4930                 ret = relocate_one_extent(root, path, &key, block_group,
4931                                           reloc_inode, pass);
4932                 BUG_ON(ret < 0);
4933                 if (ret > 0)
4934                         skipped++;
4935
4936                 key.objectid = cur_byte;
4937                 key.type = 0;
4938                 key.offset = 0;
4939         }
4940
4941         btrfs_release_path(root, path);
4942
4943         if (pass == 0) {
4944                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
4945                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
4946                 WARN_ON(reloc_inode->i_mapping->nrpages);
4947         }
4948
4949         if (total_found > 0) {
4950                 printk("btrfs found %llu extents in pass %d\n",
4951                        (unsigned long long)total_found, pass);
4952                 pass++;
4953                 if (total_found == skipped && pass > 2) {
4954                         iput(reloc_inode);
4955                         reloc_inode = create_reloc_inode(info, block_group);
4956                         pass = 0;
4957                 }
4958                 goto again;
4959         }
4960
4961         /* delete reloc_inode */
4962         iput(reloc_inode);
4963
4964         /* unpin extents in this range */
4965         trans = btrfs_start_transaction(info->tree_root, 1);
4966         btrfs_commit_transaction(trans, info->tree_root);
4967
4968         spin_lock(&block_group->lock);
4969         WARN_ON(block_group->pinned > 0);
4970         WARN_ON(block_group->reserved > 0);
4971         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
4972         spin_unlock(&block_group->lock);
4973         ret = 0;
4974 out:
4975         btrfs_free_path(path);
4976         return ret;
4977 }
4978
4979 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
4980                            struct btrfs_key *key)
4981 {
4982         int ret = 0;
4983         struct btrfs_key found_key;
4984         struct extent_buffer *leaf;
4985         int slot;
4986
4987         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
4988         if (ret < 0)
4989                 goto out;
4990
4991         while(1) {
4992                 slot = path->slots[0];
4993                 leaf = path->nodes[0];
4994                 if (slot >= btrfs_header_nritems(leaf)) {
4995                         ret = btrfs_next_leaf(root, path);
4996                         if (ret == 0)
4997                                 continue;
4998                         if (ret < 0)
4999                                 goto out;
5000                         break;
5001                 }
5002                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5003
5004                 if (found_key.objectid >= key->objectid &&
5005                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5006                         ret = 0;
5007                         goto out;
5008                 }
5009                 path->slots[0]++;
5010         }
5011         ret = -ENOENT;
5012 out:
5013         return ret;
5014 }
5015
5016 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5017 {
5018         struct btrfs_block_group_cache *block_group;
5019         struct rb_node *n;
5020
5021         spin_lock(&info->block_group_cache_lock);
5022         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5023                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5024                                        cache_node);
5025                 rb_erase(&block_group->cache_node,
5026                          &info->block_group_cache_tree);
5027                 spin_unlock(&info->block_group_cache_lock);
5028
5029                 btrfs_remove_free_space_cache(block_group);
5030                 down_write(&block_group->space_info->groups_sem);
5031                 list_del(&block_group->list);
5032                 up_write(&block_group->space_info->groups_sem);
5033                 kfree(block_group);
5034
5035                 spin_lock(&info->block_group_cache_lock);
5036         }
5037         spin_unlock(&info->block_group_cache_lock);
5038         return 0;
5039 }
5040
5041 int btrfs_read_block_groups(struct btrfs_root *root)
5042 {
5043         struct btrfs_path *path;
5044         int ret;
5045         struct btrfs_block_group_cache *cache;
5046         struct btrfs_fs_info *info = root->fs_info;
5047         struct btrfs_space_info *space_info;
5048         struct btrfs_key key;
5049         struct btrfs_key found_key;
5050         struct extent_buffer *leaf;
5051
5052         root = info->extent_root;
5053         key.objectid = 0;
5054         key.offset = 0;
5055         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5056         path = btrfs_alloc_path();
5057         if (!path)
5058                 return -ENOMEM;
5059
5060         while(1) {
5061                 ret = find_first_block_group(root, path, &key);
5062                 if (ret > 0) {
5063                         ret = 0;
5064                         goto error;
5065                 }
5066                 if (ret != 0)
5067                         goto error;
5068
