git.oblomov.eu Git - linux-2.6/blob - fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/version.h>
  23 #include "compat.h"
  24 #include "hash.h"
  25 #include "crc32c.h"
  26 #include "ctree.h"
  27 #include "disk-io.h"
  28 #include "print-tree.h"
  29 #include "transaction.h"
  30 #include "volumes.h"
  31 #include "locking.h"
  32 #include "ref-cache.h"
  33 #include "compat.h"
  34
  35 #define PENDING_EXTENT_INSERT 0
  36 #define PENDING_EXTENT_DELETE 1
  37 #define PENDING_BACKREF_UPDATE 2
  38
  39 struct pending_extent_op {
  40         int type;
  41         u64 bytenr;
  42         u64 num_bytes;
  43         u64 parent;
  44         u64 orig_parent;
  45         u64 generation;
  46         u64 orig_generation;
  47         int level;
  48         struct list_head list;
  49         int del;
  50 };
  51
  52 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  53                                  btrfs_root *extent_root, int all);
  54 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  55                                btrfs_root *extent_root, int all);
  56 static int pin_down_bytes(struct btrfs_trans_handle *trans,
  57                           struct btrfs_root *root,
  58                           u64 bytenr, u64 num_bytes, int is_data);
  59 static int update_block_group(struct btrfs_trans_handle *trans,
  60                               struct btrfs_root *root,
  61                               u64 bytenr, u64 num_bytes, int alloc,
  62                               int mark_free);
  63
  64 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  65 {
  66         return (cache->flags & bits) == bits;
  67 }
  68
  69 /*
  70  * this adds the block group to the fs_info rb tree for the block group
  71  * cache
  72  */
  73 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  74                                 struct btrfs_block_group_cache *block_group)
  75 {
  76         struct rb_node **p;
  77         struct rb_node *parent = NULL;
  78         struct btrfs_block_group_cache *cache;
  79
  80         spin_lock(&info->block_group_cache_lock);
  81         p = &info->block_group_cache_tree.rb_node;
  82
  83         while (*p) {
  84                 parent = *p;
  85                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  86                                  cache_node);
  87                 if (block_group->key.objectid < cache->key.objectid) {
  88                         p = &(*p)->rb_left;
  89                 } else if (block_group->key.objectid > cache->key.objectid) {
  90                         p = &(*p)->rb_right;
  91                 } else {
  92                         spin_unlock(&info->block_group_cache_lock);
  93                         return -EEXIST;
  94                 }
  95         }
  96
  97         rb_link_node(&block_group->cache_node, parent, p);
  98         rb_insert_color(&block_group->cache_node,
  99                         &info->block_group_cache_tree);
 100         spin_unlock(&info->block_group_cache_lock);
 101
 102         return 0;
 103 }
 104
 105 /*
 106  * This will return the block group at or after bytenr if contains is 0, else
 107  * it will return the block group that contains the bytenr
 108  */
 109 static struct btrfs_block_group_cache *
 110 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 111                               int contains)
 112 {
 113         struct btrfs_block_group_cache *cache, *ret = NULL;
 114         struct rb_node *n;
 115         u64 end, start;
 116
 117         spin_lock(&info->block_group_cache_lock);
 118         n = info->block_group_cache_tree.rb_node;
 119
 120         while (n) {
 121                 cache = rb_entry(n, struct btrfs_block_group_cache,
 122                                  cache_node);
 123                 end = cache->key.objectid + cache->key.offset - 1;
 124                 start = cache->key.objectid;
 125
 126                 if (bytenr < start) {
 127                         if (!contains && (!ret || start < ret->key.objectid))
 128                                 ret = cache;
 129                         n = n->rb_left;
 130                 } else if (bytenr > start) {
 131                         if (contains && bytenr <= end) {
 132                                 ret = cache;
 133                                 break;
 134                         }
 135                         n = n->rb_right;
 136                 } else {
 137                         ret = cache;
 138                         break;
 139                 }
 140         }
 141         if (ret)
 142                 atomic_inc(&ret->count);
 143         spin_unlock(&info->block_group_cache_lock);
 144
 145         return ret;
 146 }
 147
 148 /*
 149  * this is only called by cache_block_group, since we could have freed extents
 150  * we need to check the pinned_extents for any extents that can't be used yet
 151  * since their free space will be released as soon as the transaction commits.
 152  */
 153 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 154                               struct btrfs_fs_info *info, u64 start, u64 end)
 155 {
 156         u64 extent_start, extent_end, size;
 157         int ret;
 158
 159         mutex_lock(&info->pinned_mutex);
 160         while (start < end) {
 161                 ret = find_first_extent_bit(&info->pinned_extents, start,
 162                                             &extent_start, &extent_end,
 163                                             EXTENT_DIRTY);
 164                 if (ret)
 165                         break;
 166
 167                 if (extent_start == start) {
 168                         start = extent_end + 1;
 169                 } else if (extent_start > start && extent_start < end) {
 170                         size = extent_start - start;
 171                         ret = btrfs_add_free_space(block_group, start,
 172                                                    size);
 173                         BUG_ON(ret);
 174                         start = extent_end + 1;
 175                 } else {
 176                         break;
 177                 }
 178         }
 179
 180         if (start < end) {
 181                 size = end - start;
 182                 ret = btrfs_add_free_space(block_group, start, size);
 183                 BUG_ON(ret);
 184         }
 185         mutex_unlock(&info->pinned_mutex);
 186
 187         return 0;
 188 }
 189
 190 static int remove_sb_from_cache(struct btrfs_root *root,
 191                                 struct btrfs_block_group_cache *cache)
 192 {
 193         u64 bytenr;
 194         u64 *logical;
 195         int stripe_len;
 196         int i, nr, ret;
 197
 198         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 199                 bytenr = btrfs_sb_offset(i);
 200                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 201                                        cache->key.objectid, bytenr, 0,
 202                                        &logical, &nr, &stripe_len);
 203                 BUG_ON(ret);
 204                 while (nr--) {
 205                         btrfs_remove_free_space(cache, logical[nr],
 206                                                 stripe_len);
 207                 }
 208                 kfree(logical);
 209         }
 210         return 0;
 211 }
 212
 213 static int cache_block_group(struct btrfs_root *root,
 214                              struct btrfs_block_group_cache *block_group)
 215 {
 216         struct btrfs_path *path;
 217         int ret = 0;
 218         struct btrfs_key key;
 219         struct extent_buffer *leaf;
 220         int slot;
 221         u64 last;
 222
 223         if (!block_group)
 224                 return 0;
 225
 226         root = root->fs_info->extent_root;
 227
 228         if (block_group->cached)
 229                 return 0;
 230
 231         path = btrfs_alloc_path();
 232         if (!path)
 233                 return -ENOMEM;
 234
 235         path->reada = 2;
 236         /*
 237          * we get into deadlocks with paths held by callers of this function.
 238          * since the alloc_mutex is protecting things right now, just
 239          * skip the locking here
 240          */
 241         path->skip_locking = 1;
 242         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 243         key.objectid = last;
 244         key.offset = 0;
 245         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 246         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 247         if (ret < 0)
 248                 goto err;
 249
 250         while(1) {
 251                 leaf = path->nodes[0];
 252                 slot = path->slots[0];
 253                 if (slot >= btrfs_header_nritems(leaf)) {
 254                         ret = btrfs_next_leaf(root, path);
 255                         if (ret < 0)
 256                                 goto err;
 257                         if (ret == 0)
 258                                 continue;
 259                         else
 260                                 break;
 261                 }
 262                 btrfs_item_key_to_cpu(leaf, &key, slot);
 263                 if (key.objectid < block_group->key.objectid)
 264                         goto next;
 265
 266                 if (key.objectid >= block_group->key.objectid +
 267                     block_group->key.offset)
 268                         break;
 269
 270                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 271                         add_new_free_space(block_group, root->fs_info, last,
 272                                            key.objectid);
 273
 274                         last = key.objectid + key.offset;
 275                 }
 276 next:
 277                 path->slots[0]++;
 278         }
 279
 280         add_new_free_space(block_group, root->fs_info, last,
 281                            block_group->key.objectid +
 282                            block_group->key.offset);
 283
 284         remove_sb_from_cache(root, block_group);
 285         block_group->cached = 1;
 286         ret = 0;
 287 err:
 288         btrfs_free_path(path);
 289         return ret;
 290 }
 291
 292 /*
 293  * return the block group that starts at or after bytenr
 294  */
 295 static struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
 296                                                        btrfs_fs_info *info,
 297                                                          u64 bytenr)
 298 {
 299         struct btrfs_block_group_cache *cache;
 300
 301         cache = block_group_cache_tree_search(info, bytenr, 0);
 302
 303         return cache;
 304 }
 305
 306 /*
 307  * return the block group that contains teh given bytenr
 308  */
 309 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
 310                                                          btrfs_fs_info *info,
 311                                                          u64 bytenr)
 312 {
 313         struct btrfs_block_group_cache *cache;
 314
 315         cache = block_group_cache_tree_search(info, bytenr, 1);
 316
 317         return cache;
 318 }
 319
 320 static inline void put_block_group(struct btrfs_block_group_cache *cache)
 321 {
 322         if (atomic_dec_and_test(&cache->count))
 323                 kfree(cache);
 324 }
 325
 326 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 327                                                   u64 flags)
 328 {
 329         struct list_head *head = &info->space_info;
 330         struct list_head *cur;
 331         struct btrfs_space_info *found;
 332         list_for_each(cur, head) {
 333                 found = list_entry(cur, struct btrfs_space_info, list);
 334                 if (found->flags == flags)
 335                         return found;
 336         }
 337         return NULL;
 338 }
 339
 340 static u64 div_factor(u64 num, int factor)
 341 {
 342         if (factor == 10)
 343                 return num;
 344         num *= factor;
 345         do_div(num, 10);
 346         return num;
 347 }
 348
 349 u64 btrfs_find_block_group(struct btrfs_root *root,
 350                            u64 search_start, u64 search_hint, int owner)
 351 {
 352         struct btrfs_block_group_cache *cache;
 353         u64 used;
 354         u64 last = max(search_hint, search_start);
 355         u64 group_start = 0;
 356         int full_search = 0;
 357         int factor = 9;
 358         int wrapped = 0;
 359 again:
 360         while (1) {
 361                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 362                 if (!cache)
 363                         break;
 364
 365                 spin_lock(&cache->lock);
 366                 last = cache->key.objectid + cache->key.offset;
 367                 used = btrfs_block_group_used(&cache->item);
 368
 369                 if ((full_search || !cache->ro) &&
 370                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 371                         if (used + cache->pinned + cache->reserved <
 372                             div_factor(cache->key.offset, factor)) {
 373                                 group_start = cache->key.objectid;
 374                                 spin_unlock(&cache->lock);
 375                                 put_block_group(cache);
 376                                 goto found;
 377                         }
 378                 }
 379                 spin_unlock(&cache->lock);
 380                 put_block_group(cache);
 381                 cond_resched();
 382         }
 383         if (!wrapped) {
 384                 last = search_start;
 385                 wrapped = 1;
 386                 goto again;
 387         }
 388         if (!full_search && factor < 10) {
 389                 last = search_start;
 390                 full_search = 1;
 391                 factor = 10;
 392                 goto again;
 393         }
 394 found:
 395         return group_start;
 396 }
 397
 398 /* simple helper to search for an existing extent at a given offset */
 399 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 400 {
 401         int ret;
 402         struct btrfs_key key;
 403         struct btrfs_path *path;
 404
 405         path = btrfs_alloc_path();
 406         BUG_ON(!path);
 407         key.objectid = start;
 408         key.offset = len;
 409         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 410         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 411                                 0, 0);
 412         btrfs_free_path(path);
 413         return ret;
 414 }
 415
 416 /*
 417  * Back reference rules.  Back refs have three main goals:
 418  *
 419  * 1) differentiate between all holders of references to an extent so that
 420  *    when a reference is dropped we can make sure it was a valid reference
 421  *    before freeing the extent.
 422  *
 423  * 2) Provide enough information to quickly find the holders of an extent
 424  *    if we notice a given block is corrupted or bad.
 425  *
 426  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 427  *    maintenance.  This is actually the same as #2, but with a slightly
 428  *    different use case.
 429  *
 430  * File extents can be referenced by:
 431  *
 432  * - multiple snapshots, subvolumes, or different generations in one subvol
 433  * - different files inside a single subvolume
 434  * - different offsets inside a file (bookend extents in file.c)
 435  *
 436  * The extent ref structure has fields for:
 437  *
 438  * - Objectid of the subvolume root
 439  * - Generation number of the tree holding the reference
 440  * - objectid of the file holding the reference
 441  * - number of references holding by parent node (alway 1 for tree blocks)
 442  *
 443  * Btree leaf may hold multiple references to a file extent. In most cases,
 444  * these references are from same file and the corresponding offsets inside
 445  * the file are close together.
 446  *
 447  * When a file extent is allocated the fields are filled in:
 448  *     (root_key.objectid, trans->transid, inode objectid, 1)
 449  *
 450  * When a leaf is cow'd new references are added for every file extent found
 451  * in the leaf.  It looks similar to the create case, but trans->transid will
 452  * be different when the block is cow'd.
 453  *
 454  *     (root_key.objectid, trans->transid, inode objectid,
 455  *      number of references in the leaf)
 456  *
 457  * When a file extent is removed either during snapshot deletion or
 458  * file truncation, we find the corresponding back reference and check
 459  * the following fields:
 460  *
 461  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 462  *      inode objectid)
 463  *
 464  * Btree extents can be referenced by:
 465  *
 466  * - Different subvolumes
 467  * - Different generations of the same subvolume
 468  *
 469  * When a tree block is created, back references are inserted:
 470  *
 471  * (root->root_key.objectid, trans->transid, level, 1)
 472  *
 473  * When a tree block is cow'd, new back references are added for all the
 474  * blocks it points to. If the tree block isn't in reference counted root,
 475  * the old back references are removed. These new back references are of
 476  * the form (trans->transid will have increased since creation):
 477  *
 478  * (root->root_key.objectid, trans->transid, level, 1)
 479  *
 480  * When a backref is in deleting, the following fields are checked:
 481  *
 482  * if backref was for a tree root:
 483  *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
 484  * else
 485  *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
 486  *
 487  * Back Reference Key composing:
 488  *
 489  * The key objectid corresponds to the first byte in the extent, the key
 490  * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
 491  * byte of parent extent. If a extent is tree root, the key offset is set
 492  * to the key objectid.
 493  */
 494
 495 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
 496                                           struct btrfs_root *root,
 497                                           struct btrfs_path *path,
 498                                           u64 bytenr, u64 parent,
 499                                           u64 ref_root, u64 ref_generation,
 500                                           u64 owner_objectid, int del)
 501 {
 502         struct btrfs_key key;
 503         struct btrfs_extent_ref *ref;
 504         struct extent_buffer *leaf;
 505         u64 ref_objectid;
 506         int ret;
 507
 508         key.objectid = bytenr;
 509         key.type = BTRFS_EXTENT_REF_KEY;
 510         key.offset = parent;
 511
 512         ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
 513         if (ret < 0)
 514                 goto out;
 515         if (ret > 0) {
 516                 ret = -ENOENT;
 517                 goto out;
 518         }
 519
 520         leaf = path->nodes[0];
 521         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 522         ref_objectid = btrfs_ref_objectid(leaf, ref);
 523         if (btrfs_ref_root(leaf, ref) != ref_root ||
 524             btrfs_ref_generation(leaf, ref) != ref_generation ||
 525             (ref_objectid != owner_objectid &&
 526              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 527                 ret = -EIO;
 528                 WARN_ON(1);
 529                 goto out;
 530         }
 531         ret = 0;
 532 out:
 533         return ret;
 534 }
 535
 536 /*
 537  * updates all the backrefs that are pending on update_list for the
 538  * extent_root
 539  */
 540 static int noinline update_backrefs(struct btrfs_trans_handle *trans,
 541                                     struct btrfs_root *extent_root,
 542                                     struct btrfs_path *path,
 543                                     struct list_head *update_list)
 544 {
 545         struct btrfs_key key;
 546         struct btrfs_extent_ref *ref;
 547         struct btrfs_fs_info *info = extent_root->fs_info;
 548         struct pending_extent_op *op;
 549         struct extent_buffer *leaf;
 550         int ret = 0;
 551         struct list_head *cur = update_list->next;
 552         u64 ref_objectid;
 553         u64 ref_root = extent_root->root_key.objectid;
 554
 555         op = list_entry(cur, struct pending_extent_op, list);
 556
 557 search:
 558         key.objectid = op->bytenr;
 559         key.type = BTRFS_EXTENT_REF_KEY;
 560         key.offset = op->orig_parent;
 561
 562         ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
 563         BUG_ON(ret);
 564
 565         leaf = path->nodes[0];
 566
 567 loop:
 568         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 569
 570         ref_objectid = btrfs_ref_objectid(leaf, ref);
 571
 572         if (btrfs_ref_root(leaf, ref) != ref_root ||
 573             btrfs_ref_generation(leaf, ref) != op->orig_generation ||
 574             (ref_objectid != op->level &&
 575              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 576                 printk(KERN_ERR "couldn't find %Lu, parent %Lu, root %Lu, "
 577                        "owner %u\n", op->bytenr, op->orig_parent,
 578                        ref_root, op->level);
 579                 btrfs_print_leaf(extent_root, leaf);
 580                 BUG();
 581         }
 582
 583         key.objectid = op->bytenr;
 584         key.offset = op->parent;
 585         key.type = BTRFS_EXTENT_REF_KEY;
 586         ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
 587         BUG_ON(ret);
 588         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 589         btrfs_set_ref_generation(leaf, ref, op->generation);
 590
 591         cur = cur->next;
 592
 593         list_del_init(&op->list);
 594         unlock_extent(&info->extent_ins, op->bytenr,
 595                       op->bytenr + op->num_bytes - 1, GFP_NOFS);
 596         kfree(op);
 597
 598         if (cur == update_list) {
 599                 btrfs_mark_buffer_dirty(path->nodes[0]);
 600                 btrfs_release_path(extent_root, path);
 601                 goto out;
 602         }
 603
 604         op = list_entry(cur, struct pending_extent_op, list);
 605
 606         path->slots[0]++;
 607         while (path->slots[0] < btrfs_header_nritems(leaf)) {
 608                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 609                 if (key.objectid == op->bytenr &&
 610                     key.type == BTRFS_EXTENT_REF_KEY)
 611                         goto loop;
 612                 path->slots[0]++;
 613         }
 614
 615         btrfs_mark_buffer_dirty(path->nodes[0]);
 616         btrfs_release_path(extent_root, path);
 617         goto search;
 618
 619 out:
 620         return 0;
 621 }
 622
 623 static int noinline insert_extents(struct btrfs_trans_handle *trans,
 624                                    struct btrfs_root *extent_root,
 625                                    struct btrfs_path *path,
 626                                    struct list_head *insert_list, int nr)
 627 {
 628         struct btrfs_key *keys;
 629         u32 *data_size;
 630         struct pending_extent_op *op;
 631         struct extent_buffer *leaf;
 632         struct list_head *cur = insert_list->next;
 633         struct btrfs_fs_info *info = extent_root->fs_info;
 634         u64 ref_root = extent_root->root_key.objectid;
 635         int i = 0, last = 0, ret;
 636         int total = nr * 2;
 637
 638         if (!nr)
 639                 return 0;
 640
 641         keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
 642         if (!keys)
 643                 return -ENOMEM;
 644
 645         data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
 646         if (!data_size) {
 647                 kfree(keys);
 648                 return -ENOMEM;
 649         }
 650
 651         list_for_each_entry(op, insert_list, list) {
 652                 keys[i].objectid = op->bytenr;
 653                 keys[i].offset = op->num_bytes;
 654                 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
 655                 data_size[i] = sizeof(struct btrfs_extent_item);
 656                 i++;
 657
 658                 keys[i].objectid = op->bytenr;
 659                 keys[i].offset = op->parent;
 660                 keys[i].type = BTRFS_EXTENT_REF_KEY;
 661                 data_size[i] = sizeof(struct btrfs_extent_ref);
 662                 i++;
 663         }
 664
 665         op = list_entry(cur, struct pending_extent_op, list);
 666         i = 0;
 667         while (i < total) {
 668                 int c;
 669                 ret = btrfs_insert_some_items(trans, extent_root, path,
 670                                               keys+i, data_size+i, total-i);
 671                 BUG_ON(ret < 0);
 672
 673                 if (last && ret > 1)
 674                         BUG();
 675
 676                 leaf = path->nodes[0];
 677                 for (c = 0; c < ret; c++) {
 678                         int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
 679
 680                         /*
 681                          * if the first item we inserted was a backref, then
 682                          * the EXTENT_ITEM will be the odd c's, else it will
 683                          * be the even c's
 684                          */
 685                         if ((ref_first && (c % 2)) ||
 686                             (!ref_first && !(c % 2))) {
 687                                 struct btrfs_extent_item *itm;
 688
 689                                 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
 690                                                      struct btrfs_extent_item);
 691                                 btrfs_set_extent_refs(path->nodes[0], itm, 1);
 692                                 op->del++;
 693                         } else {
 694                                 struct btrfs_extent_ref *ref;
 695
 696                                 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
 697                                                      struct btrfs_extent_ref);
 698                                 btrfs_set_ref_root(leaf, ref, ref_root);
 699                                 btrfs_set_ref_generation(leaf, ref,
 700                                                          op->generation);
 701                                 btrfs_set_ref_objectid(leaf, ref, op->level);
 702                                 btrfs_set_ref_num_refs(leaf, ref, 1);
 703                                 op->del++;
 704                         }
 705
 706                         /*
 707                          * using del to see when its ok to free up the
 708                          * pending_extent_op.  In the case where we insert the
 709                          * last item on the list in order to help do batching
 710                          * we need to not free the extent op until we actually
 711                          * insert the extent_item
 712                          */
 713                         if (op->del == 2) {
 714                                 unlock_extent(&info->extent_ins, op->bytenr,
 715                                               op->bytenr + op->num_bytes - 1,
 716                                               GFP_NOFS);
 717                                 cur = cur->next;
 718                                 list_del_init(&op->list);
 719                                 kfree(op);
 720                                 if (cur != insert_list)
 721                                         op = list_entry(cur,
 722                                                 struct pending_extent_op,
 723                                                 list);
 724                         }
 725                 }
 726                 btrfs_mark_buffer_dirty(leaf);
 727                 btrfs_release_path(extent_root, path);
 728
 729                 /*
 730                  * Ok backref's and items usually go right next to eachother,
 731                  * but if we could only insert 1 item that means that we
 732                  * inserted on the end of a leaf, and we have no idea what may
 733                  * be on the next leaf so we just play it safe.  In order to
 734                  * try and help this case we insert the last thing on our
 735                  * insert list so hopefully it will end up being the last
 736                  * thing on the leaf and everything else will be before it,
 737                  * which will let us insert a whole bunch of items at the same
 738                  * time.
