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_irq(&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                 spin_unlock_irq(&info->delalloc_lock);
1055
1056                 root_used = btrfs_root_used(&extent_root->root_item);
1057                 btrfs_set_root_used(&extent_root->root_item,
1058                                     root_used - bytes_freed);
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 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                         ret = 1;
1424                         goto out;
1425                 }
1426                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1427                         ret = 1;
1428                         goto out;
1429                 }
1430
1431                 path->slots[0]++;
1432         }
1433         ret = 0;
1434 out:
1435         btrfs_free_path(path);
1436         return ret;
1437 }
1438
1439 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1440                     struct extent_buffer *buf, u32 nr_extents)
1441 {
1442         struct btrfs_key key;
1443         struct btrfs_file_extent_item *fi;
1444         u64 root_gen;
1445         u32 nritems;
1446         int i;
1447         int level;
1448         int ret = 0;
1449         int shared = 0;
1450
1451         if (!root->ref_cows)
1452                 return 0;
1453
1454         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1455                 shared = 0;
1456                 root_gen = root->root_key.offset;
1457         } else {
1458                 shared = 1;
1459                 root_gen = trans->transid - 1;
1460         }
1461
1462         level = btrfs_header_level(buf);
1463         nritems = btrfs_header_nritems(buf);
1464
1465         if (level == 0) {
1466                 struct btrfs_leaf_ref *ref;
1467                 struct btrfs_extent_info *info;
1468
1469                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1470                 if (!ref) {
1471                         ret = -ENOMEM;
1472                         goto out;
1473                 }
1474
1475                 ref->root_gen = root_gen;
1476                 ref->bytenr = buf->start;
1477                 ref->owner = btrfs_header_owner(buf);
1478                 ref->generation = btrfs_header_generation(buf);
1479                 ref->nritems = nr_extents;
1480                 info = ref->extents;
1481
1482                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1483                         u64 disk_bytenr;
1484                         btrfs_item_key_to_cpu(buf, &key, i);
1485                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1486                                 continue;
1487                         fi = btrfs_item_ptr(buf, i,
1488                                             struct btrfs_file_extent_item);
1489                         if (btrfs_file_extent_type(buf, fi) ==
1490                             BTRFS_FILE_EXTENT_INLINE)
1491                                 continue;
1492                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1493                         if (disk_bytenr == 0)
1494                                 continue;
1495
1496                         info->bytenr = disk_bytenr;
1497                         info->num_bytes =
1498                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1499                         info->objectid = key.objectid;
1500                         info->offset = key.offset;
1501                         info++;
1502                 }
1503
1504                 ret = btrfs_add_leaf_ref(root, ref, shared);
1505                 if (ret == -EEXIST && shared) {
1506                         struct btrfs_leaf_ref *old;
1507                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1508                         BUG_ON(!old);
1509                         btrfs_remove_leaf_ref(root, old);
1510                         btrfs_free_leaf_ref(root, old);
1511                         ret = btrfs_add_leaf_ref(root, ref, shared);
1512                 }
1513                 WARN_ON(ret);
1514                 btrfs_free_leaf_ref(root, ref);
1515         }
1516 out:
1517         return ret;
1518 }
1519
1520 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1521                   struct extent_buffer *orig_buf, struct extent_buffer *buf,
1522                   u32 *nr_extents)
1523 {
1524         u64 bytenr;
1525         u64 ref_root;
1526         u64 orig_root;
1527         u64 ref_generation;
1528         u64 orig_generation;
1529         u32 nritems;
1530         u32 nr_file_extents = 0;
1531         struct btrfs_key key;
1532         struct btrfs_file_extent_item *fi;
1533         int i;
1534         int level;
1535         int ret = 0;
1536         int faili = 0;
1537         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1538                             u64, u64, u64, u64, u64, u64, u64, u64);
1539
1540         ref_root = btrfs_header_owner(buf);
1541         ref_generation = btrfs_header_generation(buf);
1542         orig_root = btrfs_header_owner(orig_buf);
1543         orig_generation = btrfs_header_generation(orig_buf);
1544
1545         nritems = btrfs_header_nritems(buf);
1546         level = btrfs_header_level(buf);
1547
1548         if (root->ref_cows) {
1549                 process_func = __btrfs_inc_extent_ref;
1550         } else {
1551                 if (level == 0 &&
1552                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1553                         goto out;
1554                 if (level != 0 &&
1555                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1556                         goto out;
1557                 process_func = __btrfs_update_extent_ref;
1558         }
1559
1560         for (i = 0; i < nritems; i++) {
1561                 cond_resched();
1562                 if (level == 0) {
1563                         btrfs_item_key_to_cpu(buf, &key, i);
1564                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1565                                 continue;
1566                         fi = btrfs_item_ptr(buf, i,
1567                                             struct btrfs_file_extent_item);
1568                         if (btrfs_file_extent_type(buf, fi) ==
1569                             BTRFS_FILE_EXTENT_INLINE)
1570                                 continue;
1571                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1572                         if (bytenr == 0)
1573                                 continue;
1574
1575                         nr_file_extents++;
1576
1577                         ret = process_func(trans, root, bytenr,