5069                 leaf = path->nodes[0];
5070                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5071                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5072                 if (!cache) {
5073                         ret = -ENOMEM;
5074                         break;
5075                 }
5076
5077                 spin_lock_init(&cache->lock);
5078                 mutex_init(&cache->alloc_mutex);
5079                 INIT_LIST_HEAD(&cache->list);
5080                 read_extent_buffer(leaf, &cache->item,
5081                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5082                                    sizeof(cache->item));
5083                 memcpy(&cache->key, &found_key, sizeof(found_key));
5084
5085                 key.objectid = found_key.objectid + found_key.offset;
5086                 btrfs_release_path(root, path);
5087                 cache->flags = btrfs_block_group_flags(&cache->item);
5088
5089                 ret = update_space_info(info, cache->flags, found_key.offset,
5090                                         btrfs_block_group_used(&cache->item),
5091                                         &space_info);
5092                 BUG_ON(ret);
5093                 cache->space_info = space_info;
5094                 down_write(&space_info->groups_sem);
5095                 list_add_tail(&cache->list, &space_info->block_groups);
5096                 up_write(&space_info->groups_sem);
5097
5098                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5099                 BUG_ON(ret);
5100
5101                 set_avail_alloc_bits(root->fs_info, cache->flags);
5102         }
5103         ret = 0;
5104 error:
5105         btrfs_free_path(path);
5106         return ret;
5107 }
5108
5109 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5110                            struct btrfs_root *root, u64 bytes_used,
5111                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5112                            u64 size)
5113 {
5114         int ret;
5115         struct btrfs_root *extent_root;
5116         struct btrfs_block_group_cache *cache;
5117
5118         extent_root = root->fs_info->extent_root;
5119
5120         root->fs_info->last_trans_new_blockgroup = trans->transid;
5121
5122         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5123         if (!cache)
5124                 return -ENOMEM;
5125
5126         cache->key.objectid = chunk_offset;
5127         cache->key.offset = size;
5128         spin_lock_init(&cache->lock);
5129         mutex_init(&cache->alloc_mutex);
5130         INIT_LIST_HEAD(&cache->list);
5131         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5132
5133         btrfs_set_block_group_used(&cache->item, bytes_used);
5134         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5135         cache->flags = type;
5136         btrfs_set_block_group_flags(&cache->item, type);
5137
5138         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5139                                 &cache->space_info);
5140         BUG_ON(ret);
5141         down_write(&cache->space_info->groups_sem);
5142         list_add_tail(&cache->list, &cache->space_info->block_groups);
5143         up_write(&cache->space_info->groups_sem);
5144
5145         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5146         BUG_ON(ret);
5147
5148         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5149                                 sizeof(cache->item));
5150         BUG_ON(ret);
5151
5152         finish_current_insert(trans, extent_root, 0);
5153         ret = del_pending_extents(trans, extent_root, 0);
5154         BUG_ON(ret);
5155         set_avail_alloc_bits(extent_root->fs_info, type);
5156
5157         return 0;
5158 }
5159
5160 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5161                              struct btrfs_root *root, u64 group_start)
5162 {
5163         struct btrfs_path *path;
5164         struct btrfs_block_group_cache *block_group;
5165         struct btrfs_key key;
5166         int ret;
5167
5168         root = root->fs_info->extent_root;
5169
5170         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5171         BUG_ON(!block_group);
5172
5173         memcpy(&key, &block_group->key, sizeof(key));
5174
5175         path = btrfs_alloc_path();
5176         BUG_ON(!path);
5177
5178         btrfs_remove_free_space_cache(block_group);
5179         rb_erase(&block_group->cache_node,
5180                  &root->fs_info->block_group_cache_tree);
5181         down_write(&block_group->space_info->groups_sem);
5182         list_del(&block_group->list);
5183         up_write(&block_group->space_info->groups_sem);
5184
5185         /*
5186         memset(shrink_block_group, 0, sizeof(*shrink_block_group));
5187         kfree(shrink_block_group);
5188         */
5189
5190         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5191         if (ret > 0)
5192                 ret = -EIO;
5193         if (ret < 0)
5194                 goto out;
5195
5196         ret = btrfs_del_item(trans, root, path);
5197 out:
5198         btrfs_free_path(path);
5199         return ret;
5200 }