 739                  */
 740                 if (ret == 1 && !last && (i + ret < total)) {
 741                         /*
 742                          * last: where we will pick up the next time around
 743                          * i: our current key to insert, will be total - 1
 744                          * cur: the current op we are screwing with
 745                          * op: duh
 746                          */
 747                         last = i + ret;
 748                         i = total - 1;
 749                         cur = insert_list->prev;
 750                         op = list_entry(cur, struct pending_extent_op, list);
 751                 } else if (last) {
 752                         /*
 753                          * ok we successfully inserted the last item on the
 754                          * list, lets reset everything
 755                          *
 756                          * i: our current key to insert, so where we left off
 757                          *    last time
 758                          * last: done with this
 759                          * cur: the op we are messing with
 760                          * op: duh
 761                          * total: since we inserted the last key, we need to
 762                          *        decrement total so we dont overflow
 763                          */
 764                         i = last;
 765                         last = 0;
 766                         total--;
 767                         if (i < total) {
 768                                 cur = insert_list->next;
 769                                 op = list_entry(cur, struct pending_extent_op,
 770                                                 list);
 771                         }
 772                 } else {
 773                         i += ret;
 774                 }
 775
 776                 cond_resched();
 777         }
 778         ret = 0;
 779         kfree(keys);
 780         kfree(data_size);
 781         return ret;
 782 }
 783
 784 static int noinline insert_extent_backref(struct btrfs_trans_handle *trans,
 785                                           struct btrfs_root *root,
 786                                           struct btrfs_path *path,
 787                                           u64 bytenr, u64 parent,
 788                                           u64 ref_root, u64 ref_generation,
 789                                           u64 owner_objectid)
 790 {
 791         struct btrfs_key key;
 792         struct extent_buffer *leaf;
 793         struct btrfs_extent_ref *ref;
 794         u32 num_refs;
 795         int ret;
 796
 797         key.objectid = bytenr;
 798         key.type = BTRFS_EXTENT_REF_KEY;
 799         key.offset = parent;
 800
 801         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
 802         if (ret == 0) {
 803                 leaf = path->nodes[0];
 804                 ref = btrfs_item_ptr(leaf, path->slots[0],
 805                                      struct btrfs_extent_ref);
 806                 btrfs_set_ref_root(leaf, ref, ref_root);
 807                 btrfs_set_ref_generation(leaf, ref, ref_generation);
 808                 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
 809                 btrfs_set_ref_num_refs(leaf, ref, 1);
 810         } else if (ret == -EEXIST) {
 811                 u64 existing_owner;
 812                 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
 813                 leaf = path->nodes[0];
 814                 ref = btrfs_item_ptr(leaf, path->slots[0],
 815                                      struct btrfs_extent_ref);
 816                 if (btrfs_ref_root(leaf, ref) != ref_root ||
 817                     btrfs_ref_generation(leaf, ref) != ref_generation) {
 818                         ret = -EIO;
 819                         WARN_ON(1);
 820                         goto out;
 821                 }
 822
 823                 num_refs = btrfs_ref_num_refs(leaf, ref);
 824                 BUG_ON(num_refs == 0);
 825                 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
 826
 827                 existing_owner = btrfs_ref_objectid(leaf, ref);
 828                 if (existing_owner != owner_objectid &&
 829                     existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
 830                         btrfs_set_ref_objectid(leaf, ref,
 831                                         BTRFS_MULTIPLE_OBJECTIDS);
 832                 }
 833                 ret = 0;
 834         } else {
 835                 goto out;
 836         }
 837         btrfs_mark_buffer_dirty(path->nodes[0]);
 838 out:
 839         btrfs_release_path(root, path);
 840         return ret;
 841 }
 842
 843 static int noinline remove_extent_backref(struct btrfs_trans_handle *trans,
 844                                           struct btrfs_root *root,
 845                                           struct btrfs_path *path)
 846 {
 847         struct extent_buffer *leaf;
 848         struct btrfs_extent_ref *ref;
 849         u32 num_refs;
 850         int ret = 0;
 851
 852         leaf = path->nodes[0];
 853         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 854         num_refs = btrfs_ref_num_refs(leaf, ref);
 855         BUG_ON(num_refs == 0);
 856         num_refs -= 1;
 857         if (num_refs == 0) {
 858                 ret = btrfs_del_item(trans, root, path);
 859         } else {
 860                 btrfs_set_ref_num_refs(leaf, ref, num_refs);
 861                 btrfs_mark_buffer_dirty(leaf);
 862         }
 863         btrfs_release_path(root, path);
 864         return ret;
 865 }
 866
 867 #ifdef BIO_RW_DISCARD
 868 static void btrfs_issue_discard(struct block_device *bdev,
 869                                 u64 start, u64 len)
 870 {
 871 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,28)
 872         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
 873 #else
 874         blkdev_issue_discard(bdev, start >> 9, len >> 9);
 875 #endif
 876 }
 877 #endif
 878
 879 static int noinline free_extents(struct btrfs_trans_handle *trans,
 880                                  struct btrfs_root *extent_root,
 881                                  struct list_head *del_list)
 882 {
 883         struct btrfs_fs_info *info = extent_root->fs_info;
 884         struct btrfs_path *path;
 885         struct btrfs_key key, found_key;
 886         struct extent_buffer *leaf;
 887         struct list_head *cur;
 888         struct pending_extent_op *op;
 889         struct btrfs_extent_item *ei;
 890         int ret, num_to_del, extent_slot = 0, found_extent = 0;
 891         u32 refs;
 892         u64 bytes_freed = 0;
 893
 894         path = btrfs_alloc_path();
 895         if (!path)
 896                 return -ENOMEM;
 897         path->reada = 1;
 898
 899 search:
 900         /* search for the backref for the current ref we want to delete */
 901         cur = del_list->next;
 902         op = list_entry(cur, struct pending_extent_op, list);
 903         ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
 904                                     op->orig_parent,
 905                                     extent_root->root_key.objectid,
 906                                     op->orig_generation, op->level, 1);
 907         if (ret) {
 908                 printk("Unable to find backref byte nr %Lu root %Lu gen %Lu "
 909                        "owner %u\n", op->bytenr,
 910                        extent_root->root_key.objectid, op->orig_generation,
 911                        op->level);
 912                 btrfs_print_leaf(extent_root, path->nodes[0]);
 913                 WARN_ON(1);
 914                 goto out;
 915         }
 916
 917         extent_slot = path->slots[0];
 918         num_to_del = 1;
 919         found_extent = 0;
 920
 921         /*
 922          * if we aren't the first item on the leaf we can move back one and see
 923          * if our ref is right next to our extent item
 924          */
 925         if (likely(extent_slot)) {
 926                 extent_slot--;
 927                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 928                                       extent_slot);
 929                 if (found_key.objectid == op->bytenr &&
 930                     found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 931                     found_key.offset == op->num_bytes) {
 932                         num_to_del++;
 933                         found_extent = 1;
 934                 }
 935         }
 936
 937         /*
 938          * if we didn't find the extent we need to delete the backref and then
 939          * search for the extent item key so we can update its ref count
 940          */
 941         if (!found_extent) {
 942                 key.objectid = op->bytenr;
 943                 key.type = BTRFS_EXTENT_ITEM_KEY;
 944                 key.offset = op->num_bytes;
 945
 946                 ret = remove_extent_backref(trans, extent_root, path);
 947                 BUG_ON(ret);
 948                 btrfs_release_path(extent_root, path);
 949                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
 950                 BUG_ON(ret);
 951                 extent_slot = path->slots[0];
 952         }
 953
 954         /* this is where we update the ref count for the extent */
 955         leaf = path->nodes[0];
 956         ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
 957         refs = btrfs_extent_refs(leaf, ei);
 958         BUG_ON(refs == 0);
 959         refs--;
 960         btrfs_set_extent_refs(leaf, ei, refs);
 961
 962         btrfs_mark_buffer_dirty(leaf);
 963
 964         /*
 965          * This extent needs deleting.  The reason cur_slot is extent_slot +
 966          * num_to_del is because extent_slot points to the slot where the extent
 967          * is, and if the backref was not right next to the extent we will be
 968          * deleting at least 1 item, and will want to start searching at the
 969          * slot directly next to extent_slot.  However if we did find the
 970          * backref next to the extent item them we will be deleting at least 2
 971          * items and will want to start searching directly after the ref slot
 972          */
 973         if (!refs) {
 974                 struct list_head *pos, *n, *end;
 975                 int cur_slot = extent_slot+num_to_del;
 976                 u64 super_used;
 977                 u64 root_used;
 978
 979                 path->slots[0] = extent_slot;
 980                 bytes_freed = op->num_bytes;
 981
 982                 mutex_lock(&info->pinned_mutex);
 983                 ret = pin_down_bytes(trans, extent_root, op->bytenr,
 984                                      op->num_bytes, op->level >=
 985                                      BTRFS_FIRST_FREE_OBJECTID);
 986                 mutex_unlock(&info->pinned_mutex);
 987                 BUG_ON(ret < 0);
 988                 op->del = ret;
 989
 990                 /*
 991                  * we need to see if we can delete multiple things at once, so
 992                  * start looping through the list of extents we are wanting to
 993                  * delete and see if their extent/backref's are right next to
 994                  * eachother and the extents only have 1 ref
 995                  */
 996                 for (pos = cur->next; pos != del_list; pos = pos->next) {
 997                         struct pending_extent_op *tmp;
 998
 999                         tmp = list_entry(pos, struct pending_extent_op, list);
1000
1001                         /* we only want to delete extent+ref at this stage */
1002                         if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1003                                 break;
1004
1005                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1006                         if (found_key.objectid != tmp->bytenr ||
1007                             found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1008                             found_key.offset != tmp->num_bytes)
1009                                 break;
1010
1011                         /* check to make sure this extent only has one ref */
1012                         ei = btrfs_item_ptr(leaf, cur_slot,
1013                                             struct btrfs_extent_item);
1014                         if (btrfs_extent_refs(leaf, ei) != 1)
1015                                 break;
1016
1017                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1018                         if (found_key.objectid != tmp->bytenr ||
1019                             found_key.type != BTRFS_EXTENT_REF_KEY ||
1020                             found_key.offset != tmp->orig_parent)
1021                                 break;
1022
1023                         /*
1024                          * the ref is right next to the extent, we can set the
1025                          * ref count to 0 since we will delete them both now
1026                          */
1027                         btrfs_set_extent_refs(leaf, ei, 0);
1028
1029                         /* pin down the bytes for this extent */
1030                         mutex_lock(&info->pinned_mutex);
1031                         ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1032                                              tmp->num_bytes, tmp->level >=
1033                                              BTRFS_FIRST_FREE_OBJECTID);
1034                         mutex_unlock(&info->pinned_mutex);
1035                         BUG_ON(ret < 0);
1036
1037                         /*
1038                          * use the del field to tell if we need to go ahead and
1039                          * free up the extent when we delete the item or not.
1040                          */
1041                         tmp->del = ret;
1042                         bytes_freed += tmp->num_bytes;
1043
1044                         num_to_del += 2;
1045                         cur_slot += 2;
1046                 }
1047                 end = pos;
1048
1049                 /* update the free space counters */
1050                 spin_lock(&info->delalloc_lock);
1051                 super_used = btrfs_super_bytes_used(&info->super_copy);
1052                 btrfs_set_super_bytes_used(&info->super_copy,
1053                                            super_used - bytes_freed);
1054
1055                 root_used = btrfs_root_used(&extent_root->root_item);
1056                 btrfs_set_root_used(&extent_root->root_item,
1057                                     root_used - bytes_freed);
1058                 spin_unlock(&info->delalloc_lock);
1059
1060                 /* delete the items */
1061                 ret = btrfs_del_items(trans, extent_root, path,
1062                                       path->slots[0], num_to_del);
1063                 BUG_ON(ret);
1064
1065                 /*
1066                  * loop through the extents we deleted and do the cleanup work
1067                  * on them
1068                  */
1069                 for (pos = cur, n = pos->next; pos != end;
1070                      pos = n, n = pos->next) {
1071                         struct pending_extent_op *tmp;
1072 #ifdef BIO_RW_DISCARD
1073                         u64 map_length;
1074                         struct btrfs_multi_bio *multi = NULL;
1075 #endif
1076                         tmp = list_entry(pos, struct pending_extent_op, list);
1077
1078                         /*
1079                          * remember tmp->del tells us wether or not we pinned
1080                          * down the extent
1081                          */
1082                         ret = update_block_group(trans, extent_root,
1083                                                  tmp->bytenr, tmp->num_bytes, 0,
1084                                                  tmp->del);
1085                         BUG_ON(ret);
1086
1087 #ifdef BIO_RW_DISCARD
1088                         map_length = tmp->num_bytes;
1089                         ret = btrfs_map_block(&info->mapping_tree, READ,
1090                                               tmp->bytenr, &map_length, &multi,
1091                                               0);
1092                         if (!ret) {
1093                                 struct btrfs_bio_stripe *stripe;
1094                                 int i;
1095
1096                                 stripe = multi->stripes;
1097
1098                                 if (map_length > tmp->num_bytes)
1099                                         map_length = tmp->num_bytes;
1100
1101                                 for (i = 0; i < multi->num_stripes;
1102                                      i++, stripe++)
1103                                         btrfs_issue_discard(stripe->dev->bdev,
1104                                                             stripe->physical,
1105                                                             map_length);
1106                                 kfree(multi);
1107                         }
1108 #endif
1109                         list_del_init(&tmp->list);
1110                         unlock_extent(&info->extent_ins, tmp->bytenr,
1111                                       tmp->bytenr + tmp->num_bytes - 1,
1112                                       GFP_NOFS);
1113                         kfree(tmp);
1114                 }
1115         } else if (refs && found_extent) {
1116                 /*
1117                  * the ref and extent were right next to eachother, but the
1118                  * extent still has a ref, so just free the backref and keep
1119                  * going
1120                  */
1121                 ret = remove_extent_backref(trans, extent_root, path);
1122                 BUG_ON(ret);
1123
1124                 list_del_init(&op->list);
1125                 unlock_extent(&info->extent_ins, op->bytenr,
1126                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1127                 kfree(op);
1128         } else {
1129                 /*
1130                  * the extent has multiple refs and the backref we were looking
1131                  * for was not right next to it, so just unlock and go next,
1132                  * we're good to go
1133                  */
1134                 list_del_init(&op->list);
1135                 unlock_extent(&info->extent_ins, op->bytenr,
1136                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1137                 kfree(op);
1138         }
1139
1140         btrfs_release_path(extent_root, path);
1141         if (!list_empty(del_list))
1142                 goto search;
1143
1144 out:
1145         btrfs_free_path(path);
1146         return ret;
1147 }
1148
1149 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1150                                      struct btrfs_root *root, u64 bytenr,
1151                                      u64 orig_parent, u64 parent,
1152                                      u64 orig_root, u64 ref_root,
1153                                      u64 orig_generation, u64 ref_generation,
1154                                      u64 owner_objectid)
1155 {
1156         int ret;
1157         struct btrfs_root *extent_root = root->fs_info->extent_root;
1158         struct btrfs_path *path;
1159
1160         if (root == root->fs_info->extent_root) {
1161                 struct pending_extent_op *extent_op;
1162                 u64 num_bytes;
1163
1164                 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1165                 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1166                 mutex_lock(&root->fs_info->extent_ins_mutex);
1167                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1168                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1169                         u64 priv;
1170                         ret = get_state_private(&root->fs_info->extent_ins,
1171                                                 bytenr, &priv);
1172                         BUG_ON(ret);
1173                         extent_op = (struct pending_extent_op *)
1174                                                         (unsigned long)priv;
1175                         BUG_ON(extent_op->parent != orig_parent);
1176                         BUG_ON(extent_op->generation != orig_generation);
1177
1178                         extent_op->parent = parent;
1179                         extent_op->generation = ref_generation;
1180                 } else {
1181                         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1182                         BUG_ON(!extent_op);
1183
1184                         extent_op->type = PENDING_BACKREF_UPDATE;
1185                         extent_op->bytenr = bytenr;
1186                         extent_op->num_bytes = num_bytes;
1187                         extent_op->parent = parent;
1188                         extent_op->orig_parent = orig_parent;
1189                         extent_op->generation = ref_generation;
1190                         extent_op->orig_generation = orig_generation;
1191                         extent_op->level = (int)owner_objectid;
1192                         INIT_LIST_HEAD(&extent_op->list);
1193                         extent_op->del = 0;
1194
1195                         set_extent_bits(&root->fs_info->extent_ins,
1196                                         bytenr, bytenr + num_bytes - 1,
1197                                         EXTENT_WRITEBACK, GFP_NOFS);
1198                         set_state_private(&root->fs_info->extent_ins,
1199                                           bytenr, (unsigned long)extent_op);
1200                 }
1201                 mutex_unlock(&root->fs_info->extent_ins_mutex);
1202                 return 0;
1203         }
1204
1205         path = btrfs_alloc_path();
1206         if (!path)
1207                 return -ENOMEM;
1208         ret = lookup_extent_backref(trans, extent_root, path,
1209                                     bytenr, orig_parent, orig_root,
1210                                     orig_generation, owner_objectid, 1);
1211         if (ret)
1212                 goto out;
1213         ret = remove_extent_backref(trans, extent_root, path);
1214         if (ret)
1215                 goto out;
1216         ret = insert_extent_backref(trans, extent_root, path, bytenr,
1217                                     parent, ref_root, ref_generation,
1218                                     owner_objectid);
1219         BUG_ON(ret);
1220         finish_current_insert(trans, extent_root, 0);
1221         del_pending_extents(trans, extent_root, 0);
1222 out:
1223         btrfs_free_path(path);
1224         return ret;
1225 }
1226
1227 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1228                             struct btrfs_root *root, u64 bytenr,
1229                             u64 orig_parent, u64 parent,
1230                             u64 ref_root, u64 ref_generation,
1231                             u64 owner_objectid)
1232 {
1233         int ret;
1234         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1235             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1236                 return 0;
1237         ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1238                                         parent, ref_root, ref_root,
1239                                         ref_generation, ref_generation,
1240                                         owner_objectid);
1241         return ret;
1242 }
1243
1244 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1245                                   struct btrfs_root *root, u64 bytenr,
1246                                   u64 orig_parent, u64 parent,
1247                                   u64 orig_root, u64 ref_root,
1248                                   u64 orig_generation, u64 ref_generation,
1249                                   u64 owner_objectid)
1250 {
1251         struct btrfs_path *path;
1252         int ret;
1253         struct btrfs_key key;
1254         struct extent_buffer *l;
1255         struct btrfs_extent_item *item;
1256         u32 refs;
1257
1258         path = btrfs_alloc_path();
1259         if (!path)
1260                 return -ENOMEM;
1261
1262         path->reada = 1;
1263         key.objectid = bytenr;
1264         key.type = BTRFS_EXTENT_ITEM_KEY;
1265         key.offset = (u64)-1;
1266
1267         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1268                                 0, 1);
1269         if (ret < 0)
1270                 return ret;
1271         BUG_ON(ret == 0 || path->slots[0] == 0);
1272
1273         path->slots[0]--;
1274         l = path->nodes[0];
1275
1276         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1277         if (key.objectid != bytenr) {
1278                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1279                 printk("wanted %Lu found %Lu\n", bytenr, key.objectid);
1280                 BUG();
1281         }
1282         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1283
1284         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1285         refs = btrfs_extent_refs(l, item);
1286         btrfs_set_extent_refs(l, item, refs + 1);
1287         btrfs_mark_buffer_dirty(path->nodes[0]);
1288
1289         btrfs_release_path(root->fs_info->extent_root, path);
1290
1291         path->reada = 1;
1292         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1293                                     path, bytenr, parent,
1294                                     ref_root, ref_generation,
1295                                     owner_objectid);
1296         BUG_ON(ret);
1297         finish_current_insert(trans, root->fs_info->extent_root, 0);
1298         del_pending_extents(trans, root->fs_info->extent_root, 0);
1299
1300         btrfs_free_path(path);
1301         return 0;
1302 }
1303
1304 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1305                          struct btrfs_root *root,
1306                          u64 bytenr, u64 num_bytes, u64 parent,
1307                          u64 ref_root, u64 ref_generation,
1308                          u64 owner_objectid)
1309 {
1310         int ret;
1311         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1312             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1313                 return 0;
1314         ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1315                                      0, ref_root, 0, ref_generation,
1316                                      owner_objectid);
1317         return ret;
1318 }
1319
1320 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1321                          struct btrfs_root *root)
1322 {
1323         finish_current_insert(trans, root->fs_info->extent_root, 1);
1324         del_pending_extents(trans, root->fs_info->extent_root, 1);
1325         return 0;
1326 }
1327
1328 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1329                             struct btrfs_root *root, u64 bytenr,
1330                             u64 num_bytes, u32 *refs)
1331 {
1332         struct btrfs_path *path;
1333         int ret;
1334         struct btrfs_key key;
1335         struct extent_buffer *l;
1336         struct btrfs_extent_item *item;
1337
1338         WARN_ON(num_bytes < root->sectorsize);
1339         path = btrfs_alloc_path();
1340         path->reada = 1;
1341         key.objectid = bytenr;
1342         key.offset = num_bytes;
1343         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1344         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1345                                 0, 0);
1346         if (ret < 0)
1347                 goto out;
1348         if (ret != 0) {
1349                 btrfs_print_leaf(root, path->nodes[0]);
1350                 printk("failed to find block number %Lu\n", bytenr);
1351                 BUG();
1352         }
1353         l = path->nodes[0];
1354         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1355         *refs = btrfs_extent_refs(l, item);
1356 out:
1357         btrfs_free_path(path);
1358         return 0;
1359 }
1360
1361 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1362                           struct btrfs_root *root, u64 objectid, u64 bytenr)
1363 {
1364         struct btrfs_root *extent_root = root->fs_info->extent_root;
1365         struct btrfs_path *path;
1366         struct extent_buffer *leaf;
1367         struct btrfs_extent_ref *ref_item;
1368         struct btrfs_key key;
1369         struct btrfs_key found_key;
1370         u64 ref_root;
1371         u64 last_snapshot;
1372         u32 nritems;
1373         int ret;
1374
1375         key.objectid = bytenr;
1376         key.offset = (u64)-1;
1377         key.type = BTRFS_EXTENT_ITEM_KEY;
1378
1379         path = btrfs_alloc_path();
1380         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1381         if (ret < 0)
1382                 goto out;
1383         BUG_ON(ret == 0);
1384
1385         ret = -ENOENT;
1386         if (path->slots[0] == 0)
1387                 goto out;
1388
1389         path->slots[0]--;
1390         leaf = path->nodes[0];
1391         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1392
1393         if (found_key.objectid != bytenr ||
1394             found_key.type != BTRFS_EXTENT_ITEM_KEY)
1395                 goto out;
1396
1397         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1398         while (1) {
1399                 leaf = path->nodes[0];
1400                 nritems = btrfs_header_nritems(leaf);
1401                 if (path->slots[0] >= nritems) {
1402                         ret = btrfs_next_leaf(extent_root, path);
1403                         if (ret < 0)
1404                                 goto out;
1405                         if (ret == 0)
1406                                 continue;
1407                         break;
1408                 }
1409                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1410                 if (found_key.objectid != bytenr)
1411                         break;
1412
1413                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1414                         path->slots[0]++;
1415                         continue;
1416                 }
1417
1418                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1419                                           struct btrfs_extent_ref);
1420                 ref_root = btrfs_ref_root(leaf, ref_item);
1421                 if ((ref_root != root->root_key.objectid &&
1422                      ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1423                      objectid != btrfs_ref_objectid(leaf, ref_item)) {
1424                         ret = 1;
1425                         goto out;
1426                 }
1427                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1428                         ret = 1;
1429                         goto out;
1430                 }
1431
1432                 path->slots[0]++;
1433         }
1434         ret = 0;
1435 out:
1436         btrfs_free_path(path);
1437         return ret;
1438 }
1439
1440 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1441                     struct extent_buffer *buf, u32 nr_extents)
1442 {
1443         struct btrfs_key key;
1444         struct btrfs_file_extent_item *fi;
1445         u64 root_gen;
1446         u32 nritems;
1447         int i;
1448         int level;
1449         int ret = 0;
1450         int shared = 0;
1451
1452         if (!root->ref_cows)
1453                 return 0;
1454
1455         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1456                 shared = 0;
1457                 root_gen = root->root_key.offset;
1458         } else {
1459                 shared = 1;
1460                 root_gen = trans->transid - 1;
1461         }
1462
1463         level = btrfs_header_level(buf);
1464         nritems = btrfs_header_nritems(buf);
1465
1466         if (level == 0) {
1467                 struct btrfs_leaf_ref *ref;
1468                 struct btrfs_extent_info *info;
1469
1470                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1471                 if (!ref) {
1472                         ret = -ENOMEM;
1473                         goto out;
1474                 }
1475
1476                 ref->root_gen = root_gen;
1477                 ref->bytenr = buf->start;
1478                 ref->owner = btrfs_header_owner(buf);
1479                 ref->generation = btrfs_header_generation(buf);
1480                 ref->nritems = nr_extents;
1481                 info = ref->extents;
1482
1483                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1484                         u64 disk_bytenr;
1485                         btrfs_item_key_to_cpu(buf, &key, i);
1486                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1487                                 continue;
1488                         fi = btrfs_item_ptr(buf, i,
1489                                             struct btrfs_file_extent_item);
1490                         if (btrfs_file_extent_type(buf, fi) ==
1491                             BTRFS_FILE_EXTENT_INLINE)
1492                                 continue;
1493                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1494                         if (disk_bytenr == 0)
1495                                 continue;
1496
1497                         info->bytenr = disk_bytenr;
1498                         info->num_bytes =
1499                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1500                         info->objectid = key.objectid;
1501                         info->offset = key.offset;
1502                         info++;
1503                 }
1504
1505                 ret = btrfs_add_leaf_ref(root, ref, shared);
1506                 if (ret == -EEXIST && shared) {
1507                         struct btrfs_leaf_ref *old;
1508                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1509                         BUG_ON(!old);
1510                         btrfs_remove_leaf_ref(root, old);
1511                         btrfs_free_leaf_ref(root, old);
1512                         ret = btrfs_add_leaf_ref(root, ref, shared);
1513                 }
1514                 WARN_ON(ret);
1515                 btrfs_free_leaf_ref(root, ref);
1516         }
1517 out:
1518         return ret;
1519 }
1520
1521 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1522                   struct extent_buffer *orig_buf, struct extent_buffer *buf,
1523                   u32 *nr_extents)
1524 {
1525         u64 bytenr;
1526         u64 ref_root;
1527         u64 orig_root;
1528         u64 ref_generation;
1529         u64 orig_generation;
1530         u32 nritems;
1531         u32 nr_file_extents = 0;
1532         struct btrfs_key key;
1533         struct btrfs_file_extent_item *fi;
1534         int i;
1535         int level;
1536         int ret = 0;
1537         int faili = 0;
1538         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1539                             u64, u64, u64, u64, u64, u64, u64, u64);
1540
1541         ref_root = btrfs_header_owner(buf);
1542         ref_generation = btrfs_header_generation(buf);
1543         orig_root = btrfs_header_owner(orig_buf);
1544         orig_generation = btrfs_header_generation(orig_buf);
1545
1546         nritems = btrfs_header_nritems(buf);
1547         level = btrfs_header_level(buf);
1548
1549         if (root->ref_cows) {
1550                 process_func = __btrfs_inc_extent_ref;
1551         } else {
1552                 if (level == 0 &&
1553                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1554                         goto out;
1555                 if (level != 0 &&
1556                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1557                         goto out;
1558                 process_func = __btrfs_update_extent_ref;
1559         }
1560
1561         for (i = 0; i < nritems; i++) {
1562                 cond_resched();
1563                 if (level == 0) {
1564                         btrfs_item_key_to_cpu(buf, &key, i);
1565                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1566                                 continue;
1567                         fi = btrfs_item_ptr(buf, i,
1568                                             struct btrfs_file_extent_item);
1569                         if (btrfs_file_extent_type(buf, fi) ==
1570                             BTRFS_FILE_EXTENT_INLINE)
1571                                 continue;
1572                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1573                         if (bytenr == 0)
1574                                 continue;
1575
1576                         nr_file_extents++;
1577
1578                         ret = process_func(trans, root, bytenr,
1579                                            orig_buf->start, buf->start,
1580                                            orig_root, ref_root,
1581                                            orig_generation, ref_generation,
1582                                            key.objectid);
1583
1584                         if (ret) {
1585                                 faili = i;
1586                                 WARN_ON(1);
1587                                 goto fail;
1588                         }
1589                 } else {
1590                         bytenr = btrfs_node_blockptr(buf, i);
1591                         ret = process_func(trans, root, bytenr,
1592                                            orig_buf->start, buf->start,
1593                                            orig_root, ref_root,
1594                                            orig_generation, ref_generation,
1595                                            level - 1);
1596                         if (ret) {
1597                                 faili = i;
1598                                 WARN_ON(1);
1599                                 goto fail;
1600                         }
1601                 }
1602         }
1603 out:
1604         if (nr_extents) {
1605                 if (level == 0)
1606                         *nr_extents = nr_file_extents;
1607                 else
1608                         *nr_extents = nritems;
1609         }
1610         return 0;
1611 fail:
1612         WARN_ON(1);
1613         return ret;
1614 }
1615
1616 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1617                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1618                      struct extent_buffer *buf, int start_slot, int nr)
1619
1620 {
1621         u64 bytenr;
1622         u64 ref_root;
1623         u64 orig_root;
1624         u64 ref_generation;
1625         u64 orig_generation;
1626         struct btrfs_key key;
1627         struct btrfs_file_extent_item *fi;
1628         int i;
1629         int ret;
1630         int slot;
1631         int level;
1632
1633         BUG_ON(start_slot < 0);
1634         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1635
1636         ref_root = btrfs_header_owner(buf);
1637         ref_generation = btrfs_header_generation(buf);
1638         orig_root = btrfs_header_owner(orig_buf);
1639         orig_generation = btrfs_header_generation(orig_buf);
1640         level = btrfs_header_level(buf);
1641
1642         if (!root->ref_cows) {
1643                 if (level == 0 &&
1644                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1645                         return 0;
1646                 if (level != 0 &&
1647                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1648                         return 0;
1649         }
1650
1651         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1652                 cond_resched();
1653                 if (level == 0) {
1654                         btrfs_item_key_to_cpu(buf, &key, slot);
1655                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1656                                 continue;
1657                         fi = btrfs_item_ptr(buf, slot,
1658                                             struct btrfs_file_extent_item);
1659                         if (btrfs_file_extent_type(buf, fi) ==
1660                             BTRFS_FILE_EXTENT_INLINE)
1661                                 continue;
1662                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1663                         if (bytenr == 0)
1664                                 continue;
1665                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1666                                             orig_buf->start, buf->start,
1667                                             orig_root, ref_root,
1668                                             orig_generation, ref_generation,
1669                                             key.objectid);
1670                         if (ret)
1671                                 goto fail;
1672                 } else {
1673                         bytenr = btrfs_node_blockptr(buf, slot);
1674                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1675                                             orig_buf->start, buf->start,
1676                                             orig_root, ref_root,
1677                                             orig_generation, ref_generation,
1678                                             level - 1);
1679                         if (ret)
1680                                 goto fail;
1681                 }
1682         }
1683         return 0;
1684 fail:
1685         WARN_ON(1);
1686         return -1;
1687 }
1688
1689 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1690                                  struct btrfs_root *root,
1691                                  struct btrfs_path *path,
1692                                  struct btrfs_block_group_cache *cache)
1693 {
1694         int ret;
1695         int pending_ret;
1696         struct btrfs_root *extent_root = root->fs_info->extent_root;
1697         unsigned long bi;
1698         struct extent_buffer *leaf;
1699
1700         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1701         if (ret < 0)
1702                 goto fail;
1703         BUG_ON(ret);
1704
1705         leaf = path->nodes[0];
1706         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1707         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1708         btrfs_mark_buffer_dirty(leaf);
1709         btrfs_release_path(extent_root, path);
1710 fail:
1711         finish_current_insert(trans, extent_root, 0);
1712         pending_ret = del_pending_extents(trans, extent_root, 0);
1713         if (ret)
1714                 return ret;
1715         if (pending_ret)
1716                 return pending_ret;
1717         return 0;
1718
1719 }
1720
1721 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1722                                    struct btrfs_root *root)
1723 {
1724         struct btrfs_block_group_cache *cache, *entry;
1725         struct rb_node *n;
1726         int err = 0;
1727         int werr = 0;
1728         struct btrfs_path *path;
1729         u64 last = 0;
1730
1731         path = btrfs_alloc_path();
1732         if (!path)
1733                 return -ENOMEM;
1734
1735         while(1) {
1736                 cache = NULL;
1737                 spin_lock(&root->fs_info->block_group_cache_lock);
1738                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1739                      n; n = rb_next(n)) {
1740                         entry = rb_entry(n, struct btrfs_block_group_cache,
1741                                          cache_node);
1742                         if (entry->dirty) {
1743                                 cache = entry;
1744                                 break;
1745                         }
1746                 }
1747                 spin_unlock(&root->fs_info->block_group_cache_lock);
1748
1749                 if (!cache)
1750                         break;
1751
1752                 cache->dirty = 0;
1753                 last += cache->key.offset;
1754
1755                 err = write_one_cache_group(trans, root,
1756                                             path, cache);
1757                 /*
1758                  * if we fail to write the cache group, we want
1759                  * to keep it marked dirty in hopes that a later
1760                  * write will work
1761                  */
1762                 if (err) {
1763                         werr = err;
1764                         continue;
1765                 }
1766         }
1767         btrfs_free_path(path);
1768         return werr;
1769 }
1770
1771 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1772 {
1773         struct btrfs_block_group_cache *block_group;
1774         int readonly = 0;
1775
1776         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1777         if (!block_group || block_group->ro)
1778                 readonly = 1;
1779         if (block_group)
1780                 put_block_group(block_group);
1781         return readonly;
1782 }
1783
1784 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1785                              u64 total_bytes, u64 bytes_used,
1786                              struct btrfs_space_info **space_info)
1787 {
1788         struct btrfs_space_info *found;
1789
1790         found = __find_space_info(info, flags);
1791         if (found) {
1792                 spin_lock(&found->lock);
1793                 found->total_bytes += total_bytes;
1794                 found->bytes_used += bytes_used;
1795                 found->full = 0;
1796                 spin_unlock(&found->lock);
1797                 *space_info = found;
1798                 return 0;
1799         }
1800         found = kzalloc(sizeof(*found), GFP_NOFS);
1801         if (!found)
1802                 return -ENOMEM;
1803
1804         list_add(&found->list, &info->space_info);
1805         INIT_LIST_HEAD(&found->block_groups);
1806         init_rwsem(&found->groups_sem);
1807         spin_lock_init(&found->lock);
1808         found->flags = flags;
1809         found->total_bytes = total_bytes;
1810         found->bytes_used = bytes_used;
1811         found->bytes_pinned = 0;
1812         found->bytes_reserved = 0;
1813         found->bytes_readonly = 0;
1814         found->full = 0;
1815         found->force_alloc = 0;
1816         *space_info = found;
1817         return 0;
1818 }
1819
1820 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1821 {
1822         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1823                                    BTRFS_BLOCK_GROUP_RAID1 |
1824                                    BTRFS_BLOCK_GROUP_RAID10 |
1825                                    BTRFS_BLOCK_GROUP_DUP);
1826         if (extra_flags) {
1827                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1828                         fs_info->avail_data_alloc_bits |= extra_flags;
1829                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1830                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1831                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1832                         fs_info->avail_system_alloc_bits |= extra_flags;
1833         }
1834 }
1835
1836 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1837 {
1838         spin_lock(&cache->space_info->lock);
1839         spin_lock(&cache->lock);
1840         if (!cache->ro) {
1841                 cache->space_info->bytes_readonly += cache->key.offset -
1842                                         btrfs_block_group_used(&cache->item);
1843                 cache->ro = 1;
1844         }
1845         spin_unlock(&cache->lock);
1846         spin_unlock(&cache->space_info->lock);
1847 }
1848
1849 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1850 {
1851         u64 num_devices = root->fs_info->fs_devices->rw_devices;
1852
1853         if (num_devices == 1)
1854                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1855         if (num_devices < 4)
1856                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1857
1858         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1859             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1860                       BTRFS_BLOCK_GROUP_RAID10))) {
1861                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1862         }
1863
1864         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1865             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1866                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1867         }
1868
1869         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1870             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1871              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1872              (flags & BTRFS_BLOCK_GROUP_DUP)))
1873                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1874         return flags;
1875 }
1876
1877 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1878                           struct btrfs_root *extent_root, u64 alloc_bytes,
1879                           u64 flags, int force)
1880 {
1881         struct btrfs_space_info *space_info;
1882         u64 thresh;
1883         int ret = 0;
1884
1885         mutex_lock(&extent_root->fs_info->chunk_mutex);
1886
1887         flags = btrfs_reduce_alloc_profile(extent_root, flags);
1888
1889         space_info = __find_space_info(extent_root->fs_info, flags);
1890         if (!space_info) {
1891                 ret = update_space_info(extent_root->fs_info, flags,
1892                                         0, 0, &space_info);
1893                 BUG_ON(ret);
1894         }
1895         BUG_ON(!space_info);
1896
1897         spin_lock(&space_info->lock);
1898         if (space_info->force_alloc) {
1899                 force = 1;
1900                 space_info->force_alloc = 0;
1901         }
1902         if (space_info->full) {
1903                 spin_unlock(&space_info->lock);
1904                 goto out;
1905         }
1906
1907         thresh = space_info->total_bytes - space_info->bytes_readonly;
1908         thresh = div_factor(thresh, 6);
1909         if (!force &&
1910            (space_info->bytes_used + space_info->bytes_pinned +
1911             space_info->bytes_reserved + alloc_bytes) < thresh) {
1912                 spin_unlock(&space_info->lock);
1913                 goto out;
1914         }
1915         spin_unlock(&space_info->lock);
1916
1917         ret = btrfs_alloc_chunk(trans, extent_root, flags);
1918         if (ret) {
1919 printk("space info full %Lu\n", flags);
1920                 space_info->full = 1;
1921         }
1922 out:
1923         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1924         return ret;
1925 }
1926
1927 static int update_block_group(struct btrfs_trans_handle *trans,
1928                               struct btrfs_root *root,
1929                               u64 bytenr, u64 num_bytes, int alloc,
1930                               int mark_free)
1931 {
1932         struct btrfs_block_group_cache *cache;
1933         struct btrfs_fs_info *info = root->fs_info;
1934         u64 total = num_bytes;
1935         u64 old_val;
1936         u64 byte_in_group;
1937
1938         while(total) {
1939                 cache = btrfs_lookup_block_group(info, bytenr);
1940                 if (!cache)
1941                         return -1;
1942                 byte_in_group = bytenr - cache->key.objectid;
1943                 WARN_ON(byte_in_group > cache->key.offset);
1944
1945                 spin_lock(&cache->space_info->lock);
1946                 spin_lock(&cache->lock);
1947                 cache->dirty = 1;
1948                 old_val = btrfs_block_group_used(&cache->item);
1949                 num_bytes = min(total, cache->key.offset - byte_in_group);
1950                 if (alloc) {
1951                         old_val += num_bytes;
1952                         cache->space_info->bytes_used += num_bytes;
1953                         if (cache->ro)
1954                                 cache->space_info->bytes_readonly -= num_bytes;
1955                         btrfs_set_block_group_used(&cache->item, old_val);
1956                         spin_unlock(&cache->lock);
1957                         spin_unlock(&cache->space_info->lock);
1958                 } else {
1959                         old_val -= num_bytes;
1960                         cache->space_info->bytes_used -= num_bytes;
1961                         if (cache->ro)
1962                                 cache->space_info->bytes_readonly += num_bytes;
1963                         btrfs_set_block_group_used(&cache->item, old_val);
1964                         spin_unlock(&cache->lock);
1965                         spin_unlock(&cache->space_info->lock);
1966                         if (mark_free) {
1967                                 int ret;
1968                                 ret = btrfs_add_free_space(cache, bytenr,
1969                                                            num_bytes);
1970                                 WARN_ON(ret);
1971                         }
1972                 }
1973                 put_block_group(cache);
1974                 total -= num_bytes;
1975                 bytenr += num_bytes;
1976         }
1977         return 0;
1978 }
1979
1980 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1981 {
1982         struct btrfs_block_group_cache *cache;
1983         u64 bytenr;
1984
1985         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1986         if (!cache)
1987                 return 0;
1988
1989         bytenr = cache->key.objectid;
1990         put_block_group(cache);
1991
1992         return bytenr;
1993 }
1994
1995 int btrfs_update_pinned_extents(struct btrfs_root *root,
1996                                 u64 bytenr, u64 num, int pin)
1997 {
1998         u64 len;
1999         struct btrfs_block_group_cache *cache;
2000         struct btrfs_fs_info *fs_info = root->fs_info;
2001
2002         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2003         if (pin) {
2004                 set_extent_dirty(&fs_info->pinned_extents,
2005                                 bytenr, bytenr + num - 1, GFP_NOFS);
2006         } else {
2007                 clear_extent_dirty(&fs_info->pinned_extents,
2008                                 bytenr, bytenr + num - 1, GFP_NOFS);
2009         }
2010         while (num > 0) {
2011                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2012                 BUG_ON(!cache);
2013                 len = min(num, cache->key.offset -
2014                           (bytenr - cache->key.objectid));
2015                 if (pin) {
2016                         spin_lock(&cache->space_info->lock);
2017                         spin_lock(&cache->lock);
2018                         cache->pinned += len;
2019                         cache->space_info->bytes_pinned += len;
2020                         spin_unlock(&cache->lock);
2021                         spin_unlock(&cache->space_info->lock);
2022                         fs_info->total_pinned += len;
2023                 } else {
2024                         spin_lock(&cache->space_info->lock);
2025                         spin_lock(&cache->lock);
2026                         cache->pinned -= len;
2027                         cache->space_info->bytes_pinned -= len;
2028                         spin_unlock(&cache->lock);
2029                         spin_unlock(&cache->space_info->lock);
2030                         fs_info->total_pinned -= len;
2031                         if (cache->cached)
2032                                 btrfs_add_free_space(cache, bytenr, len);
2033                 }
2034                 put_block_group(cache);
2035                 bytenr += len;
2036                 num -= len;
2037         }
2038         return 0;
2039 }
2040
2041 static int update_reserved_extents(struct btrfs_root *root,
2042                                    u64 bytenr, u64 num, int reserve)
2043 {
2044         u64 len;
2045         struct btrfs_block_group_cache *cache;
2046         struct btrfs_fs_info *fs_info = root->fs_info;
2047
2048         while (num > 0) {
2049                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2050                 BUG_ON(!cache);
2051                 len = min(num, cache->key.offset -
2052                           (bytenr - cache->key.objectid));
2053
2054                 spin_lock(&cache->space_info->lock);
2055                 spin_lock(&cache->lock);
2056                 if (reserve) {
2057                         cache->reserved += len;
2058                         cache->space_info->bytes_reserved += len;
2059                 } else {
2060                         cache->reserved -= len;
2061                         cache->space_info->bytes_reserved -= len;
2062                 }
2063                 spin_unlock(&cache->lock);
2064                 spin_unlock(&cache->space_info->lock);
2065                 put_block_group(cache);
2066                 bytenr += len;
2067                 num -= len;
2068         }
2069         return 0;
2070 }
2071
2072 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2073 {
2074         u64 last = 0;
2075         u64 start;
2076         u64 end;
2077         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2078         int ret;
2079
2080         mutex_lock(&root->fs_info->pinned_mutex);
2081         while(1) {
2082                 ret = find_first_extent_bit(pinned_extents, last,
2083                                             &start, &end, EXTENT_DIRTY);
2084                 if (ret)
2085                         break;
2086                 set_extent_dirty(copy, start, end, GFP_NOFS);
2087                 last = end + 1;
2088         }
2089         mutex_unlock(&root->fs_info->pinned_mutex);
2090         return 0;
2091 }
2092
2093 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2094                                struct btrfs_root *root,
2095                                struct extent_io_tree *unpin)
2096 {
2097         u64 start;
2098         u64 end;
2099         int ret;
2100
2101         mutex_lock(&root->fs_info->pinned_mutex);
2102         while(1) {
2103                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2104                                             EXTENT_DIRTY);
2105                 if (ret)
2106                         break;
2107                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2108                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2109                 if (need_resched()) {
2110                         mutex_unlock(&root->fs_info->pinned_mutex);
2111                         cond_resched();
2112                         mutex_lock(&root->fs_info->pinned_mutex);
2113                 }
2114         }
2115         mutex_unlock(&root->fs_info->pinned_mutex);
2116         return 0;
2117 }
2118
2119 static int finish_current_insert(struct btrfs_trans_handle *trans,
2120                                  struct btrfs_root *extent_root, int all)
2121 {
2122         u64 start;
2123         u64 end;
2124         u64 priv;
2125         u64 search = 0;
2126         u64 skipped = 0;
2127         struct btrfs_fs_info *info = extent_root->fs_info;
2128         struct btrfs_path *path;
2129         struct pending_extent_op *extent_op, *tmp;
2130         struct list_head insert_list, update_list;
2131         int ret;
2132         int num_inserts = 0, max_inserts;
2133
2134         path = btrfs_alloc_path();
2135         INIT_LIST_HEAD(&insert_list);
2136         INIT_LIST_HEAD(&update_list);
2137
2138         max_inserts = extent_root->leafsize /
2139                 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2140                  sizeof(struct btrfs_extent_ref) +
2141                  sizeof(struct btrfs_extent_item));
2142 again:
2143         mutex_lock(&info->extent_ins_mutex);
2144         while (1) {
2145                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2146                                             &end, EXTENT_WRITEBACK);
2147                 if (ret) {
2148                         if (skipped && all && !num_inserts) {
2149                                 skipped = 0;
2150                                 search = 0;
2151                                 continue;
2152                         }
2153                         mutex_unlock(&info->extent_ins_mutex);
2154                         break;
2155                 }
2156
2157                 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2158                 if (!ret) {
2159                         skipped = 1;
2160                         search = end + 1;
2161                         if (need_resched()) {
2162                                 mutex_unlock(&info->extent_ins_mutex);
2163                                 cond_resched();
2164                                 mutex_lock(&info->extent_ins_mutex);
2165                         }
2166                         continue;
2167                 }
2168
2169                 ret = get_state_private(&info->extent_ins, start, &priv);
2170                 BUG_ON(ret);
2171                 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2172
2173                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2174                         num_inserts++;
2175                         list_add_tail(&extent_op->list, &insert_list);
2176                         search = end + 1;
2177                         if (num_inserts == max_inserts) {
2178                                 mutex_unlock(&info->extent_ins_mutex);
2179                                 break;
2180                         }
2181                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2182                         list_add_tail(&extent_op->list, &update_list);
2183                         search = end + 1;
2184                 } else {
2185                         BUG();
2186                 }
2187         }
2188
2189         /*
2190          * process the update list, clear the writeback bit for it, and if
2191          * somebody marked this thing for deletion then just unlock it and be
2192          * done, the free_extents will handle it
2193          */
2194         mutex_lock(&info->extent_ins_mutex);
2195         list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2196                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2197                                   extent_op->bytenr + extent_op->num_bytes - 1,
2198                                   EXTENT_WRITEBACK, GFP_NOFS);
2199                 if (extent_op->del) {
2200                         list_del_init(&extent_op->list);
2201                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2202                                       extent_op->bytenr + extent_op->num_bytes
2203                                       - 1, GFP_NOFS);
2204                         kfree(extent_op);
2205                 }
2206         }
2207         mutex_unlock(&info->extent_ins_mutex);
2208
2209         /*
2210          * still have things left on the update list, go ahead an update
2211          * everything
2212          */
2213         if (!list_empty(&update_list)) {
2214                 ret = update_backrefs(trans, extent_root, path, &update_list);
2215                 BUG_ON(ret);
2216         }
2217
2218         /*
2219          * if no inserts need to be done, but we skipped some extents and we
2220          * need to make sure everything is cleaned then reset everything and
2221          * go back to the beginning
2222          */
2223         if (!num_inserts && all && skipped) {
2224                 search = 0;
2225                 skipped = 0;
2226                 INIT_LIST_HEAD(&update_list);
2227                 INIT_LIST_HEAD(&insert_list);
2228                 goto again;
2229         } else if (!num_inserts) {
2230                 goto out;
2231         }
2232
2233         /*
2234          * process the insert extents list.  Again if we are deleting this
2235          * extent, then just unlock it, pin down the bytes if need be, and be
2236          * done with it.  Saves us from having to actually insert the extent
2237          * into the tree and then subsequently come along and delete it
2238          */
2239         mutex_lock(&info->extent_ins_mutex);
2240         list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2241                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2242                                   extent_op->bytenr + extent_op->num_bytes - 1,
2243                                   EXTENT_WRITEBACK, GFP_NOFS);
2244                 if (extent_op->del) {
2245                         u64 used;
2246                         list_del_init(&extent_op->list);
2247                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2248                                       extent_op->bytenr + extent_op->num_bytes
2249                                       - 1, GFP_NOFS);
2250
2251                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2252                         ret = pin_down_bytes(trans, extent_root,
2253                                              extent_op->bytenr,
2254                                              extent_op->num_bytes, 0);
2255                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2256
2257                         spin_lock(&info->delalloc_lock);
2258                         used = btrfs_super_bytes_used(&info->super_copy);
2259                         btrfs_set_super_bytes_used(&info->super_copy,
2260                                         used - extent_op->num_bytes);
2261                         used = btrfs_root_used(&extent_root->root_item);
2262                         btrfs_set_root_used(&extent_root->root_item,
2263                                         used - extent_op->num_bytes);
2264                         spin_unlock(&info->delalloc_lock);
2265
2266                         ret = update_block_group(trans, extent_root,
2267                                                  extent_op->bytenr,
2268                                                  extent_op->num_bytes,
2269                                                  0, ret > 0);
2270                         BUG_ON(ret);
2271                         kfree(extent_op);
2272                         num_inserts--;
2273                 }
2274         }
2275         mutex_unlock(&info->extent_ins_mutex);
2276
2277         ret = insert_extents(trans, extent_root, path, &insert_list,
2278                              num_inserts);
2279         BUG_ON(ret);
2280
2281         /*
2282          * if we broke out of the loop in order to insert stuff because we hit
2283          * the maximum number of inserts at a time we can handle, then loop
2284          * back and pick up where we left off
2285          */
2286         if (num_inserts == max_inserts) {
2287                 INIT_LIST_HEAD(&insert_list);
2288                 INIT_LIST_HEAD(&update_list);
2289                 num_inserts = 0;
2290                 goto again;
2291         }
2292
2293         /*
2294          * again, if we need to make absolutely sure there are no more pending
2295          * extent operations left and we know that we skipped some, go back to
2296          * the beginning and do it all again
2297          */
2298         if (all && skipped) {
2299                 INIT_LIST_HEAD(&insert_list);
2300                 INIT_LIST_HEAD(&update_list);
2301                 search = 0;
2302                 skipped = 0;
2303                 num_inserts = 0;
2304                 goto again;
2305         }
2306 out:
2307         btrfs_free_path(path);
2308         return 0;
2309 }
2310
2311 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2312                           struct btrfs_root *root,
2313                           u64 bytenr, u64 num_bytes, int is_data)
2314 {
2315         int err = 0;
2316         struct extent_buffer *buf;
2317
2318         if (is_data)
2319                 goto pinit;
2320
2321         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2322         if (!buf)
2323                 goto pinit;
2324
2325         /* we can reuse a block if it hasn't been written
2326          * and it is from this transaction.  We can't
2327          * reuse anything from the tree log root because
2328          * it has tiny sub-transactions.
2329          */
2330         if (btrfs_buffer_uptodate(buf, 0) &&
2331             btrfs_try_tree_lock(buf)) {
2332                 u64 header_owner = btrfs_header_owner(buf);
2333                 u64 header_transid = btrfs_header_generation(buf);
2334                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2335                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2336                     header_transid == trans->transid &&
2337                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2338                         clean_tree_block(NULL, root, buf);
2339                         btrfs_tree_unlock(buf);
2340                         free_extent_buffer(buf);
2341                         return 1;
2342                 }
2343                 btrfs_tree_unlock(buf);
2344         }
2345         free_extent_buffer(buf);
2346 pinit:
2347         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2348
2349         BUG_ON(err < 0);
2350         return 0;
2351 }
2352
2353 /*
2354  * remove an extent from the root, returns 0 on success
2355  */
2356 static int __free_extent(struct btrfs_trans_handle *trans,
2357                          struct btrfs_root *root,
2358                          u64 bytenr, u64 num_bytes, u64 parent,
2359                          u64 root_objectid, u64 ref_generation,
2360                          u64 owner_objectid, int pin, int mark_free)
2361 {
2362         struct btrfs_path *path;
2363         struct btrfs_key key;
2364         struct btrfs_fs_info *info = root->fs_info;
2365         struct btrfs_root *extent_root = info->extent_root;
2366         struct extent_buffer *leaf;
2367         int ret;
2368         int extent_slot = 0;
2369         int found_extent = 0;
2370         int num_to_del = 1;
2371         struct btrfs_extent_item *ei;
2372         u32 refs;
2373
2374         key.objectid = bytenr;
2375         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2376         key.offset = num_bytes;
2377         path = btrfs_alloc_path();
2378         if (!path)
2379                 return -ENOMEM;
2380
2381         path->reada = 1;
2382         ret = lookup_extent_backref(trans, extent_root, path,
2383                                     bytenr, parent, root_objectid,
2384                                     ref_generation, owner_objectid, 1);
2385         if (ret == 0) {
2386                 struct btrfs_key found_key;
2387                 extent_slot = path->slots[0];
2388                 while(extent_slot > 0) {
2389                         extent_slot--;
2390                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2391                                               extent_slot);
2392                         if (found_key.objectid != bytenr)
2393                                 break;
2394                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2395                             found_key.offset == num_bytes) {
2396                                 found_extent = 1;
2397                                 break;
2398                         }
2399                         if (path->slots[0] - extent_slot > 5)
2400                                 break;
2401                 }
2402                 if (!found_extent) {
2403                         ret = remove_extent_backref(trans, extent_root, path);
2404                         BUG_ON(ret);
2405                         btrfs_release_path(extent_root, path);
2406                         ret = btrfs_search_slot(trans, extent_root,
2407                                                 &key, path, -1, 1);
2408                         if (ret) {
2409                                 printk(KERN_ERR "umm, got %d back from search"
2410                                        ", was looking for %Lu\n", ret,
2411                                        bytenr);
2412                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2413                         }
2414                         BUG_ON(ret);
2415                         extent_slot = path->slots[0];
2416                 }
2417         } else {
2418                 btrfs_print_leaf(extent_root, path->nodes[0]);
2419                 WARN_ON(1);
2420                 printk("Unable to find ref byte nr %Lu root %Lu "
2421                        "gen %Lu owner %Lu\n", bytenr,
2422                        root_objectid, ref_generation, owner_objectid);
2423         }
2424
2425         leaf = path->nodes[0];
2426         ei = btrfs_item_ptr(leaf, extent_slot,
2427                             struct btrfs_extent_item);
2428         refs = btrfs_extent_refs(leaf, ei);
2429         BUG_ON(refs == 0);
2430         refs -= 1;
2431         btrfs_set_extent_refs(leaf, ei, refs);
2432
2433         btrfs_mark_buffer_dirty(leaf);
2434
2435         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2436                 struct btrfs_extent_ref *ref;
2437                 ref = btrfs_item_ptr(leaf, path->slots[0],
2438                                      struct btrfs_extent_ref);
2439                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2440                 /* if the back ref and the extent are next to each other
2441                  * they get deleted below in one shot
2442                  */
2443                 path->slots[0] = extent_slot;
2444                 num_to_del = 2;
2445         } else if (found_extent) {
2446                 /* otherwise delete the extent back ref */
2447                 ret = remove_extent_backref(trans, extent_root, path);
2448                 BUG_ON(ret);
2449                 /* if refs are 0, we need to setup the path for deletion */
2450                 if (refs == 0) {
2451                         btrfs_release_path(extent_root, path);
2452                         ret = btrfs_search_slot(trans, extent_root, &key, path,
2453                                                 -1, 1);
2454                         BUG_ON(ret);
2455                 }
2456         }
2457
2458         if (refs == 0) {
2459                 u64 super_used;
2460                 u64 root_used;
2461 #ifdef BIO_RW_DISCARD
2462                 u64 map_length = num_bytes;
2463                 struct btrfs_multi_bio *multi = NULL;
2464 #endif
2465
2466                 if (pin) {
2467                         mutex_lock(&root->fs_info->pinned_mutex);
2468                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2469                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2470                         mutex_unlock(&root->fs_info->pinned_mutex);
2471                         if (ret > 0)
2472                                 mark_free = 1;
2473                         BUG_ON(ret < 0);
2474                 }
2475                 /* block accounting for super block */
2476                 spin_lock(&info->delalloc_lock);
2477                 super_used = btrfs_super_bytes_used(&info->super_copy);
2478                 btrfs_set_super_bytes_used(&info->super_copy,
2479                                            super_used - num_bytes);
2480
2481                 /* block accounting for root item */
2482                 root_used = btrfs_root_used(&root->root_item);
2483                 btrfs_set_root_used(&root->root_item,
2484                                            root_used - num_bytes);
2485                 spin_unlock(&info->delalloc_lock);
2486                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2487                                       num_to_del);
2488                 BUG_ON(ret);
2489                 btrfs_release_path(extent_root, path);
2490
2491                 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2492                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2493                         BUG_ON(ret);
2494                 }
2495
2496                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2497                                          mark_free);
2498                 BUG_ON(ret);
2499 #ifdef BIO_RW_DISCARD
2500                 /* Tell the block device(s) that the sectors can be discarded */
2501                 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2502                                       bytenr, &map_length, &multi, 0);
2503                 if (!ret) {
2504                         struct btrfs_bio_stripe *stripe = multi->stripes;
2505                         int i;
2506
2507                         if (map_length > num_bytes)
2508                                 map_length = num_bytes;
2509
2510                         for (i = 0; i < multi->num_stripes; i++, stripe++) {
2511                                 btrfs_issue_discard(stripe->dev->bdev,
2512                                                     stripe->physical,
2513                                                      map_length);
2514                         }
2515                         kfree(multi);
2516                 }
2517 #endif
2518         }
2519         btrfs_free_path(path);
2520         finish_current_insert(trans, extent_root, 0);
2521         return ret;
2522 }
2523
2524 /*
2525  * find all the blocks marked as pending in the radix tree and remove
2526  * them from the extent map
2527  */
2528 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2529                                btrfs_root *extent_root, int all)
2530 {
2531         int ret;
2532         int err = 0;
2533         u64 start;
2534         u64 end;
2535         u64 priv;
2536         u64 search = 0;
2537         int nr = 0, skipped = 0;
2538         struct extent_io_tree *pending_del;
2539         struct extent_io_tree *extent_ins;
2540         struct pending_extent_op *extent_op;
2541         struct btrfs_fs_info *info = extent_root->fs_info;
2542         struct list_head delete_list;
2543
2544         INIT_LIST_HEAD(&delete_list);
2545         extent_ins = &extent_root->fs_info->extent_ins;
2546         pending_del = &extent_root->fs_info->pending_del;
2547
2548 again:
2549         mutex_lock(&info->extent_ins_mutex);
2550         while(1) {
2551                 ret = find_first_extent_bit(pending_del, search, &start, &end,
2552                                             EXTENT_WRITEBACK);
2553                 if (ret) {
2554                         if (all && skipped && !nr) {
2555                                 search = 0;
2556                                 continue;
2557                         }
2558                         mutex_unlock(&info->extent_ins_mutex);
2559                         break;
2560                 }
2561
2562                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2563                 if (!ret) {
2564                         search = end+1;
2565                         skipped = 1;
2566
2567                         if (need_resched()) {
2568                                 mutex_unlock(&info->extent_ins_mutex);
2569                                 cond_resched();
2570                                 mutex_lock(&info->extent_ins_mutex);
2571                         }
2572
2573                         continue;
2574                 }
2575                 BUG_ON(ret < 0);
2576
2577                 ret = get_state_private(pending_del, start, &priv);
2578                 BUG_ON(ret);
2579                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2580
2581                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2582                                   GFP_NOFS);
2583                 if (!test_range_bit(extent_ins, start, end,
2584                                     EXTENT_WRITEBACK, 0)) {
2585                         list_add_tail(&extent_op->list, &delete_list);
2586                         nr++;
2587                 } else {
2588                         kfree(extent_op);
2589
2590                         ret = get_state_private(&info->extent_ins, start,
2591                                                 &priv);
2592                         BUG_ON(ret);
2593                         extent_op = (struct pending_extent_op *)
2594                                                 (unsigned long)priv;
2595
2596                         clear_extent_bits(&info->extent_ins, start, end,
2597                                           EXTENT_WRITEBACK, GFP_NOFS);
2598
2599                         if (extent_op->type == PENDING_BACKREF_UPDATE) {
2600                                 list_add_tail(&extent_op->list, &delete_list);
2601                                 search = end + 1;
2602                                 nr++;
2603                                 continue;
2604                         }
2605
2606                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2607                         ret = pin_down_bytes(trans, extent_root, start,
2608                                              end + 1 - start, 0);
2609                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2610
2611                         ret = update_block_group(trans, extent_root, start,
2612                                                 end + 1 - start, 0, ret > 0);
2613
2614                         unlock_extent(extent_ins, start, end, GFP_NOFS);
2615                         BUG_ON(ret);
2616                         kfree(extent_op);
2617                 }
2618                 if (ret)
2619                         err = ret;
2620
2621                 search = end + 1;
2622
2623                 if (need_resched()) {
2624                         mutex_unlock(&info->extent_ins_mutex);
2625                         cond_resched();
2626                         mutex_lock(&info->extent_ins_mutex);
2627                 }
2628         }
2629
2630         if (nr) {
2631                 ret = free_extents(trans, extent_root, &delete_list);
2632                 BUG_ON(ret);
2633         }
2634
2635         if (all && skipped) {
2636                 INIT_LIST_HEAD(&delete_list);
2637                 search = 0;
2638                 nr = 0;
2639                 goto again;
2640         }
2641
2642         return err;
2643 }
2644
2645 /*
2646  * remove an extent from the root, returns 0 on success
2647  */
2648 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2649                                struct btrfs_root *root,
2650                                u64 bytenr, u64 num_bytes, u64 parent,
2651                                u64 root_objectid, u64 ref_generation,
2652                                u64 owner_objectid, int pin)
2653 {
2654         struct btrfs_root *extent_root = root->fs_info->extent_root;
2655         int pending_ret;
2656         int ret;
2657
2658         WARN_ON(num_bytes < root->sectorsize);
2659         if (root == extent_root) {
2660                 struct pending_extent_op *extent_op = NULL;
2661
2662                 mutex_lock(&root->fs_info->extent_ins_mutex);
2663                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2664                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2665                         u64 priv;
2666                         ret = get_state_private(&root->fs_info->extent_ins,
2667                                                 bytenr, &priv);
2668                         BUG_ON(ret);
2669                         extent_op = (struct pending_extent_op *)
2670                                                 (unsigned long)priv;
2671
2672                         extent_op->del = 1;
2673                         if (extent_op->type == PENDING_EXTENT_INSERT) {
2674                                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2675                                 return 0;
2676                         }
2677                 }
2678
2679                 if (extent_op) {
2680                         ref_generation = extent_op->orig_generation;
2681                         parent = extent_op->orig_parent;
2682                 }
2683
2684                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2685                 BUG_ON(!extent_op);
2686
2687                 extent_op->type = PENDING_EXTENT_DELETE;
2688                 extent_op->bytenr = bytenr;
2689                 extent_op->num_bytes = num_bytes;
2690                 extent_op->parent = parent;
2691                 extent_op->orig_parent = parent;
2692                 extent_op->generation = ref_generation;
2693                 extent_op->orig_generation = ref_generation;
2694                 extent_op->level = (int)owner_objectid;
2695                 INIT_LIST_HEAD(&extent_op->list);
2696                 extent_op->del = 0;
2697
2698                 set_extent_bits(&root->fs_info->pending_del,
2699                                 bytenr, bytenr + num_bytes - 1,
2700                                 EXTENT_WRITEBACK, GFP_NOFS);
2701                 set_state_private(&root->fs_info->pending_del,
2702                                   bytenr, (unsigned long)extent_op);
2703                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2704                 return 0;
2705         }
2706         /* if metadata always pin */
2707         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2708                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2709                         struct btrfs_block_group_cache *cache;
2710
2711                         /* btrfs_free_reserved_extent */
2712                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2713                         BUG_ON(!cache);
2714                         btrfs_add_free_space(cache, bytenr, num_bytes);
2715                         put_block_group(cache);
2716                         update_reserved_extents(root, bytenr, num_bytes, 0);
2717                         return 0;
2718                 }
2719                 pin = 1;
2720         }
2721
2722         /* if data pin when any transaction has committed this */
2723         if (ref_generation != trans->transid)
2724                 pin = 1;
2725
2726         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2727                             root_objectid, ref_generation,
2728                             owner_objectid, pin, pin == 0);
2729
2730         finish_current_insert(trans, root->fs_info->extent_root, 0);
2731         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2732         return ret ? ret : pending_ret;
2733 }
2734
2735 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2736                       struct btrfs_root *root,
2737                       u64 bytenr, u64 num_bytes, u64 parent,
2738                       u64 root_objectid, u64 ref_generation,
2739                       u64 owner_objectid, int pin)
2740 {
2741         int ret;
2742
2743         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2744                                   root_objectid, ref_generation,
2745                                   owner_objectid, pin);
2746         return ret;
2747 }
2748
2749 static u64 stripe_align(struct btrfs_root *root, u64 val)
2750 {
2751         u64 mask = ((u64)root->stripesize - 1);
2752         u64 ret = (val + mask) & ~mask;
2753         return ret;
2754 }
2755
2756 /*
2757  * walks the btree of allocated extents and find a hole of a given size.
2758  * The key ins is changed to record the hole:
2759  * ins->objectid == block start
2760  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2761  * ins->offset == number of blocks
2762  * Any available blocks before search_start are skipped.
2763  */
2764 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2765                                      struct btrfs_root *orig_root,
2766                                      u64 num_bytes, u64 empty_size,
2767                                      u64 search_start, u64 search_end,
2768                                      u64 hint_byte, struct btrfs_key *ins,
2769                                      u64 exclude_start, u64 exclude_nr,
2770                                      int data)
2771 {
2772         int ret = 0;
2773         struct btrfs_root * root = orig_root->fs_info->extent_root;
2774         u64 total_needed = num_bytes;
2775         u64 *last_ptr = NULL;
2776         u64 last_wanted = 0;
2777         struct btrfs_block_group_cache *block_group = NULL;
2778         int chunk_alloc_done = 0;
2779         int empty_cluster = 2 * 1024 * 1024;
2780         int allowed_chunk_alloc = 0;
2781         struct list_head *head = NULL, *cur = NULL;
2782         int loop = 0;
2783         int extra_loop = 0;
2784         struct btrfs_space_info *space_info;
2785
2786         WARN_ON(num_bytes < root->sectorsize);
2787         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2788         ins->objectid = 0;
2789         ins->offset = 0;
2790
2791         if (orig_root->ref_cows || empty_size)
2792                 allowed_chunk_alloc = 1;
2793
2794         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2795                 last_ptr = &root->fs_info->last_alloc;
2796                 empty_cluster = 64 * 1024;
2797         }
2798
2799         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2800                 last_ptr = &root->fs_info->last_data_alloc;
2801
2802         if (last_ptr) {
2803                 if (*last_ptr) {
2804                         hint_byte = *last_ptr;
2805                         last_wanted = *last_ptr;
2806                 } else
2807                         empty_size += empty_cluster;
2808         } else {
2809                 empty_cluster = 0;
2810         }
2811         search_start = max(search_start, first_logical_byte(root, 0));
2812         search_start = max(search_start, hint_byte);
2813
2814         if (last_wanted && search_start != last_wanted) {
2815                 last_wanted = 0;
2816                 empty_size += empty_cluster;
2817         }
2818
2819         total_needed += empty_size;
2820         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2821         if (!block_group)
2822                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2823                                                              search_start);
2824         space_info = __find_space_info(root->fs_info, data);
2825
2826         down_read(&space_info->groups_sem);
2827         while (1) {
2828                 struct btrfs_free_space *free_space;
2829                 /*
2830                  * the only way this happens if our hint points to a block
2831                  * group thats not of the proper type, while looping this
2832                  * should never happen
2833                  */
2834                 if (empty_size)
2835                         extra_loop = 1;
2836
2837                 if (!block_group)
2838                         goto new_group_no_lock;
2839
2840                 if (unlikely(!block_group->cached)) {
2841                         mutex_lock(&block_group->cache_mutex);
2842                         ret = cache_block_group(root, block_group);
2843                         mutex_unlock(&block_group->cache_mutex);
2844                         if (ret)
2845                                 break;
2846                 }
2847
2848                 mutex_lock(&block_group->alloc_mutex);
2849                 if (unlikely(!block_group_bits(block_group, data)))
2850                         goto new_group;
2851
2852                 if (unlikely(block_group->ro))
2853                         goto new_group;
2854
2855                 free_space = btrfs_find_free_space(block_group, search_start,
2856                                                    total_needed);
2857                 if (free_space) {
2858                         u64 start = block_group->key.objectid;
2859                         u64 end = block_group->key.objectid +
2860                                 block_group->key.offset;
2861
2862                         search_start = stripe_align(root, free_space->offset);
2863
2864                         /* move on to the next group */
2865                         if (search_start + num_bytes >= search_end)
2866                                 goto new_group;
2867
2868                         /* move on to the next group */
2869                         if (search_start + num_bytes > end)
2870                                 goto new_group;
2871
2872                         if (last_wanted && search_start != last_wanted) {
2873                                 total_needed += empty_cluster;
2874                                 empty_size += empty_cluster;
2875                                 last_wanted = 0;
2876                                 /*
2877                                  * if search_start is still in this block group
2878                                  * then we just re-search this block group
2879                                  */
2880                                 if (search_start >= start &&
2881                                     search_start < end) {
2882                                         mutex_unlock(&block_group->alloc_mutex);
2883                                         continue;
2884                                 }
2885
2886                                 /* else we go to the next block group */
2887                                 goto new_group;
2888                         }
2889
2890                         if (exclude_nr > 0 &&
2891                             (search_start + num_bytes > exclude_start &&
2892                              search_start < exclude_start + exclude_nr)) {
2893                                 search_start = exclude_start + exclude_nr;
2894                                 /*
2895                                  * if search_start is still in this block group
2896                                  * then we just re-search this block group
2897                                  */
2898                                 if (search_start >= start &&
2899                                     search_start < end) {
2900                                         mutex_unlock(&block_group->alloc_mutex);
2901                                         last_wanted = 0;
2902                                         continue;
2903                                 }
2904
2905                                 /* else we go to the next block group */
2906                                 goto new_group;
2907                         }
2908
2909                         ins->objectid = search_start;
2910                         ins->offset = num_bytes;
2911
2912                         btrfs_remove_free_space_lock(block_group, search_start,
2913                                                      num_bytes);
2914                         /* we are all good, lets return */
2915                         mutex_unlock(&block_group->alloc_mutex);
2916                         break;
2917                 }
2918 new_group:
2919                 mutex_unlock(&block_group->alloc_mutex);
2920                 put_block_group(block_group);
2921                 block_group = NULL;
2922 new_group_no_lock:
2923                 /* don't try to compare new allocations against the
2924                  * last allocation any more
2925                  */
2926                 last_wanted = 0;
2927
2928                 /*
2929                  * Here's how this works.
2930                  * loop == 0: we were searching a block group via a hint
2931                  *              and didn't find anything, so we start at
2932                  *              the head of the block groups and keep searching
2933                  * loop == 1: we're searching through all of the block groups
2934                  *              if we hit the head again we have searched
2935                  *              all of the block groups for this space and we
2936                  *              need to try and allocate, if we cant error out.
2937                  * loop == 2: we allocated more space and are looping through
2938                  *              all of the block groups again.
2939                  */
2940                 if (loop == 0) {
2941                         head = &space_info->block_groups;
2942                         cur = head->next;
2943                         loop++;
2944                 } else if (loop == 1 && cur == head) {
2945                         int keep_going;
2946
2947                         /* at this point we give up on the empty_size
2948                          * allocations and just try to allocate the min
2949                          * space.
2950                          *
2951                          * The extra_loop field was set if an empty_size
2952                          * allocation was attempted above, and if this
2953                          * is try we need to try the loop again without
2954                          * the additional empty_size.
2955                          */
2956                         total_needed -= empty_size;
2957                         empty_size = 0;
2958                         keep_going = extra_loop;
2959                         loop++;
2960
2961                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2962                                 up_read(&space_info->groups_sem);
2963                                 ret = do_chunk_alloc(trans, root, num_bytes +
2964                                                      2 * 1024 * 1024, data, 1);
2965                                 down_read(&space_info->groups_sem);
2966                                 if (ret < 0)
2967                                         goto loop_check;
2968                                 head = &space_info->block_groups;
2969                                 /*
2970                                  * we've allocated a new chunk, keep
2971                                  * trying
2972                                  */
2973                                 keep_going = 1;
2974                                 chunk_alloc_done = 1;
2975                         } else if (!allowed_chunk_alloc) {
2976                                 space_info->force_alloc = 1;
2977                         }
2978 loop_check:
2979                         if (keep_going) {
2980                                 cur = head->next;
2981                                 extra_loop = 0;
2982                         } else {
2983                                 break;
2984                         }
2985                 } else if (cur == head) {
2986                         break;
2987                 }
2988
2989                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2990                                          list);
2991                 atomic_inc(&block_group->count);
2992
2993                 search_start = block_group->key.objectid;
2994                 cur = cur->next;
2995         }
2996
2997         /* we found what we needed */
2998         if (ins->objectid) {
2999                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3000                         trans->block_group = block_group->key.objectid;
3001
3002                 if (last_ptr)
3003                         *last_ptr = ins->objectid + ins->offset;
3004                 ret = 0;
3005         } else if (!ret) {
3006                 printk(KERN_ERR "we were searching for %Lu bytes, num_bytes %Lu,"
3007                        " loop %d, allowed_alloc %d\n", total_needed, num_bytes,
3008                        loop, allowed_chunk_alloc);
3009                 ret = -ENOSPC;
3010         }
3011         if (block_group)
3012                 put_block_group(block_group);
3013
3014         up_read(&space_info->groups_sem);
3015         return ret;
3016 }
3017
3018 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3019 {
3020         struct btrfs_block_group_cache *cache;
3021         struct list_head *l;
3022
3023         printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
3024                info->total_bytes - info->bytes_used - info->bytes_pinned -
3025                info->bytes_reserved, (info->full) ? "" : "not ");
3026
3027         down_read(&info->groups_sem);
3028         list_for_each(l, &info->block_groups) {
3029                 cache = list_entry(l, struct btrfs_block_group_cache, list);
3030                 spin_lock(&cache->lock);
3031                 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
3032                        "%Lu pinned %Lu reserved\n",
3033                        cache->key.objectid, cache->key.offset,
3034                        btrfs_block_group_used(&cache->item),
3035                        cache->pinned, cache->reserved);
3036                 btrfs_dump_free_space(cache, bytes);
3037                 spin_unlock(&cache->lock);
3038         }
3039         up_read(&info->groups_sem);
3040 }
3041
3042 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3043                                   struct btrfs_root *root,
3044                                   u64 num_bytes, u64 min_alloc_size,
3045                                   u64 empty_size, u64 hint_byte,
3046                                   u64 search_end, struct btrfs_key *ins,
3047                                   u64 data)
3048 {
3049         int ret;
3050         u64 search_start = 0;
3051         u64 alloc_profile;
3052         struct btrfs_fs_info *info = root->fs_info;
3053
3054         if (data) {
3055                 alloc_profile = info->avail_data_alloc_bits &
3056                                 info->data_alloc_profile;
3057                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3058         } else if (root == root->fs_info->chunk_root) {
3059                 alloc_profile = info->avail_system_alloc_bits &
3060                                 info->system_alloc_profile;
3061                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3062         } else {
3063                 alloc_profile = info->avail_metadata_alloc_bits &
3064                                 info->metadata_alloc_profile;
3065                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3066         }
3067 again:
3068         data = btrfs_reduce_alloc_profile(root, data);
3069         /*
3070          * the only place that sets empty_size is btrfs_realloc_node, which
3071          * is not called recursively on allocations
3072          */
3073         if (empty_size || root->ref_cows) {
3074                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3075                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3076                                      2 * 1024 * 1024,
3077                                      BTRFS_BLOCK_GROUP_METADATA |
3078                                      (info->metadata_alloc_profile &
3079                                       info->avail_metadata_alloc_bits), 0);
3080                 }
3081                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3082                                      num_bytes + 2 * 1024 * 1024, data, 0);
3083         }
3084
3085         WARN_ON(num_bytes < root->sectorsize);
3086         ret = find_free_extent(trans, root, num_bytes, empty_size,
3087                                search_start, search_end, hint_byte, ins,
3088                                trans->alloc_exclude_start,
3089                                trans->alloc_exclude_nr, data);
3090
3091         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3092                 num_bytes = num_bytes >> 1;
3093                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3094                 num_bytes = max(num_bytes, min_alloc_size);
3095                 do_chunk_alloc(trans, root->fs_info->extent_root,
3096                                num_bytes, data, 1);
3097                 goto again;
3098         }
3099         if (ret) {
3100                 struct btrfs_space_info *sinfo;
3101
3102                 sinfo = __find_space_info(root->fs_info, data);
3103                 printk("allocation failed flags %Lu, wanted %Lu\n",
3104                        data, num_bytes);
3105                 dump_space_info(sinfo, num_bytes);
3106                 BUG();
3107         }
3108
3109         return ret;
3110 }
3111
3112 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3113 {
3114         struct btrfs_block_group_cache *cache;
3115
3116         cache = btrfs_lookup_block_group(root->fs_info, start);
3117         if (!cache) {
3118                 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
3119                 return -ENOSPC;
3120         }
3121         btrfs_add_free_space(cache, start, len);
3122         put_block_group(cache);
3123         update_reserved_extents(root, start, len, 0);
3124         return 0;
3125 }
3126
3127 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3128                                   struct btrfs_root *root,
3129                                   u64 num_bytes, u64 min_alloc_size,
3130                                   u64 empty_size, u64 hint_byte,
3131                                   u64 search_end, struct btrfs_key *ins,
3132                                   u64 data)
3133 {
3134         int ret;
3135         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3136                                      empty_size, hint_byte, search_end, ins,
3137                                      data);
3138         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3139         return ret;
3140 }
3141
3142 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3143                                          struct btrfs_root *root, u64 parent,
3144                                          u64 root_objectid, u64 ref_generation,
3145                                          u64 owner, struct btrfs_key *ins)
3146 {
3147         int ret;
3148         int pending_ret;
3149         u64 super_used;
3150         u64 root_used;
3151         u64 num_bytes = ins->offset;
3152         u32 sizes[2];
3153         struct btrfs_fs_info *info = root->fs_info;
3154         struct btrfs_root *extent_root = info->extent_root;
3155         struct btrfs_extent_item *extent_item;
3156         struct btrfs_extent_ref *ref;
3157         struct btrfs_path *path;
3158         struct btrfs_key keys[2];
3159
3160         if (parent == 0)
3161                 parent = ins->objectid;
3162
3163         /* block accounting for super block */
3164         spin_lock(&info->delalloc_lock);
3165         super_used = btrfs_super_bytes_used(&info->super_copy);
3166         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3167
3168         /* block accounting for root item */
3169         root_used = btrfs_root_used(&root->root_item);
3170         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3171         spin_unlock(&info->delalloc_lock);
3172
3173         if (root == extent_root) {
3174                 struct pending_extent_op *extent_op;
3175
3176                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3177                 BUG_ON(!extent_op);
3178
3179                 extent_op->type = PENDING_EXTENT_INSERT;
3180                 extent_op->bytenr = ins->objectid;
3181                 extent_op->num_bytes = ins->offset;
3182                 extent_op->parent = parent;
3183                 extent_op->orig_parent = 0;
3184                 extent_op->generation = ref_generation;
3185                 extent_op->orig_generation = 0;
3186                 extent_op->level = (int)owner;
3187                 INIT_LIST_HEAD(&extent_op->list);
3188                 extent_op->del = 0;
3189
3190                 mutex_lock(&root->fs_info->extent_ins_mutex);
3191                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3192                                 ins->objectid + ins->offset - 1,
3193                                 EXTENT_WRITEBACK, GFP_NOFS);
3194                 set_state_private(&root->fs_info->extent_ins,
3195                                   ins->objectid, (unsigned long)extent_op);
3196                 mutex_unlock(&root->fs_info->extent_ins_mutex);
3197                 goto update_block;
3198         }
3199
3200         memcpy(&keys[0], ins, sizeof(*ins));
3201         keys[1].objectid = ins->objectid;
3202         keys[1].type = BTRFS_EXTENT_REF_KEY;
3203         keys[1].offset = parent;
3204         sizes[0] = sizeof(*extent_item);
3205         sizes[1] = sizeof(*ref);
3206
3207         path = btrfs_alloc_path();
3208         BUG_ON(!path);
3209
3210         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3211                                        sizes, 2);
3212         BUG_ON(ret);
3213
3214         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3215                                      struct btrfs_extent_item);
3216         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3217         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3218                              struct btrfs_extent_ref);
3219
3220         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3221         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3222         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3223         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3224
3225         btrfs_mark_buffer_dirty(path->nodes[0]);
3226
3227         trans->alloc_exclude_start = 0;
3228         trans->alloc_exclude_nr = 0;
3229         btrfs_free_path(path);
3230         finish_current_insert(trans, extent_root, 0);
3231         pending_ret = del_pending_extents(trans, extent_root, 0);
3232
3233         if (ret)
3234                 goto out;
3235         if (pending_ret) {
3236                 ret = pending_ret;
3237                 goto out;
3238         }
3239
3240 update_block:
3241         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
3242         if (ret) {
3243                 printk("update block group failed for %Lu %Lu\n",
3244                        ins->objectid, ins->offset);
3245                 BUG();
3246         }
3247 out:
3248         return ret;
3249 }
3250
3251 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3252                                 struct btrfs_root *root, u64 parent,
3253                                 u64 root_objectid, u64 ref_generation,
3254                                 u64 owner, struct btrfs_key *ins)
3255 {
3256         int ret;
3257
3258         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3259                 return 0;
3260         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3261                                             ref_generation, owner, ins);
3262         update_reserved_extents(root, ins->objectid, ins->offset, 0);
3263         return ret;
3264 }
3265
3266 /*
3267  * this is used by the tree logging recovery code.  It records that
3268  * an extent has been allocated and makes sure to clear the free
3269  * space cache bits as well
3270  */
3271 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3272                                 struct btrfs_root *root, u64 parent,
3273                                 u64 root_objectid, u64 ref_generation,
3274                                 u64 owner, struct btrfs_key *ins)
3275 {
3276         int ret;
3277         struct btrfs_block_group_cache *block_group;
3278
3279         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3280         mutex_lock(&block_group->cache_mutex);
3281         cache_block_group(root, block_group);
3282         mutex_unlock(&block_group->cache_mutex);
3283
3284         ret = btrfs_remove_free_space(block_group, ins->objectid,
3285                                       ins->offset);
3286         BUG_ON(ret);
3287         put_block_group(block_group);
3288         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3289                                             ref_generation, owner, ins);
3290         return ret;
3291 }
3292
3293 /*
3294  * finds a free extent and does all the dirty work required for allocation
3295  * returns the key for the extent through ins, and a tree buffer for
3296  * the first block of the extent through buf.
3297  *
3298  * returns 0 if everything worked, non-zero otherwise.
3299  */
3300 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3301                        struct btrfs_root *root,
3302                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3303                        u64 root_objectid, u64 ref_generation,
3304                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3305                        u64 search_end, struct btrfs_key *ins, u64 data)
3306 {
3307         int ret;
3308
3309         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3310                                      min_alloc_size, empty_size, hint_byte,
3311                                      search_end, ins, data);
3312         BUG_ON(ret);
3313         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3314                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3315                                         root_objectid, ref_generation,
3316                                         owner_objectid, ins);
3317                 BUG_ON(ret);
3318
3319         } else {
3320                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3321         }
3322         return ret;
3323 }
3324
3325 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3326                                             struct btrfs_root *root,
3327                                             u64 bytenr, u32 blocksize)
3328 {
3329         struct extent_buffer *buf;
3330
3331         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3332         if (!buf)
3333                 return ERR_PTR(-ENOMEM);
3334         btrfs_set_header_generation(buf, trans->transid);
3335         btrfs_tree_lock(buf);
3336         clean_tree_block(trans, root, buf);
3337         btrfs_set_buffer_uptodate(buf);
3338         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3339                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3340                          buf->start + buf->len - 1, GFP_NOFS);
3341         } else {
3342                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3343                          buf->start + buf->len - 1, GFP_NOFS);
3344         }
3345         trans->blocks_used++;
3346         return buf;
3347 }
3348
3349 /*
3350  * helper function to allocate a block for a given tree
3351  * returns the tree buffer or NULL.
3352  */
3353 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3354                                              struct btrfs_root *root,
3355                                              u32 blocksize, u64 parent,
3356                                              u64 root_objectid,
3357                                              u64 ref_generation,
3358                                              int level,
3359                                              u64 hint,
3360                                              u64 empty_size)
3361 {
3362         struct btrfs_key ins;
3363         int ret;
3364         struct extent_buffer *buf;
3365
3366         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3367                                  root_objectid, ref_generation, level,
3368                                  empty_size, hint, (u64)-1, &ins, 0);
3369         if (ret) {
3370                 BUG_ON(ret > 0);
3371                 return ERR_PTR(ret);
3372         }
3373
3374         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3375         return buf;
3376 }
3377
3378 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3379                         struct btrfs_root *root, struct extent_buffer *leaf)
3380 {
3381         u64 leaf_owner;
3382         u64 leaf_generation;
3383         struct btrfs_key key;
3384         struct btrfs_file_extent_item *fi;
3385         int i;
3386         int nritems;
3387         int ret;
3388
3389         BUG_ON(!btrfs_is_leaf(leaf));
3390         nritems = btrfs_header_nritems(leaf);
3391         leaf_owner = btrfs_header_owner(leaf);
3392         leaf_generation = btrfs_header_generation(leaf);
3393
3394         for (i = 0; i < nritems; i++) {
3395                 u64 disk_bytenr;
3396                 cond_resched();
3397
3398                 btrfs_item_key_to_cpu(leaf, &key, i);
3399                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3400                         continue;
3401                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3402                 if (btrfs_file_extent_type(leaf, fi) ==
3403                     BTRFS_FILE_EXTENT_INLINE)
3404                         continue;
3405                 /*
3406                  * FIXME make sure to insert a trans record that
3407                  * repeats the snapshot del on crash
3408                  */
3409                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3410                 if (disk_bytenr == 0)
3411                         continue;
3412
3413                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3414                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3415                                 leaf->start, leaf_owner, leaf_generation,
3416                                 key.objectid, 0);
3417                 BUG_ON(ret);
3418
3419                 atomic_inc(&root->fs_info->throttle_gen);
3420                 wake_up(&root->fs_info->transaction_throttle);
3421                 cond_resched();
3422         }
3423         return 0;
3424 }
3425
3426 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3427                                         struct btrfs_root *root,
3428                                         struct btrfs_leaf_ref *ref)
3429 {
3430         int i;
3431         int ret;
3432         struct btrfs_extent_info *info = ref->extents;
3433
3434         for (i = 0; i < ref->nritems; i++) {
3435                 ret = __btrfs_free_extent(trans, root, info->bytenr,
3436                                           info->num_bytes, ref->bytenr,
3437                                           ref->owner, ref->generation,
3438                                           info->objectid, 0);
3439
3440                 atomic_inc(&root->fs_info->throttle_gen);
3441                 wake_up(&root->fs_info->transaction_throttle);
3442                 cond_resched();
3443
3444                 BUG_ON(ret);
3445                 info++;
3446         }
3447
3448         return 0;
3449 }
3450
3451 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
3452                               u32 *refs)
3453 {
3454         int ret;
3455
3456         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3457         BUG_ON(ret);
3458
3459 #if 0 // some debugging code in case we see problems here
3460         /* if the refs count is one, it won't get increased again.  But
3461          * if the ref count is > 1, someone may be decreasing it at
3462          * the same time we are.
3463          */
3464         if (*refs != 1) {
3465                 struct extent_buffer *eb = NULL;
3466                 eb = btrfs_find_create_tree_block(root, start, len);
3467                 if (eb)
3468                         btrfs_tree_lock(eb);
3469
3470                 mutex_lock(&root->fs_info->alloc_mutex);
3471                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3472                 BUG_ON(ret);
3473                 mutex_unlock(&root->fs_info->alloc_mutex);
3474
3475                 if (eb) {
3476                         btrfs_tree_unlock(eb);
3477                         free_extent_buffer(eb);
3478                 }
3479                 if (*refs == 1) {
3480                         printk("block %llu went down to one during drop_snap\n",
3481                                (unsigned long long)start);
3482                 }
3483
3484         }
3485 #endif
3486
3487         cond_resched();
3488         return ret;
3489 }
3490
3491 /*
3492  * helper function for drop_snapshot, this walks down the tree dropping ref
3493  * counts as it goes.
3494  */
3495 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
3496                                    struct btrfs_root *root,
3497                                    struct btrfs_path *path, int *level)
3498 {
3499         u64 root_owner;
3500         u64 root_gen;
3501         u64 bytenr;
3502         u64 ptr_gen;
3503         struct extent_buffer *next;
3504         struct extent_buffer *cur;
3505         struct extent_buffer *parent;
3506         struct btrfs_leaf_ref *ref;
3507         u32 blocksize;
3508         int ret;
3509         u32 refs;
3510
3511         WARN_ON(*level < 0);
3512         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3513         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3514                                 path->nodes[*level]->len, &refs);
3515         BUG_ON(ret);
3516         if (refs > 1)
3517                 goto out;
3518
3519         /*
3520          * walk down to the last node level and free all the leaves
3521          */
3522         while(*level >= 0) {
3523                 WARN_ON(*level < 0);
3524                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3525                 cur = path->nodes[*level];
3526
3527                 if (btrfs_header_level(cur) != *level)
3528                         WARN_ON(1);
3529
3530                 if (path->slots[*level] >=
3531                     btrfs_header_nritems(cur))
3532                         break;
3533                 if (*level == 0) {
3534                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3535                         BUG_ON(ret);
3536                         break;
3537                 }
3538                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3539                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3540                 blocksize = btrfs_level_size(root, *level - 1);
3541
3542                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3543                 BUG_ON(ret);
3544                 if (refs != 1) {
3545                         parent = path->nodes[*level];
3546                         root_owner = btrfs_header_owner(parent);
3547                         root_gen = btrfs_header_generation(parent);
3548                         path->slots[*level]++;
3549
3550                         ret = __btrfs_free_extent(trans, root, bytenr,
3551                                                 blocksize, parent->start,
3552                                                 root_owner, root_gen,
3553                                                 *level - 1, 1);
3554                         BUG_ON(ret);
3555
3556                         atomic_inc(&root->fs_info->throttle_gen);
3557                         wake_up(&root->fs_info->transaction_throttle);
3558                         cond_resched();
3559
3560                         continue;
3561                 }
3562                 /*
3563                  * at this point, we have a single ref, and since the
3564                  * only place referencing this extent is a dead root
3565                  * the reference count should never go higher.
3566                  * So, we don't need to check it again
3567                  */
3568                 if (*level == 1) {
3569                         ref = btrfs_lookup_leaf_ref(root, bytenr);
3570                         if (ref && ref->generation != ptr_gen) {
3571                                 btrfs_free_leaf_ref(root, ref);
3572                                 ref = NULL;
3573                         }
3574                         if (ref) {
3575                                 ret = cache_drop_leaf_ref(trans, root, ref);
3576                                 BUG_ON(ret);
3577                                 btrfs_remove_leaf_ref(root, ref);
3578                                 btrfs_free_leaf_ref(root, ref);
3579                                 *level = 0;
3580                                 break;
3581                         }
3582                         if (printk_ratelimit()) {
3583                                 printk("leaf ref miss for bytenr %llu\n",
3584                                        (unsigned long long)bytenr);
3585                         }
3586                 }
3587                 next = btrfs_find_tree_block(root, bytenr, blocksize);
3588                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3589                         free_extent_buffer(next);
3590
3591                         next = read_tree_block(root, bytenr, blocksize,
3592                                                ptr_gen);
3593                         cond_resched();
3594 #if 0
3595                         /*
3596                          * this is a debugging check and can go away
3597                          * the ref should never go all the way down to 1
3598                          * at this point
3599                          */
3600                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3601                                                 &refs);
3602                         BUG_ON(ret);
3603                         WARN_ON(refs != 1);
3604 #endif
3605                 }
3606                 WARN_ON(*level <= 0);
3607                 if (path->nodes[*level-1])
3608                         free_extent_buffer(path->nodes[*level-1]);
3609                 path->nodes[*level-1] = next;
3610                 *level = btrfs_header_level(next);
3611                 path->slots[*level] = 0;
3612                 cond_resched();
3613         }
3614 out:
3615         WARN_ON(*level < 0);
3616         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3617
3618         if (path->nodes[*level] == root->node) {
3619                 parent = path->nodes[*level];
3620                 bytenr = path->nodes[*level]->start;
3621         } else {
3622                 parent = path->nodes[*level + 1];
3623                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3624         }
3625
3626         blocksize = btrfs_level_size(root, *level);
3627         root_owner = btrfs_header_owner(parent);
3628         root_gen = btrfs_header_generation(parent);
3629
3630         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3631                                   parent->start, root_owner, root_gen,
3632                                   *level, 1);
3633         free_extent_buffer(path->nodes[*level]);
3634         path->nodes[*level] = NULL;
3635         *level += 1;
3636         BUG_ON(ret);
3637
3638         cond_resched();
3639         return 0;
3640 }
3641
3642 /*
3643  * helper function for drop_subtree, this function is similar to
3644  * walk_down_tree. The main difference is that it checks reference
3645  * counts while tree blocks are locked.
3646  */
3647 static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
3648                                       struct btrfs_root *root,
3649                                       struct btrfs_path *path, int *level)
3650 {
3651         struct extent_buffer *next;
3652         struct extent_buffer *cur;
3653         struct extent_buffer *parent;
3654         u64 bytenr;
3655         u64 ptr_gen;
3656         u32 blocksize;
3657         u32 refs;
3658         int ret;
3659
3660         cur = path->nodes[*level];
3661         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3662                                       &refs);
3663         BUG_ON(ret);
3664         if (refs > 1)
3665                 goto out;
3666
3667         while (*level >= 0) {
3668                 cur = path->nodes[*level];
3669                 if (*level == 0) {
3670                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3671                         BUG_ON(ret);
3672                         clean_tree_block(trans, root, cur);
3673                         break;
3674                 }
3675                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3676                         clean_tree_block(trans, root, cur);
3677                         break;
3678                 }
3679
3680                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3681                 blocksize = btrfs_level_size(root, *level - 1);
3682                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3683
3684                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3685                 btrfs_tree_lock(next);
3686
3687                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3688                                               &refs);
3689                 BUG_ON(ret);
3690                 if (refs > 1) {
3691                         parent = path->nodes[*level];
3692                         ret = btrfs_free_extent(trans, root, bytenr,
3693                                         blocksize, parent->start,
3694                                         btrfs_header_owner(parent),
3695                                         btrfs_header_generation(parent),
3696                                         *level - 1, 1);
3697                         BUG_ON(ret);
3698                         path->slots[*level]++;
3699                         btrfs_tree_unlock(next);
3700                         free_extent_buffer(next);
3701                         continue;
3702                 }
3703
3704                 *level = btrfs_header_level(next);
3705                 path->nodes[*level] = next;
3706                 path->slots[*level] = 0;
3707                 path->locks[*level] = 1;
3708                 cond_resched();
3709         }
3710 out:
3711         parent = path->nodes[*level + 1];
3712         bytenr = path->nodes[*level]->start;
3713         blocksize = path->nodes[*level]->len;
3714
3715         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3716                         parent->start, btrfs_header_owner(parent),
3717                         btrfs_header_generation(parent), *level, 1);
3718         BUG_ON(ret);
3719
3720         if (path->locks[*level]) {
3721                 btrfs_tree_unlock(path->nodes[*level]);
3722                 path->locks[*level] = 0;
3723         }
3724         free_extent_buffer(path->nodes[*level]);
3725         path->nodes[*level] = NULL;
3726         *level += 1;
3727         cond_resched();
3728         return 0;
3729 }
3730
3731 /*
3732  * helper for dropping snapshots.  This walks back up the tree in the path
3733  * to find the first node higher up where we haven't yet gone through
3734  * all the slots
3735  */
3736 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3737                                  struct btrfs_root *root,
3738                                  struct btrfs_path *path,
3739                                  int *level, int max_level)
3740 {
3741         u64 root_owner;
3742         u64 root_gen;
3743         struct btrfs_root_item *root_item = &root->root_item;
3744         int i;
3745         int slot;
3746         int ret;
3747
3748         for (i = *level; i < max_level && path->nodes[i]; i++) {
3749                 slot = path->slots[i];
3750                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3751                         struct extent_buffer *node;
3752                         struct btrfs_disk_key disk_key;
3753                         node = path->nodes[i];
3754                         path->slots[i]++;
3755                         *level = i;
3756                         WARN_ON(*level == 0);
3757                         btrfs_node_key(node, &disk_key, path->slots[i]);
3758                         memcpy(&root_item->drop_progress,
3759                                &disk_key, sizeof(disk_key));
3760                         root_item->drop_level = i;
3761                         return 0;
3762                 } else {
3763                         struct extent_buffer *parent;
3764                         if (path->nodes[*level] == root->node)
3765                                 parent = path->nodes[*level];
3766                         else
3767                                 parent = path->nodes[*level + 1];
3768
3769                         root_owner = btrfs_header_owner(parent);
3770                         root_gen = btrfs_header_generation(parent);
3771
3772                         clean_tree_block(trans, root, path->nodes[*level]);
3773                         ret = btrfs_free_extent(trans, root,
3774                                                 path->nodes[*level]->start,
3775                                                 path->nodes[*level]->len,
3776                                                 parent->start, root_owner,
3777                                                 root_gen, *level, 1);
3778                         BUG_ON(ret);
3779                         if (path->locks[*level]) {
3780                                 btrfs_tree_unlock(path->nodes[*level]);
3781                                 path->locks[*level] = 0;
3782                         }
3783                         free_extent_buffer(path->nodes[*level]);
3784                         path->nodes[*level] = NULL;
3785                         *level = i + 1;
3786                 }
3787         }
3788         return 1;
3789 }
3790
3791 /*
3792  * drop the reference count on the tree rooted at 'snap'.  This traverses
3793  * the tree freeing any blocks that have a ref count of zero after being
3794  * decremented.
3795  */
3796 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3797                         *root)
3798 {
3799         int ret = 0;
3800         int wret;
3801         int level;
3802         struct btrfs_path *path;
3803         int i;
3804         int orig_level;
3805         struct btrfs_root_item *root_item = &root->root_item;
3806
3807         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3808         path = btrfs_alloc_path();
3809         BUG_ON(!path);
3810
3811         level = btrfs_header_level(root->node);
3812         orig_level = level;
3813         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3814                 path->nodes[level] = root->node;
3815                 extent_buffer_get(root->node);
3816                 path->slots[level] = 0;
3817         } else {
3818                 struct btrfs_key key;
3819                 struct btrfs_disk_key found_key;
3820                 struct extent_buffer *node;
3821
3822                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3823                 level = root_item->drop_level;
3824                 path->lowest_level = level;
3825                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3826                 if (wret < 0) {
3827                         ret = wret;
3828                         goto out;
3829                 }
3830                 node = path->nodes[level];
3831                 btrfs_node_key(node, &found_key, path->slots[level]);
3832                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3833                                sizeof(found_key)));
3834                 /*
3835                  * unlock our path, this is safe because only this
3836                  * function is allowed to delete this snapshot
3837                  */
3838                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3839                         if (path->nodes[i] && path->locks[i]) {
3840                                 path->locks[i] = 0;
3841                                 btrfs_tree_unlock(path->nodes[i]);
3842                         }
3843                 }
3844         }
3845         while(1) {
3846                 wret = walk_down_tree(trans, root, path, &level);
3847                 if (wret > 0)
3848                         break;
3849                 if (wret < 0)
3850                         ret = wret;
3851
3852                 wret = walk_up_tree(trans, root, path, &level,
3853                                     BTRFS_MAX_LEVEL);
3854                 if (wret > 0)
3855                         break;
3856                 if (wret < 0)
3857                         ret = wret;
3858                 if (trans->transaction->in_commit) {
3859                         ret = -EAGAIN;
3860                         break;
3861                 }
3862                 atomic_inc(&root->fs_info->throttle_gen);
3863                 wake_up(&root->fs_info->transaction_throttle);
3864         }
3865         for (i = 0; i <= orig_level; i++) {
3866                 if (path->nodes[i]) {
3867                         free_extent_buffer(path->nodes[i]);
3868                         path->nodes[i] = NULL;
3869                 }
3870         }
3871 out:
3872         btrfs_free_path(path);
3873         return ret;
3874 }
3875
3876 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3877                         struct btrfs_root *root,
3878                         struct extent_buffer *node,
3879                         struct extent_buffer *parent)
3880 {
3881         struct btrfs_path *path;
3882         int level;
3883         int parent_level;
3884         int ret = 0;
3885         int wret;
3886
3887         path = btrfs_alloc_path();
3888         BUG_ON(!path);
3889
3890         BUG_ON(!btrfs_tree_locked(parent));
3891         parent_level = btrfs_header_level(parent);
3892         extent_buffer_get(parent);
3893         path->nodes[parent_level] = parent;
3894         path->slots[parent_level] = btrfs_header_nritems(parent);
3895
3896         BUG_ON(!btrfs_tree_locked(node));
3897         level = btrfs_header_level(node);
3898         extent_buffer_get(node);
3899         path->nodes[level] = node;
3900         path->slots[level] = 0;
3901
3902         while (1) {
3903                 wret = walk_down_subtree(trans, root, path, &level);
3904                 if (wret < 0)
3905                         ret = wret;
3906                 if (wret != 0)
3907                         break;
3908
3909                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3910                 if (wret < 0)
3911                         ret = wret;
3912                 if (wret != 0)
3913                         break;
3914         }
3915
3916         btrfs_free_path(path);
3917         return ret;
3918 }
3919
3920 static unsigned long calc_ra(unsigned long start, unsigned long last,
3921                              unsigned long nr)
3922 {
3923         return min(last, start + nr - 1);
3924 }
3925
3926 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3927                                          u64 len)
3928 {
3929         u64 page_start;
3930         u64 page_end;
3931         unsigned long first_index;
3932         unsigned long last_index;
3933         unsigned long i;
3934         struct page *page;
3935         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3936         struct file_ra_state *ra;
3937         struct btrfs_ordered_extent *ordered;
3938         unsigned int total_read = 0;
3939         unsigned int total_dirty = 0;
3940         int ret = 0;
3941
3942         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3943
3944         mutex_lock(&inode->i_mutex);
3945         first_index = start >> PAGE_CACHE_SHIFT;
3946         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3947
3948         /* make sure the dirty trick played by the caller work */
3949         ret = invalidate_inode_pages2_range(inode->i_mapping,
3950                                             first_index, last_index);
3951         if (ret)
3952                 goto out_unlock;
3953
3954         file_ra_state_init(ra, inode->i_mapping);
3955
3956         for (i = first_index ; i <= last_index; i++) {
3957                 if (total_read % ra->ra_pages == 0) {
3958                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3959                                        calc_ra(i, last_index, ra->ra_pages));
3960                 }
3961                 total_read++;
3962 again:
3963                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3964                         BUG_ON(1);
3965                 page = grab_cache_page(inode->i_mapping, i);
3966                 if (!page) {
3967                         ret = -ENOMEM;
3968                         goto out_unlock;
3969                 }
3970                 if (!PageUptodate(page)) {
3971                         btrfs_readpage(NULL, page);
3972                         lock_page(page);
3973                         if (!PageUptodate(page)) {
3974                                 unlock_page(page);
3975                                 page_cache_release(page);
3976                                 ret = -EIO;
3977                                 goto out_unlock;
3978                         }
3979                 }
3980                 wait_on_page_writeback(page);
3981
3982                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3983                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3984                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3985
3986                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3987                 if (ordered) {
3988                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3989                         unlock_page(page);
3990                         page_cache_release(page);
3991                         btrfs_start_ordered_extent(inode, ordered, 1);
3992                         btrfs_put_ordered_extent(ordered);
3993                         goto again;
3994                 }
3995                 set_page_extent_mapped(page);
3996
3997                 if (i == first_index)
3998                         set_extent_bits(io_tree, page_start, page_end,
3999                                         EXTENT_BOUNDARY, GFP_NOFS);
4000                 btrfs_set_extent_delalloc(inode, page_start, page_end);
4001
4002                 set_page_dirty(page);
4003                 total_dirty++;
4004
4005                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4006                 unlock_page(page);
4007                 page_cache_release(page);
4008         }
4009
4010 out_unlock:
4011         kfree(ra);
4012         mutex_unlock(&inode->i_mutex);
4013         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4014         return ret;
4015 }
4016
4017 static int noinline relocate_data_extent(struct inode *reloc_inode,
4018                                          struct btrfs_key *extent_key,
4019                                          u64 offset)
4020 {
4021         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4022         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4023         struct extent_map *em;
4024         u64 start = extent_key->objectid - offset;
4025         u64 end = start + extent_key->offset - 1;
4026
4027         em = alloc_extent_map(GFP_NOFS);
4028         BUG_ON(!em || IS_ERR(em));
4029
4030         em->start = start;
4031         em->len = extent_key->offset;
4032         em->block_len = extent_key->offset;
4033         em->block_start = extent_key->objectid;
4034         em->bdev = root->fs_info->fs_devices->latest_bdev;
4035         set_bit(EXTENT_FLAG_PINNED, &em->flags);
4036
4037         /* setup extent map to cheat btrfs_readpage */
4038         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4039         while (1) {
4040                 int ret;
4041                 spin_lock(&em_tree->lock);
4042                 ret = add_extent_mapping(em_tree, em);
4043                 spin_unlock(&em_tree->lock);
4044                 if (ret != -EEXIST) {
4045                         free_extent_map(em);
4046                         break;
4047                 }
4048                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4049         }
4050         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4051
4052         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4053 }
4054
4055 struct btrfs_ref_path {
4056         u64 extent_start;
4057         u64 nodes[BTRFS_MAX_LEVEL];
4058         u64 root_objectid;
4059         u64 root_generation;
4060         u64 owner_objectid;
4061         u32 num_refs;
4062         int lowest_level;
4063         int current_level;
4064         int shared_level;
4065
4066         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4067         u64 new_nodes[BTRFS_MAX_LEVEL];
4068 };
4069
4070 struct disk_extent {
4071         u64 ram_bytes;
4072         u64 disk_bytenr;
4073         u64 disk_num_bytes;
4074         u64 offset;
4075         u64 num_bytes;
4076         u8 compression;
4077         u8 encryption;
4078         u16 other_encoding;
4079 };
4080
4081 static int is_cowonly_root(u64 root_objectid)
4082 {
4083         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4084             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4085             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4086             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4087             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4088             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4089                 return 1;
4090         return 0;
4091 }
4092
4093 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
4094                                     struct btrfs_root *extent_root,
4095                                     struct btrfs_ref_path *ref_path,
4096                                     int first_time)
4097 {
4098         struct extent_buffer *leaf;
4099         struct btrfs_path *path;
4100         struct btrfs_extent_ref *ref;
4101         struct btrfs_key key;
4102         struct btrfs_key found_key;
4103         u64 bytenr;
4104         u32 nritems;
4105         int level;
4106         int ret = 1;
4107
4108         path = btrfs_alloc_path();
4109         if (!path)
4110                 return -ENOMEM;
4111
4112         if (first_time) {
4113                 ref_path->lowest_level = -1;
4114                 ref_path->current_level = -1;
4115                 ref_path->shared_level = -1;
4116                 goto walk_up;
4117         }
4118 walk_down:
4119         level = ref_path->current_level - 1;
4120         while (level >= -1) {
4121                 u64 parent;
4122                 if (level < ref_path->lowest_level)
4123                         break;
4124
4125                 if (level >= 0) {
4126                         bytenr = ref_path->nodes[level];
4127                 } else {
4128                         bytenr = ref_path->extent_start;
4129                 }
4130                 BUG_ON(bytenr == 0);
4131
4132                 parent = ref_path->nodes[level + 1];
4133                 ref_path->nodes[level + 1] = 0;
4134                 ref_path->current_level = level;
4135                 BUG_ON(parent == 0);
4136
4137                 key.objectid = bytenr;
4138                 key.offset = parent + 1;
4139                 key.type = BTRFS_EXTENT_REF_KEY;
4140
4141                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4142                 if (ret < 0)
4143                         goto out;
4144                 BUG_ON(ret == 0);
4145
4146                 leaf = path->nodes[0];
4147                 nritems = btrfs_header_nritems(leaf);
4148                 if (path->slots[0] >= nritems) {
4149                         ret = btrfs_next_leaf(extent_root, path);
4150                         if (ret < 0)
4151                                 goto out;
4152                         if (ret > 0)
4153                                 goto next;
4154                         leaf = path->nodes[0];
4155                 }
4156
4157                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4158                 if (found_key.objectid == bytenr &&
4159                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4160                         if (level < ref_path->shared_level)
4161                                 ref_path->shared_level = level;
4162                         goto found;
4163                 }
4164 next:
4165                 level--;
4166                 btrfs_release_path(extent_root, path);
4167                 cond_resched();
4168         }
4169         /* reached lowest level */
4170         ret = 1;
4171         goto out;
4172 walk_up:
4173         level = ref_path->current_level;
4174         while (level < BTRFS_MAX_LEVEL - 1) {
4175                 u64 ref_objectid;
4176                 if (level >= 0) {
4177                         bytenr = ref_path->nodes[level];
4178                 } else {
4179                         bytenr = ref_path->extent_start;
4180                 }
4181                 BUG_ON(bytenr == 0);
4182
4183                 key.objectid = bytenr;
4184                 key.offset = 0;
4185                 key.type = BTRFS_EXTENT_REF_KEY;
4186
4187                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4188                 if (ret < 0)
4189                         goto out;
4190
4191                 leaf = path->nodes[0];
4192                 nritems = btrfs_header_nritems(leaf);
4193                 if (path->slots[0] >= nritems) {
4194                         ret = btrfs_next_leaf(extent_root, path);
4195                         if (ret < 0)
4196                                 goto out;
4197                         if (ret > 0) {
4198                                 /* the extent was freed by someone */
4199                                 if (ref_path->lowest_level == level)
4200                                         goto out;
4201                                 btrfs_release_path(extent_root, path);
4202                                 goto walk_down;
4203                         }
4204                         leaf = path->nodes[0];
4205                 }
4206
4207                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4208                 if (found_key.objectid != bytenr ||
4209                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4210                         /* the extent was freed by someone */
4211                         if (ref_path->lowest_level == level) {
4212                                 ret = 1;
4213                                 goto out;
4214                         }
4215                         btrfs_release_path(extent_root, path);
4216                         goto walk_down;
4217                 }
4218 found:
4219                 ref = btrfs_item_ptr(leaf, path->slots[0],
4220                                 struct btrfs_extent_ref);
4221                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4222                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4223                         if (first_time) {
4224                                 level = (int)ref_objectid;
4225                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4226                                 ref_path->lowest_level = level;
4227                                 ref_path->current_level = level;
4228                                 ref_path->nodes[level] = bytenr;
4229                         } else {
4230                                 WARN_ON(ref_objectid != level);
4231                         }
4232                 } else {
4233                         WARN_ON(level != -1);
4234                 }
4235                 first_time = 0;
4236
4237                 if (ref_path->lowest_level == level) {
4238                         ref_path->owner_objectid = ref_objectid;
4239                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4240                 }
4241
4242                 /*
4243                  * the block is tree root or the block isn't in reference
4244                  * counted tree.
4245                  */
4246                 if (found_key.objectid == found_key.offset ||
4247                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4248                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4249                         ref_path->root_generation =
4250                                 btrfs_ref_generation(leaf, ref);
4251                         if (level < 0) {
4252                                 /* special reference from the tree log */
4253                                 ref_path->nodes[0] = found_key.offset;
4254                                 ref_path->current_level = 0;
4255                         }
4256                         ret = 0;
4257                         goto out;
4258                 }
4259
4260                 level++;
4261                 BUG_ON(ref_path->nodes[level] != 0);
4262                 ref_path->nodes[level] = found_key.offset;
4263                 ref_path->current_level = level;
4264
4265                 /*
4266                  * the reference was created in the running transaction,
4267                  * no need to continue walking up.
4268                  */
4269                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4270                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4271                         ref_path->root_generation =
4272                                 btrfs_ref_generation(leaf, ref);
4273                         ret = 0;
4274                         goto out;
4275                 }
4276
4277                 btrfs_release_path(extent_root, path);
4278                 cond_resched();
4279         }
4280         /* reached max tree level, but no tree root found. */
4281         BUG();
4282 out:
4283         btrfs_free_path(path);
4284         return ret;
4285 }
4286
4287 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4288                                 struct btrfs_root *extent_root,
4289                                 struct btrfs_ref_path *ref_path,
4290                                 u64 extent_start)
4291 {
4292         memset(ref_path, 0, sizeof(*ref_path));
4293         ref_path->extent_start = extent_start;
4294
4295         return __next_ref_path(trans, extent_root, ref_path, 1);
4296 }
4297
4298 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4299                                struct btrfs_root *extent_root,
4300                                struct btrfs_ref_path *ref_path)
4301 {
4302         return __next_ref_path(trans, extent_root, ref_path, 0);
4303 }
4304
4305 static int noinline get_new_locations(struct inode *reloc_inode,
4306                                       struct btrfs_key *extent_key,
4307                                       u64 offset, int no_fragment,
4308                                       struct disk_extent **extents,
4309                                       int *nr_extents)
4310 {
4311         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4312         struct btrfs_path *path;
4313         struct btrfs_file_extent_item *fi;
4314         struct extent_buffer *leaf;
4315         struct disk_extent *exts = *extents;
4316         struct btrfs_key found_key;
4317         u64 cur_pos;
4318         u64 last_byte;
4319         u32 nritems;
4320         int nr = 0;
4321         int max = *nr_extents;
4322         int ret;
4323
4324         WARN_ON(!no_fragment && *extents);
4325         if (!exts) {
4326                 max = 1;
4327                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4328                 if (!exts)
4329                         return -ENOMEM;
4330         }
4331
4332         path = btrfs_alloc_path();
4333         BUG_ON(!path);
4334
4335         cur_pos = extent_key->objectid - offset;
4336         last_byte = extent_key->objectid + extent_key->offset;
4337         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4338                                        cur_pos, 0);
4339         if (ret < 0)
4340                 goto out;
4341         if (ret > 0) {
4342                 ret = -ENOENT;
4343                 goto out;
4344         }
4345
4346         while (1) {
4347                 leaf = path->nodes[0];
4348                 nritems = btrfs_header_nritems(leaf);
4349                 if (path->slots[0] >= nritems) {
4350                         ret = btrfs_next_leaf(root, path);
4351                         if (ret < 0)
4352                                 goto out;
4353                         if (ret > 0)
4354                                 break;
4355                         leaf = path->nodes[0];
4356                 }
4357
4358                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4359                 if (found_key.offset != cur_pos ||
4360                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4361                     found_key.objectid != reloc_inode->i_ino)
4362                         break;
4363
4364                 fi = btrfs_item_ptr(leaf, path->slots[0],
4365                                     struct btrfs_file_extent_item);
4366                 if (btrfs_file_extent_type(leaf, fi) !=
4367                     BTRFS_FILE_EXTENT_REG ||
4368                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4369                         break;
4370
4371                 if (nr == max) {
4372                         struct disk_extent *old = exts;
4373                         max *= 2;
4374                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4375                         memcpy(exts, old, sizeof(*exts) * nr);
4376                         if (old != *extents)
4377                                 kfree(old);
4378                 }
4379
4380                 exts[nr].disk_bytenr =
4381                         btrfs_file_extent_disk_bytenr(leaf, fi);
4382                 exts[nr].disk_num_bytes =
4383                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4384                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4385                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4386                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4387                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4388                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4389                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4390                                                                            fi);
4391                 BUG_ON(exts[nr].offset > 0);
4392                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4393                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4394
4395                 cur_pos += exts[nr].num_bytes;
4396                 nr++;
4397
4398                 if (cur_pos + offset >= last_byte)
4399                         break;
4400
4401                 if (no_fragment) {
4402                         ret = 1;
4403                         goto out;
4404                 }
4405                 path->slots[0]++;
4406         }
4407
4408         BUG_ON(cur_pos + offset > last_byte);
4409         if (cur_pos + offset < last_byte) {
4410                 ret = -ENOENT;
4411                 goto out;
4412         }
4413         ret = 0;
4414 out:
4415         btrfs_free_path(path);
4416         if (ret) {
4417                 if (exts != *extents)
4418                         kfree(exts);
4419         } else {
4420                 *extents = exts;
4421                 *nr_extents = nr;
4422         }
4423         return ret;
4424 }
4425
4426 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
4427                                         struct btrfs_root *root,
4428                                         struct btrfs_path *path,
4429                                         struct btrfs_key *extent_key,
4430                                         struct btrfs_key *leaf_key,
4431                                         struct btrfs_ref_path *ref_path,
4432                                         struct disk_extent *new_extents,
4433                                         int nr_extents)
4434 {
4435         struct extent_buffer *leaf;
4436         struct btrfs_file_extent_item *fi;
4437         struct inode *inode = NULL;
4438         struct btrfs_key key;
4439         u64 lock_start = 0;
4440         u64 lock_end = 0;
4441         u64 num_bytes;
4442         u64 ext_offset;
4443         u64 first_pos;
4444         u32 nritems;
4445         int nr_scaned = 0;
4446         int extent_locked = 0;
4447         int extent_type;
4448         int ret;
4449
4450         memcpy(&key, leaf_key, sizeof(key));
4451         first_pos = INT_LIMIT(loff_t) - extent_key->offset;
4452         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4453                 if (key.objectid < ref_path->owner_objectid ||
4454                     (key.objectid == ref_path->owner_objectid &&
4455                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4456                         key.objectid = ref_path->owner_objectid;
4457                         key.type = BTRFS_EXTENT_DATA_KEY;
4458                         key.offset = 0;
4459                 }
4460         }
4461
4462         while (1) {
4463                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4464                 if (ret < 0)
4465                         goto out;
4466
4467                 leaf = path->nodes[0];
4468                 nritems = btrfs_header_nritems(leaf);
4469 next:
4470                 if (extent_locked && ret > 0) {
4471                         /*
4472                          * the file extent item was modified by someone
4473                          * before the extent got locked.
4474                          */
4475                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4476                                       lock_end, GFP_NOFS);
4477                         extent_locked = 0;
4478                 }
4479
4480                 if (path->slots[0] >= nritems) {
4481                         if (++nr_scaned > 2)
4482                                 break;
4483
4484                         BUG_ON(extent_locked);
4485                         ret = btrfs_next_leaf(root, path);
4486                         if (ret < 0)
4487                                 goto out;
4488                         if (ret > 0)
4489                                 break;
4490                         leaf = path->nodes[0];
4491                         nritems = btrfs_header_nritems(leaf);
4492                 }
4493
4494                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4495
4496                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4497                         if ((key.objectid > ref_path->owner_objectid) ||
4498                             (key.objectid == ref_path->owner_objectid &&
4499                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4500                             (key.offset >= first_pos + extent_key->offset))
4501                                 break;
4502                 }
4503
4504                 if (inode && key.objectid != inode->i_ino) {
4505                         BUG_ON(extent_locked);
4506                         btrfs_release_path(root, path);
4507                         mutex_unlock(&inode->i_mutex);
4508                         iput(inode);
4509                         inode = NULL;
4510                         continue;
4511                 }
4512
4513                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4514                         path->slots[0]++;
4515                         ret = 1;
4516                         goto next;
4517                 }
4518                 fi = btrfs_item_ptr(leaf, path->slots[0],
4519                                     struct btrfs_file_extent_item);
4520                 extent_type = btrfs_file_extent_type(leaf, fi);
4521                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4522                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4523                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4524                      extent_key->objectid)) {
4525                         path->slots[0]++;
4526                         ret = 1;
4527                         goto next;
4528                 }
4529
4530                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4531                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4532
4533                 if (first_pos > key.offset - ext_offset)
4534                         first_pos = key.offset - ext_offset;
4535
4536                 if (!extent_locked) {
4537                         lock_start = key.offset;
4538                         lock_end = lock_start + num_bytes - 1;
4539                 } else {
4540                         if (lock_start > key.offset ||
4541                             lock_end + 1 < key.offset + num_bytes) {
4542                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4543                                               lock_start, lock_end, GFP_NOFS);
4544                                 extent_locked = 0;
4545                         }
4546                 }
4547
4548                 if (!inode) {
4549                         btrfs_release_path(root, path);
4550
4551                         inode = btrfs_iget_locked(root->fs_info->sb,
4552                                                   key.objectid, root);
4553                         if (inode->i_state & I_NEW) {
4554                                 BTRFS_I(inode)->root = root;
4555                                 BTRFS_I(inode)->location.objectid =
4556                                         key.objectid;
4557                                 BTRFS_I(inode)->location.type =
4558                                         BTRFS_INODE_ITEM_KEY;
4559                                 BTRFS_I(inode)->location.offset = 0;
4560                                 btrfs_read_locked_inode(inode);
4561                                 unlock_new_inode(inode);
4562                         }
4563                         /*
4564                          * some code call btrfs_commit_transaction while
4565                          * holding the i_mutex, so we can't use mutex_lock
4566                          * here.
4567                          */
4568                         if (is_bad_inode(inode) ||
4569                             !mutex_trylock(&inode->i_mutex)) {
4570                                 iput(inode);
4571                                 inode = NULL;
4572                                 key.offset = (u64)-1;
4573                                 goto skip;
4574                         }
4575                 }
4576
4577                 if (!extent_locked) {
4578                         struct btrfs_ordered_extent *ordered;
4579
4580                         btrfs_release_path(root, path);
4581
4582                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4583                                     lock_end, GFP_NOFS);
4584                         ordered = btrfs_lookup_first_ordered_extent(inode,
4585                                                                     lock_end);
4586                         if (ordered &&
4587                             ordered->file_offset <= lock_end &&
4588                             ordered->file_offset + ordered->len > lock_start) {
4589                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4590                                               lock_start, lock_end, GFP_NOFS);
4591                                 btrfs_start_ordered_extent(inode, ordered, 1);
4592                                 btrfs_put_ordered_extent(ordered);
4593                                 key.offset += num_bytes;
4594                                 goto skip;
4595                         }
4596                         if (ordered)
4597                                 btrfs_put_ordered_extent(ordered);
4598
4599                         extent_locked = 1;
4600                         continue;
4601                 }
4602
4603                 if (nr_extents == 1) {
4604                         /* update extent pointer in place */
4605                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4606                                                 new_extents[0].disk_bytenr);
4607                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4608                                                 new_extents[0].disk_num_bytes);
4609                         btrfs_mark_buffer_dirty(leaf);
4610
4611                         btrfs_drop_extent_cache(inode, key.offset,
4612                                                 key.offset + num_bytes - 1, 0);
4613
4614                         ret = btrfs_inc_extent_ref(trans, root,
4615                                                 new_extents[0].disk_bytenr,
4616                                                 new_extents[0].disk_num_bytes,
4617                                                 leaf->start,
4618                                                 root->root_key.objectid,
4619                                                 trans->transid,
4620                                                 key.objectid);
4621                         BUG_ON(ret);
4622
4623                         ret = btrfs_free_extent(trans, root,
4624                                                 extent_key->objectid,
4625                                                 extent_key->offset,
4626                                                 leaf->start,
4627                                                 btrfs_header_owner(leaf),
4628                                                 btrfs_header_generation(leaf),
4629                                                 key.objectid, 0);
4630                         BUG_ON(ret);
4631
4632                         btrfs_release_path(root, path);
4633                         key.offset += num_bytes;
4634                 } else {
4635                         BUG_ON(1);
4636 #if 0
4637                         u64 alloc_hint;
4638                         u64 extent_len;
4639                         int i;
4640                         /*
4641                          * drop old extent pointer at first, then insert the
4642                          * new pointers one bye one
4643                          */
4644                         btrfs_release_path(root, path);
4645                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4646                                                  key.offset + num_bytes,
4647                                                  key.offset, &alloc_hint);
4648                         BUG_ON(ret);
4649
4650                         for (i = 0; i < nr_extents; i++) {
4651                                 if (ext_offset >= new_extents[i].num_bytes) {
4652                                         ext_offset -= new_extents[i].num_bytes;
4653                                         continue;
4654                                 }
4655                                 extent_len = min(new_extents[i].num_bytes -
4656                                                  ext_offset, num_bytes);
4657
4658                                 ret = btrfs_insert_empty_item(trans, root,
4659                                                               path, &key,
4660                                                               sizeof(*fi));
4661                                 BUG_ON(ret);
4662
4663                                 leaf = path->nodes[0];
4664                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4665                                                 struct btrfs_file_extent_item);
4666                                 btrfs_set_file_extent_generation(leaf, fi,
4667                                                         trans->transid);
4668                                 btrfs_set_file_extent_type(leaf, fi,
4669                                                         BTRFS_FILE_EXTENT_REG);
4670                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4671                                                 new_extents[i].disk_bytenr);
4672                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4673                                                 new_extents[i].disk_num_bytes);
4674                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4675                                                 new_extents[i].ram_bytes);
4676
4677                                 btrfs_set_file_extent_compression(leaf, fi,
4678                                                 new_extents[i].compression);
4679                                 btrfs_set_file_extent_encryption(leaf, fi,
4680                                                 new_extents[i].encryption);
4681                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4682                                                 new_extents[i].other_encoding);
4683
4684                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4685                                                         extent_len);
4686                                 ext_offset += new_extents[i].offset;
4687                                 btrfs_set_file_extent_offset(leaf, fi,
4688                                                         ext_offset);
4689                                 btrfs_mark_buffer_dirty(leaf);
4690
4691                                 btrfs_drop_extent_cache(inode, key.offset,
4692                                                 key.offset + extent_len - 1, 0);
4693
4694                                 ret = btrfs_inc_extent_ref(trans, root,
4695                                                 new_extents[i].disk_bytenr,
4696                                                 new_extents[i].disk_num_bytes,
4697                                                 leaf->start,
4698                                                 root->root_key.objectid,
4699                                                 trans->transid, key.objectid);
4700                                 BUG_ON(ret);
4701                                 btrfs_release_path(root, path);
4702
4703                                 inode_add_bytes(inode, extent_len);
4704
4705                                 ext_offset = 0;
4706                                 num_bytes -= extent_len;
4707                                 key.offset += extent_len;
4708
4709                                 if (num_bytes == 0)
4710                                         break;
4711                         }
4712                         BUG_ON(i >= nr_extents);
4713 #endif
4714                 }
4715
4716                 if (extent_locked) {
4717                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4718                                       lock_end, GFP_NOFS);
4719                         extent_locked = 0;
4720                 }
4721 skip:
4722                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4723                     key.offset >= first_pos + extent_key->offset)
4724                         break;
4725
4726                 cond_resched();
4727         }
4728         ret = 0;
4729 out:
4730         btrfs_release_path(root, path);
4731         if (inode) {
4732                 mutex_unlock(&inode->i_mutex);
4733                 if (extent_locked) {
4734                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4735                                       lock_end, GFP_NOFS);
4736                 }
4737                 iput(inode);
4738         }
4739         return ret;
4740 }
4741
4742 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4743                                struct btrfs_root *root,
4744                                struct extent_buffer *buf, u64 orig_start)
4745 {
4746         int level;
4747         int ret;
4748
4749         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4750         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4751
4752         level = btrfs_header_level(buf);
4753         if (level == 0) {
4754                 struct btrfs_leaf_ref *ref;
4755                 struct btrfs_leaf_ref *orig_ref;
4756
4757                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4758                 if (!orig_ref)
4759                         return -ENOENT;
4760
4761                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4762                 if (!ref) {
4763                         btrfs_free_leaf_ref(root, orig_ref);
4764                         return -ENOMEM;
4765                 }
4766
4767                 ref->nritems = orig_ref->nritems;
4768                 memcpy(ref->extents, orig_ref->extents,
4769                         sizeof(ref->extents[0]) * ref->nritems);
4770
4771                 btrfs_free_leaf_ref(root, orig_ref);
4772
4773                 ref->root_gen = trans->transid;
4774                 ref->bytenr = buf->start;
4775                 ref->owner = btrfs_header_owner(buf);
4776                 ref->generation = btrfs_header_generation(buf);
4777                 ret = btrfs_add_leaf_ref(root, ref, 0);
4778                 WARN_ON(ret);
4779                 btrfs_free_leaf_ref(root, ref);
4780         }
4781         return 0;
4782 }
4783
4784 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4785                                         struct extent_buffer *leaf,
4786                                         struct btrfs_block_group_cache *group,
4787                                         struct btrfs_root *target_root)
4788 {
4789         struct btrfs_key key;
4790         struct inode *inode = NULL;
4791         struct btrfs_file_extent_item *fi;
4792         u64 num_bytes;
4793         u64 skip_objectid = 0;
4794         u32 nritems;
4795         u32 i;
4796
4797         nritems = btrfs_header_nritems(leaf);
4798         for (i = 0; i < nritems; i++) {
4799                 btrfs_item_key_to_cpu(leaf, &key, i);
4800                 if (key.objectid == skip_objectid ||
4801                     key.type != BTRFS_EXTENT_DATA_KEY)
4802                         continue;
4803                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4804                 if (btrfs_file_extent_type(leaf, fi) ==
4805                     BTRFS_FILE_EXTENT_INLINE)
4806                         continue;
4807                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4808                         continue;
4809                 if (!inode || inode->i_ino != key.objectid) {
4810                         iput(inode);
4811                         inode = btrfs_ilookup(target_root->fs_info->sb,
4812                                               key.objectid, target_root, 1);
4813                 }
4814                 if (!inode) {
4815                         skip_objectid = key.objectid;
4816                         continue;
4817                 }
4818                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4819
4820                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4821                             key.offset + num_bytes - 1, GFP_NOFS);
4822                 btrfs_drop_extent_cache(inode, key.offset,
4823                                         key.offset + num_bytes - 1, 1);
4824                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4825                               key.offset + num_bytes - 1, GFP_NOFS);
4826                 cond_resched();
4827         }
4828         iput(inode);
4829         return 0;
4830 }
4831
4832 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4833                                         struct btrfs_root *root,
4834                                         struct extent_buffer *leaf,
4835                                         struct btrfs_block_group_cache *group,
4836                                         struct inode *reloc_inode)
4837 {
4838         struct btrfs_key key;
4839         struct btrfs_key extent_key;
4840         struct btrfs_file_extent_item *fi;
4841         struct btrfs_leaf_ref *ref;
4842         struct disk_extent *new_extent;
4843         u64 bytenr;
4844         u64 num_bytes;
4845         u32 nritems;
4846         u32 i;
4847         int ext_index;
4848         int nr_extent;
4849         int ret;
4850
4851         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4852         BUG_ON(!new_extent);
4853
4854         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4855         BUG_ON(!ref);
4856
4857         ext_index = -1;
4858         nritems = btrfs_header_nritems(leaf);
4859         for (i = 0; i < nritems; i++) {
4860                 btrfs_item_key_to_cpu(leaf, &key, i);
4861                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4862                         continue;
4863                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4864                 if (btrfs_file_extent_type(leaf, fi) ==
4865                     BTRFS_FILE_EXTENT_INLINE)
4866                         continue;
4867                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4868                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4869                 if (bytenr == 0)
4870                         continue;
4871
4872                 ext_index++;
4873                 if (bytenr >= group->key.objectid + group->key.offset ||
4874                     bytenr + num_bytes <= group->key.objectid)
4875                         continue;
4876
4877                 extent_key.objectid = bytenr;
4878                 extent_key.offset = num_bytes;
4879                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4880                 nr_extent = 1;
4881                 ret = get_new_locations(reloc_inode, &extent_key,
4882                                         group->key.objectid, 1,
4883                                         &new_extent, &nr_extent);
4884                 if (ret > 0)
4885                         continue;
4886                 BUG_ON(ret < 0);
4887
4888                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4889                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4890                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4891                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4892
4893                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4894                                                 new_extent->disk_bytenr);
4895                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4896                                                 new_extent->disk_num_bytes);
4897                 btrfs_mark_buffer_dirty(leaf);
4898
4899                 ret = btrfs_inc_extent_ref(trans, root,
4900                                         new_extent->disk_bytenr,
4901                                         new_extent->disk_num_bytes,
4902                                         leaf->start,
4903                                         root->root_key.objectid,
4904                                         trans->transid, key.objectid);
4905                 BUG_ON(ret);
4906                 ret = btrfs_free_extent(trans, root,
4907                                         bytenr, num_bytes, leaf->start,
4908                                         btrfs_header_owner(leaf),
4909                                         btrfs_header_generation(leaf),
4910                                         key.objectid, 0);
4911                 BUG_ON(ret);
4912                 cond_resched();
4913         }
4914         kfree(new_extent);
4915         BUG_ON(ext_index + 1 != ref->nritems);
4916         btrfs_free_leaf_ref(root, ref);
4917         return 0;
4918 }
4919
4920 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4921                           struct btrfs_root *root)
4922 {
4923         struct btrfs_root *reloc_root;
4924         int ret;
4925
4926         if (root->reloc_root) {
4927                 reloc_root = root->reloc_root;
4928                 root->reloc_root = NULL;
4929                 list_add(&reloc_root->dead_list,
4930                          &root->fs_info->dead_reloc_roots);
4931
4932                 btrfs_set_root_bytenr(&reloc_root->root_item,
4933                                       reloc_root->node->start);
4934                 btrfs_set_root_level(&root->root_item,
4935                                      btrfs_header_level(reloc_root->node));
4936                 memset(&reloc_root->root_item.drop_progress, 0,
4937                         sizeof(struct btrfs_disk_key));
4938                 reloc_root->root_item.drop_level = 0;
4939
4940                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4941                                         &reloc_root->root_key,
4942                                         &reloc_root->root_item);
4943                 BUG_ON(ret);
4944         }
4945         return 0;
4946 }
4947
4948 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4949 {
4950         struct btrfs_trans_handle *trans;
4951         struct btrfs_root *reloc_root;
4952         struct btrfs_root *prev_root = NULL;
4953         struct list_head dead_roots;
4954         int ret;
4955         unsigned long nr;
4956
4957         INIT_LIST_HEAD(&dead_roots);
4958         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4959
4960         while (!list_empty(&dead_roots)) {
4961                 reloc_root = list_entry(dead_roots.prev,
4962                                         struct btrfs_root, dead_list);
4963                 list_del_init(&reloc_root->dead_list);
4964
4965                 BUG_ON(reloc_root->commit_root != NULL);
4966                 while (1) {
4967                         trans = btrfs_join_transaction(root, 1);
4968                         BUG_ON(!trans);
4969
4970                         mutex_lock(&root->fs_info->drop_mutex);
4971                         ret = btrfs_drop_snapshot(trans, reloc_root);
4972                         if (ret != -EAGAIN)
4973                                 break;
4974                         mutex_unlock(&root->fs_info->drop_mutex);
4975
4976                         nr = trans->blocks_used;
4977                         ret = btrfs_end_transaction(trans, root);
4978                         BUG_ON(ret);
4979                         btrfs_btree_balance_dirty(root, nr);
4980                 }
4981
4982                 free_extent_buffer(reloc_root->node);
4983
4984                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4985                                      &reloc_root->root_key);
4986                 BUG_ON(ret);
4987                 mutex_unlock(&root->fs_info->drop_mutex);
4988
4989                 nr = trans->blocks_used;
4990                 ret = btrfs_end_transaction(trans, root);
4991                 BUG_ON(ret);
4992                 btrfs_btree_balance_dirty(root, nr);
4993
4994                 kfree(prev_root);
4995                 prev_root = reloc_root;
4996         }
4997         if (prev_root) {
4998                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4999                 kfree(prev_root);
5000         }
5001         return 0;
5002 }
5003
5004 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5005 {
5006         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5007         return 0;
5008 }
5009
5010 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5011 {
5012         struct btrfs_root *reloc_root;
5013         struct btrfs_trans_handle *trans;
5014         struct btrfs_key location;
5015         int found;
5016         int ret;
5017
5018         mutex_lock(&root->fs_info->tree_reloc_mutex);
5019         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5020         BUG_ON(ret);
5021         found = !list_empty(&root->fs_info->dead_reloc_roots);
5022         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5023
5024         if (found) {
5025                 trans = btrfs_start_transaction(root, 1);
5026                 BUG_ON(!trans);
5027                 ret = btrfs_commit_transaction(trans, root);
5028                 BUG_ON(ret);
5029         }
5030
5031         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5032         location.offset = (u64)-1;
5033         location.type = BTRFS_ROOT_ITEM_KEY;
5034
5035         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5036         BUG_ON(!reloc_root);
5037         btrfs_orphan_cleanup(reloc_root);
5038         return 0;
5039 }
5040
5041 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
5042                                     struct btrfs_root *root)
5043 {
5044         struct btrfs_root *reloc_root;
5045         struct extent_buffer *eb;
5046         struct btrfs_root_item *root_item;
5047         struct btrfs_key root_key;
5048         int ret;
5049
5050         BUG_ON(!root->ref_cows);
5051         if (root->reloc_root)
5052                 return 0;
5053
5054         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5055         BUG_ON(!root_item);
5056
5057         ret = btrfs_copy_root(trans, root, root->commit_root,
5058                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5059         BUG_ON(ret);
5060
5061         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5062         root_key.offset = root->root_key.objectid;
5063         root_key.type = BTRFS_ROOT_ITEM_KEY;
5064
5065         memcpy(root_item, &root->root_item, sizeof(root_item));
5066         btrfs_set_root_refs(root_item, 0);
5067         btrfs_set_root_bytenr(root_item, eb->start);
5068         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5069         btrfs_set_root_generation(root_item, trans->transid);
5070
5071         btrfs_tree_unlock(eb);
5072         free_extent_buffer(eb);
5073
5074         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5075                                 &root_key, root_item);
5076         BUG_ON(ret);
5077         kfree(root_item);
5078
5079         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5080                                                  &root_key);
5081         BUG_ON(!reloc_root);
5082         reloc_root->last_trans = trans->transid;
5083         reloc_root->commit_root = NULL;
5084         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5085
5086         root->reloc_root = reloc_root;
5087         return 0;
5088 }
5089
5090 /*
5091  * Core function of space balance.
5092  *
5093  * The idea is using reloc trees to relocate tree blocks in reference
5094  * counted roots. There is one reloc tree for each subvol, and all
5095  * reloc trees share same root key objectid. Reloc trees are snapshots
5096  * of the latest committed roots of subvols (root->commit_root).
5097  *
5098  * To relocate a tree block referenced by a subvol, there are two steps.
5099  * COW the block through subvol's reloc tree, then update block pointer
5100  * in the subvol to point to the new block. Since all reloc trees share
5101  * same root key objectid, doing special handing for tree blocks owned
5102  * by them is easy. Once a tree block has been COWed in one reloc tree,
5103  * we can use the resulting new block directly when the same block is
5104  * required to COW again through other reloc trees. By this way, relocated
5105  * tree blocks are shared between reloc trees, so they are also shared
5106  * between subvols.
5107  */
5108 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
5109                                       struct btrfs_root *root,
5110                                       struct btrfs_path *path,
5111                                       struct btrfs_key *first_key,
5112                                       struct btrfs_ref_path *ref_path,
5113                                       struct btrfs_block_group_cache *group,
5114                                       struct inode *reloc_inode)
5115 {
5116         struct btrfs_root *reloc_root;
5117         struct extent_buffer *eb = NULL;
5118         struct btrfs_key *keys;
5119         u64 *nodes;
5120         int level;
5121         int shared_level;
5122         int lowest_level = 0;
5123         int ret;
5124
5125         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5126                 lowest_level = ref_path->owner_objectid;
5127
5128         if (!root->ref_cows) {
5129                 path->lowest_level = lowest_level;
5130                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5131                 BUG_ON(ret < 0);
5132                 path->lowest_level = 0;
5133                 btrfs_release_path(root, path);
5134                 return 0;
5135         }
5136
5137         mutex_lock(&root->fs_info->tree_reloc_mutex);
5138         ret = init_reloc_tree(trans, root);
5139         BUG_ON(ret);
5140         reloc_root = root->reloc_root;
5141
5142         shared_level = ref_path->shared_level;
5143         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5144
5145         keys = ref_path->node_keys;
5146         nodes = ref_path->new_nodes;
5147         memset(&keys[shared_level + 1], 0,
5148                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5149         memset(&nodes[shared_level + 1], 0,
5150                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5151
5152         if (nodes[lowest_level] == 0) {
5153                 path->lowest_level = lowest_level;
5154                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5155                                         0, 1);
5156                 BUG_ON(ret);
5157                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5158                         eb = path->nodes[level];
5159                         if (!eb || eb == reloc_root->node)
5160                                 break;
5161                         nodes[level] = eb->start;
5162                         if (level == 0)
5163                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5164                         else
5165                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5166                 }
5167                 if (nodes[0] &&
5168                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5169                         eb = path->nodes[0];
5170                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5171                                                       group, reloc_inode);
5172                         BUG_ON(ret);
5173                 }
5174                 btrfs_release_path(reloc_root, path);
5175         } else {
5176                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5177                                        lowest_level);
5178                 BUG_ON(ret);
5179         }
5180
5181         /*
5182          * replace tree blocks in the fs tree with tree blocks in
5183          * the reloc tree.
5184          */
5185         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5186         BUG_ON(ret < 0);
5187
5188         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5189                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5190                                         0, 0);
5191                 BUG_ON(ret);
5192                 extent_buffer_get(path->nodes[0]);
5193                 eb = path->nodes[0];
5194                 btrfs_release_path(reloc_root, path);
5195                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5196                 BUG_ON(ret);
5197                 free_extent_buffer(eb);
5198         }
5199
5200         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5201         path->lowest_level = 0;
5202         return 0;
5203 }
5204
5205 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
5206                                         struct btrfs_root *root,
5207                                         struct btrfs_path *path,
5208                                         struct btrfs_key *first_key,
5209                                         struct btrfs_ref_path *ref_path)
5210 {
5211         int ret;
5212
5213         ret = relocate_one_path(trans, root, path, first_key,
5214                                 ref_path, NULL, NULL);
5215         BUG_ON(ret);
5216
5217         if (root == root->fs_info->extent_root)
5218                 btrfs_extent_post_op(trans, root);
5219
5220         return 0;
5221 }
5222
5223 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
5224                                     struct btrfs_root *extent_root,
5225                                     struct btrfs_path *path,
5226                                     struct btrfs_key *extent_key)
5227 {
5228         int ret;
5229
5230         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5231         if (ret)
5232                 goto out;
5233         ret = btrfs_del_item(trans, extent_root, path);
5234 out:
5235         btrfs_release_path(extent_root, path);
5236         return ret;
5237 }
5238
5239 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
5240                                                 struct btrfs_ref_path *ref_path)
5241 {
5242         struct btrfs_key root_key;
5243
5244         root_key.objectid = ref_path->root_objectid;
5245         root_key.type = BTRFS_ROOT_ITEM_KEY;
5246         if (is_cowonly_root(ref_path->root_objectid))
5247                 root_key.offset = 0;
5248         else
5249                 root_key.offset = (u64)-1;
5250
5251         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5252 }
5253
5254 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
5255                                         struct btrfs_path *path,
5256                                         struct btrfs_key *extent_key,
5257                                         struct btrfs_block_group_cache *group,
5258                                         struct inode *reloc_inode, int pass)
5259 {
5260         struct btrfs_trans_handle *trans;
5261         struct btrfs_root *found_root;
5262         struct btrfs_ref_path *ref_path = NULL;
5263         struct disk_extent *new_extents = NULL;
5264         int nr_extents = 0;
5265         int loops;
5266         int ret;
5267         int level;
5268         struct btrfs_key first_key;
5269         u64 prev_block = 0;
5270
5271
5272         trans = btrfs_start_transaction(extent_root, 1);
5273         BUG_ON(!trans);
5274
5275         if (extent_key->objectid == 0) {
5276                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5277                 goto out;
5278         }
5279
5280         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5281         if (!ref_path) {
5282                ret = -ENOMEM;
5283                goto out;
5284         }
5285
5286         for (loops = 0; ; loops++) {
5287                 if (loops == 0) {
5288                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5289                                                    extent_key->objectid);
5290                 } else {
5291                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5292                 }
5293                 if (ret < 0)
5294                         goto out;
5295                 if (ret > 0)
5296                         break;
5297
5298                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5299                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5300                         continue;
5301
5302                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5303                 BUG_ON(!found_root);
5304                 /*
5305                  * for reference counted tree, only process reference paths
5306                  * rooted at the latest committed root.
5307                  */
5308                 if (found_root->ref_cows &&
5309                     ref_path->root_generation != found_root->root_key.offset)
5310                         continue;
5311
5312                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5313                         if (pass == 0) {
5314                                 /*
5315                                  * copy data extents to new locations
5316                                  */
5317                                 u64 group_start = group->key.objectid;
5318                                 ret = relocate_data_extent(reloc_inode,
5319                                                            extent_key,
5320                                                            group_start);
5321                                 if (ret < 0)
5322                                         goto out;
5323                                 break;
5324                         }
5325                         level = 0;
5326                 } else {
5327                         level = ref_path->owner_objectid;
5328                 }
5329
5330                 if (prev_block != ref_path->nodes[level]) {
5331                         struct extent_buffer *eb;
5332                         u64 block_start = ref_path->nodes[level];
5333                         u64 block_size = btrfs_level_size(found_root, level);
5334
5335                         eb = read_tree_block(found_root, block_start,
5336                                              block_size, 0);
5337                         btrfs_tree_lock(eb);
5338                         BUG_ON(level != btrfs_header_level(eb));
5339
5340                         if (level == 0)
5341                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5342                         else
5343                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5344
5345                         btrfs_tree_unlock(eb);
5346                         free_extent_buffer(eb);
5347                         prev_block = block_start;
5348                 }
5349
5350                 btrfs_record_root_in_trans(found_root);
5351                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5352                         /*
5353                          * try to update data extent references while
5354                          * keeping metadata shared between snapshots.
5355                          */
5356                         if (pass == 1) {
5357                                 ret = relocate_one_path(trans, found_root,
5358                                                 path, &first_key, ref_path,
5359                                                 group, reloc_inode);
5360                                 if (ret < 0)
5361                                         goto out;
5362                                 continue;
5363                         }
5364                         /*
5365                          * use fallback method to process the remaining
5366                          * references.
5367                          */
5368                         if (!new_extents) {
5369                                 u64 group_start = group->key.objectid;
5370                                 new_extents = kmalloc(sizeof(*new_extents),
5371                                                       GFP_NOFS);
5372                                 nr_extents = 1;
5373                                 ret = get_new_locations(reloc_inode,
5374                                                         extent_key,
5375                                                         group_start, 1,
5376                                                         &new_extents,
5377                                                         &nr_extents);
5378                                 if (ret)
5379                                         goto out;
5380                         }
5381                         ret = replace_one_extent(trans, found_root,
5382                                                 path, extent_key,
5383                                                 &first_key, ref_path,
5384                                                 new_extents, nr_extents);
5385                 } else {
5386                         ret = relocate_tree_block(trans, found_root, path,
5387                                                   &first_key, ref_path);
5388                 }
5389                 if (ret < 0)
5390                         goto out;
5391         }
5392         ret = 0;
5393 out:
5394         btrfs_end_transaction(trans, extent_root);
5395         kfree(new_extents);
5396         kfree(ref_path);
5397         return ret;
5398 }
5399
5400 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5401 {
5402         u64 num_devices;
5403         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5404                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5405
5406         num_devices = root->fs_info->fs_devices->rw_devices;
5407         if (num_devices == 1) {
5408                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5409                 stripped = flags & ~stripped;
5410
5411                 /* turn raid0 into single device chunks */
5412                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5413                         return stripped;
5414
5415                 /* turn mirroring into duplication */
5416                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5417                              BTRFS_BLOCK_GROUP_RAID10))
5418                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5419                 return flags;
5420         } else {
5421                 /* they already had raid on here, just return */
5422                 if (flags & stripped)
5423                         return flags;
5424
5425                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5426                 stripped = flags & ~stripped;
5427
5428                 /* switch duplicated blocks with raid1 */
5429                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5430                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5431
5432                 /* turn single device chunks into raid0 */
5433                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5434         }
5435         return flags;
5436 }
5437
5438 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5439                      struct btrfs_block_group_cache *shrink_block_group,
5440                      int force)
5441 {
5442         struct btrfs_trans_handle *trans;
5443         u64 new_alloc_flags;
5444         u64 calc;
5445
5446         spin_lock(&shrink_block_group->lock);
5447         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5448                 spin_unlock(&shrink_block_group->lock);
5449
5450                 trans = btrfs_start_transaction(root, 1);
5451                 spin_lock(&shrink_block_group->lock);
5452
5453                 new_alloc_flags = update_block_group_flags(root,
5454                                                    shrink_block_group->flags);
5455                 if (new_alloc_flags != shrink_block_group->flags) {
5456                         calc =
5457                              btrfs_block_group_used(&shrink_block_group->item);
5458                 } else {
5459                         calc = shrink_block_group->key.offset;
5460                 }
5461                 spin_unlock(&shrink_block_group->lock);
5462
5463                 do_chunk_alloc(trans, root->fs_info->extent_root,
5464                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5465
5466                 btrfs_end_transaction(trans, root);
5467         } else
5468                 spin_unlock(&shrink_block_group->lock);
5469         return 0;
5470 }
5471
5472 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5473                                  struct btrfs_root *root,
5474                                  u64 objectid, u64 size)
5475 {
5476         struct btrfs_path *path;
5477         struct btrfs_inode_item *item;
5478         struct extent_buffer *leaf;
5479         int ret;
5480
5481         path = btrfs_alloc_path();
5482         if (!path)
5483                 return -ENOMEM;
5484
5485         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5486         if (ret)
5487                 goto out;
5488
5489         leaf = path->nodes[0];
5490         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5491         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5492         btrfs_set_inode_generation(leaf, item, 1);
5493         btrfs_set_inode_size(leaf, item, size);
5494         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5495         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5496         btrfs_mark_buffer_dirty(leaf);
5497         btrfs_release_path(root, path);
5498 out:
5499         btrfs_free_path(path);
5500         return ret;
5501 }
5502
5503 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
5504                                         struct btrfs_block_group_cache *group)
5505 {
5506         struct inode *inode = NULL;
5507         struct btrfs_trans_handle *trans;
5508         struct btrfs_root *root;
5509         struct btrfs_key root_key;
5510         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5511         int err = 0;
5512
5513         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5514         root_key.type = BTRFS_ROOT_ITEM_KEY;
5515         root_key.offset = (u64)-1;
5516         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5517         if (IS_ERR(root))
5518                 return ERR_CAST(root);
5519
5520         trans = btrfs_start_transaction(root, 1);
5521         BUG_ON(!trans);
5522
5523         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5524         if (err)
5525                 goto out;
5526
5527         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5528         BUG_ON(err);
5529
5530         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5531                                        group->key.offset, 0, group->key.offset,
5532                                        0, 0, 0);
5533         BUG_ON(err);
5534
5535         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5536         if (inode->i_state & I_NEW) {
5537                 BTRFS_I(inode)->root = root;
5538                 BTRFS_I(inode)->location.objectid = objectid;
5539                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5540                 BTRFS_I(inode)->location.offset = 0;
5541                 btrfs_read_locked_inode(inode);
5542                 unlock_new_inode(inode);
5543                 BUG_ON(is_bad_inode(inode));
5544         } else {
5545                 BUG_ON(1);
5546         }
5547         BTRFS_I(inode)->index_cnt = group->key.objectid;
5548
5549         err = btrfs_orphan_add(trans, inode);
5550 out:
5551         btrfs_end_transaction(trans, root);
5552         if (err) {
5553                 if (inode)
5554                         iput(inode);
5555                 inode = ERR_PTR(err);
5556         }
5557         return inode;
5558 }
5559
5560 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5561 {
5562
5563         struct btrfs_ordered_sum *sums;
5564         struct btrfs_sector_sum *sector_sum;
5565         struct btrfs_ordered_extent *ordered;
5566         struct btrfs_root *root = BTRFS_I(inode)->root;
5567         struct list_head list;
5568         size_t offset;
5569         int ret;
5570         u64 disk_bytenr;
5571
5572         INIT_LIST_HEAD(&list);
5573
5574         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5575         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5576
5577         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5578         ret = btrfs_lookup_csums_range(root, disk_bytenr,
5579                                        disk_bytenr + len - 1, &list);
5580
5581         while (!list_empty(&list)) {
5582                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5583                 list_del_init(&sums->list);
5584
5585                 sector_sum = sums->sums;
5586                 sums->bytenr = ordered->start;
5587
5588                 offset = 0;
5589                 while (offset < sums->len) {
5590                         sector_sum->bytenr += ordered->start - disk_bytenr;
5591                         sector_sum++;
5592                         offset += root->sectorsize;
5593                 }
5594
5595                 btrfs_add_ordered_sum(inode, ordered, sums);
5596         }
5597         btrfs_put_ordered_extent(ordered);
5598         return 0;
5599 }
5600
5601 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5602 {
5603         struct btrfs_trans_handle *trans;
5604         struct btrfs_path *path;
5605         struct btrfs_fs_info *info = root->fs_info;
5606         struct extent_buffer *leaf;
5607         struct inode *reloc_inode;
5608         struct btrfs_block_group_cache *block_group;
5609         struct btrfs_key key;
5610         u64 skipped;
5611         u64 cur_byte;
5612         u64 total_found;
5613         u32 nritems;
5614         int ret;
5615         int progress;
5616         int pass = 0;
5617
5618         root = root->fs_info->extent_root;
5619
5620         block_group = btrfs_lookup_block_group(info, group_start);
5621         BUG_ON(!block_group);
5622
5623         printk("btrfs relocating block group %llu flags %llu\n",
5624                (unsigned long long)block_group->key.objectid,
5625                (unsigned long long)block_group->flags);
5626
5627         path = btrfs_alloc_path();
5628         BUG_ON(!path);
5629
5630         reloc_inode = create_reloc_inode(info, block_group);
5631         BUG_ON(IS_ERR(reloc_inode));
5632
5633         __alloc_chunk_for_shrink(root, block_group, 1);
5634         set_block_group_readonly(block_group);
5635
5636         btrfs_start_delalloc_inodes(info->tree_root);
5637         btrfs_wait_ordered_extents(info->tree_root, 0);
5638 again:
5639         skipped = 0;
5640         total_found = 0;
5641         progress = 0;
5642         key.objectid = block_group->key.objectid;
5643         key.offset = 0;
5644         key.type = 0;
5645         cur_byte = key.objectid;
5646
5647         trans = btrfs_start_transaction(info->tree_root, 1);
5648         btrfs_commit_transaction(trans, info->tree_root);
5649
5650         mutex_lock(&root->fs_info->cleaner_mutex);
5651         btrfs_clean_old_snapshots(info->tree_root);
5652         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5653         mutex_unlock(&root->fs_info->cleaner_mutex);
5654
5655         while(1) {
5656                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5657                 if (ret < 0)
5658                         goto out;
5659 next:
5660                 leaf = path->nodes[0];
5661                 nritems = btrfs_header_nritems(leaf);
5662                 if (path->slots[0] >= nritems) {
5663                         ret = btrfs_next_leaf(root, path);
5664                         if (ret < 0)
5665                                 goto out;
5666                         if (ret == 1) {
5667                                 ret = 0;
5668                                 break;
5669                         }
5670                         leaf = path->nodes[0];
5671                         nritems = btrfs_header_nritems(leaf);
5672                 }
5673
5674                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5675
5676                 if (key.objectid >= block_group->key.objectid +
5677                     block_group->key.offset)
5678                         break;
5679
5680                 if (progress && need_resched()) {
5681                         btrfs_release_path(root, path);
5682                         cond_resched();
5683                         progress = 0;
5684                         continue;
5685                 }
5686                 progress = 1;
5687
5688                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5689                     key.objectid + key.offset <= cur_byte) {
5690                         path->slots[0]++;
5691                         goto next;
5692                 }
5693
5694                 total_found++;
5695                 cur_byte = key.objectid + key.offset;
5696                 btrfs_release_path(root, path);
5697
5698                 __alloc_chunk_for_shrink(root, block_group, 0);
5699                 ret = relocate_one_extent(root, path, &key, block_group,
5700                                           reloc_inode, pass);
5701                 BUG_ON(ret < 0);
5702                 if (ret > 0)
5703                         skipped++;
5704
5705                 key.objectid = cur_byte;
5706                 key.type = 0;
5707                 key.offset = 0;
5708         }
5709
5710         btrfs_release_path(root, path);
5711
5712         if (pass == 0) {
5713                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5714                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5715         }
5716
5717         if (total_found > 0) {
5718                 printk("btrfs found %llu extents in pass %d\n",
5719                        (unsigned long long)total_found, pass);
5720                 pass++;
5721                 if (total_found == skipped && pass > 2) {
5722                         iput(reloc_inode);
5723                         reloc_inode = create_reloc_inode(info, block_group);
5724                         pass = 0;
5725                 }
5726                 goto again;
5727         }
5728
5729         /* delete reloc_inode */
5730         iput(reloc_inode);
5731
5732         /* unpin extents in this range */
5733         trans = btrfs_start_transaction(info->tree_root, 1);
5734         btrfs_commit_transaction(trans, info->tree_root);
5735
5736         spin_lock(&block_group->lock);
5737         WARN_ON(block_group->pinned > 0);
5738         WARN_ON(block_group->reserved > 0);
5739         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5740         spin_unlock(&block_group->lock);
5741         put_block_group(block_group);
5742         ret = 0;
5743 out:
5744         btrfs_free_path(path);
5745         return ret;
5746 }
5747
5748 static int find_first_block_group(struct btrfs_root *root,
5749                 struct btrfs_path *path, struct btrfs_key *key)
5750 {
5751         int ret = 0;
5752         struct btrfs_key found_key;
5753         struct extent_buffer *leaf;
5754         int slot;
5755
5756         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5757         if (ret < 0)
5758                 goto out;
5759
5760         while(1) {
5761                 slot = path->slots[0];
5762                 leaf = path->nodes[0];
5763                 if (slot >= btrfs_header_nritems(leaf)) {
5764                         ret = btrfs_next_leaf(root, path);
5765                         if (ret == 0)
5766                                 continue;
5767                         if (ret < 0)
5768                                 goto out;
5769                         break;
5770                 }
5771                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5772
5773                 if (found_key.objectid >= key->objectid &&
5774                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5775                         ret = 0;
5776                         goto out;
5777                 }
5778                 path->slots[0]++;
5779         }
5780         ret = -ENOENT;
5781 out:
5782         return ret;
5783 }
5784
5785 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5786 {
5787         struct btrfs_block_group_cache *block_group;
5788         struct rb_node *n;
5789
5790         spin_lock(&info->block_group_cache_lock);
5791         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5792                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5793                                        cache_node);
5794                 rb_erase(&block_group->cache_node,
5795                          &info->block_group_cache_tree);
5796                 spin_unlock(&info->block_group_cache_lock);
5797
5798                 btrfs_remove_free_space_cache(block_group);
5799                 down_write(&block_group->space_info->groups_sem);
5800                 list_del(&block_group->list);
5801                 up_write(&block_group->space_info->groups_sem);
5802
5803                 WARN_ON(atomic_read(&block_group->count) != 1);
5804                 kfree(block_group);
5805
5806                 spin_lock(&info->block_group_cache_lock);
5807         }
5808         spin_unlock(&info->block_group_cache_lock);
5809         return 0;
5810 }
5811
5812 int btrfs_read_block_groups(struct btrfs_root *root)
5813 {
5814         struct btrfs_path *path;
5815         int ret;
5816         struct btrfs_block_group_cache *cache;
5817         struct btrfs_fs_info *info = root->fs_info;
5818         struct btrfs_space_info *space_info;
5819         struct btrfs_key key;
5820         struct btrfs_key found_key;
5821         struct extent_buffer *leaf;
5822
5823         root = info->extent_root;
5824         key.objectid = 0;
5825         key.offset = 0;
5826         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5827         path = btrfs_alloc_path();
5828         if (!path)
5829                 return -ENOMEM;
5830
5831         while(1) {
5832                 ret = find_first_block_group(root, path, &key);
5833                 if (ret > 0) {
5834                         ret = 0;
5835                         goto error;
5836                 }
5837                 if (ret != 0)
5838                         goto error;
5839
5840                 leaf = path->nodes[0];
5841                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5842                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5843                 if (!cache) {
5844                         ret = -ENOMEM;
5845                         break;
5846                 }
5847
5848                 atomic_set(&cache->count, 1);
5849                 spin_lock_init(&cache->lock);
5850                 mutex_init(&cache->alloc_mutex);
5851                 mutex_init(&cache->cache_mutex);
5852                 INIT_LIST_HEAD(&cache->list);
5853                 read_extent_buffer(leaf, &cache->item,
5854                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5855                                    sizeof(cache->item));
5856                 memcpy(&cache->key, &found_key, sizeof(found_key));
5857
5858                 key.objectid = found_key.objectid + found_key.offset;
5859                 btrfs_release_path(root, path);
5860                 cache->flags = btrfs_block_group_flags(&cache->item);
5861
5862                 ret = update_space_info(info, cache->flags, found_key.offset,
5863                                         btrfs_block_group_used(&cache->item),
5864                                         &space_info);
5865                 BUG_ON(ret);
5866                 cache->space_info = space_info;
5867                 down_write(&space_info->groups_sem);
5868                 list_add_tail(&cache->list, &space_info->block_groups);
5869                 up_write(&space_info->groups_sem);
5870
5871                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5872                 BUG_ON(ret);
5873
5874                 set_avail_alloc_bits(root->fs_info, cache->flags);
5875                 if (btrfs_chunk_readonly(root, cache->key.objectid))
5876                         set_block_group_readonly(cache);
5877         }
5878         ret = 0;
5879 error:
5880         btrfs_free_path(path);
5881         return ret;
5882 }
5883
5884 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5885                            struct btrfs_root *root, u64 bytes_used,
5886                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5887                            u64 size)
5888 {
5889         int ret;
5890         struct btrfs_root *extent_root;
5891         struct btrfs_block_group_cache *cache;
5892
5893         extent_root = root->fs_info->extent_root;
5894
5895         root->fs_info->last_trans_new_blockgroup = trans->transid;
5896
5897         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5898         if (!cache)
5899                 return -ENOMEM;
5900
5901         cache->key.objectid = chunk_offset;
5902         cache->key.offset = size;
5903         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5904         atomic_set(&cache->count, 1);
5905         spin_lock_init(&cache->lock);
5906         mutex_init(&cache->alloc_mutex);
5907         mutex_init(&cache->cache_mutex);
5908         INIT_LIST_HEAD(&cache->list);
5909
5910         btrfs_set_block_group_used(&cache->item, bytes_used);
5911         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5912         cache->flags = type;
5913         btrfs_set_block_group_flags(&cache->item, type);
5914
5915         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5916                                 &cache->space_info);
5917         BUG_ON(ret);
5918         down_write(&cache->space_info->groups_sem);
5919         list_add_tail(&cache->list, &cache->space_info->block_groups);
5920         up_write(&cache->space_info->groups_sem);
5921
5922         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5923         BUG_ON(ret);
5924
5925         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5926                                 sizeof(cache->item));
5927         BUG_ON(ret);
5928
5929         finish_current_insert(trans, extent_root, 0);
5930         ret = del_pending_extents(trans, extent_root, 0);
5931         BUG_ON(ret);
5932         set_avail_alloc_bits(extent_root->fs_info, type);
5933
5934         return 0;
5935 }
5936
5937 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5938                              struct btrfs_root *root, u64 group_start)
5939 {
5940         struct btrfs_path *path;
5941         struct btrfs_block_group_cache *block_group;
5942         struct btrfs_key key;
5943         int ret;
5944
5945         root = root->fs_info->extent_root;
5946
5947         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5948         BUG_ON(!block_group);
5949         BUG_ON(!block_group->ro);
5950
5951         memcpy(&key, &block_group->key, sizeof(key));
5952
5953         path = btrfs_alloc_path();
5954         BUG_ON(!path);
5955
5956         btrfs_remove_free_space_cache(block_group);
5957         rb_erase(&block_group->cache_node,
5958                  &root->fs_info->block_group_cache_tree);
5959         down_write(&block_group->space_info->groups_sem);
5960         list_del(&block_group->list);
5961         up_write(&block_group->space_info->groups_sem);
5962
5963         spin_lock(&block_group->space_info->lock);
5964         block_group->space_info->total_bytes -= block_group->key.offset;
5965         block_group->space_info->bytes_readonly -= block_group->key.offset;
5966         spin_unlock(&block_group->space_info->lock);
5967         block_group->space_info->full = 0;
5968
5969         put_block_group(block_group);
5970         put_block_group(block_group);
5971
5972         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5973         if (ret > 0)
5974                 ret = -EIO;
5975         if (ret < 0)
5976                 goto out;
5977
5978         ret = btrfs_del_item(trans, root, path);
5979 out:
5980         btrfs_free_path(path);
5981         return ret;
5982 }