1578                                            orig_buf->start, buf->start,
1579                                            orig_root, ref_root,
1580                                            orig_generation, ref_generation,
1581                                            key.objectid);
1582
1583                         if (ret) {
1584                                 faili = i;
1585                                 WARN_ON(1);
1586                                 goto fail;
1587                         }
1588                 } else {
1589                         bytenr = btrfs_node_blockptr(buf, i);
1590                         ret = process_func(trans, root, bytenr,
1591                                            orig_buf->start, buf->start,
1592                                            orig_root, ref_root,
1593                                            orig_generation, ref_generation,
1594                                            level - 1);
1595                         if (ret) {
1596                                 faili = i;
1597                                 WARN_ON(1);
1598                                 goto fail;
1599                         }
1600                 }
1601         }
1602 out:
1603         if (nr_extents) {
1604                 if (level == 0)
1605                         *nr_extents = nr_file_extents;
1606                 else
1607                         *nr_extents = nritems;
1608         }
1609         return 0;
1610 fail:
1611         WARN_ON(1);
1612         return ret;
1613 }
1614
1615 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1616                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1617                      struct extent_buffer *buf, int start_slot, int nr)
1618
1619 {
1620         u64 bytenr;
1621         u64 ref_root;
1622         u64 orig_root;
1623         u64 ref_generation;
1624         u64 orig_generation;
1625         struct btrfs_key key;
1626         struct btrfs_file_extent_item *fi;
1627         int i;
1628         int ret;
1629         int slot;
1630         int level;
1631
1632         BUG_ON(start_slot < 0);
1633         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1634
1635         ref_root = btrfs_header_owner(buf);
1636         ref_generation = btrfs_header_generation(buf);
1637         orig_root = btrfs_header_owner(orig_buf);
1638         orig_generation = btrfs_header_generation(orig_buf);
1639         level = btrfs_header_level(buf);
1640
1641         if (!root->ref_cows) {
1642                 if (level == 0 &&
1643                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1644                         return 0;
1645                 if (level != 0 &&
1646                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1647                         return 0;
1648         }
1649
1650         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1651                 cond_resched();
1652                 if (level == 0) {
1653                         btrfs_item_key_to_cpu(buf, &key, slot);
1654                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1655                                 continue;
1656                         fi = btrfs_item_ptr(buf, slot,
1657                                             struct btrfs_file_extent_item);
1658                         if (btrfs_file_extent_type(buf, fi) ==
1659                             BTRFS_FILE_EXTENT_INLINE)
1660                                 continue;
1661                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1662                         if (bytenr == 0)
1663                                 continue;
1664                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1665                                             orig_buf->start, buf->start,
1666                                             orig_root, ref_root,
1667                                             orig_generation, ref_generation,
1668                                             key.objectid);
1669                         if (ret)
1670                                 goto fail;
1671                 } else {
1672                         bytenr = btrfs_node_blockptr(buf, slot);
1673                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1674                                             orig_buf->start, buf->start,
1675                                             orig_root, ref_root,
1676                                             orig_generation, ref_generation,
1677                                             level - 1);
1678                         if (ret)
1679                                 goto fail;
1680                 }
1681         }
1682         return 0;
1683 fail:
1684         WARN_ON(1);
1685         return -1;
1686 }
1687
1688 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1689                                  struct btrfs_root *root,
1690                                  struct btrfs_path *path,
1691                                  struct btrfs_block_group_cache *cache)
1692 {
1693         int ret;
1694         int pending_ret;
1695         struct btrfs_root *extent_root = root->fs_info->extent_root;
1696         unsigned long bi;
1697         struct extent_buffer *leaf;
1698
1699         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1700         if (ret < 0)
1701                 goto fail;
1702         BUG_ON(ret);
1703
1704         leaf = path->nodes[0];
1705         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1706         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1707         btrfs_mark_buffer_dirty(leaf);
1708         btrfs_release_path(extent_root, path);
1709 fail:
1710         finish_current_insert(trans, extent_root, 0);
1711         pending_ret = del_pending_extents(trans, extent_root, 0);
1712         if (ret)
1713                 return ret;
1714         if (pending_ret)
1715                 return pending_ret;
1716         return 0;
1717
1718 }
1719
1720 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1721                                    struct btrfs_root *root)
1722 {
1723         struct btrfs_block_group_cache *cache, *entry;
1724         struct rb_node *n;
1725         int err = 0;
1726         int werr = 0;
1727         struct btrfs_path *path;
1728         u64 last = 0;
1729
1730         path = btrfs_alloc_path();
1731         if (!path)
1732                 return -ENOMEM;
1733
1734         while(1) {
1735                 cache = NULL;
1736                 spin_lock(&root->fs_info->block_group_cache_lock);
1737                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1738                      n; n = rb_next(n)) {
1739                         entry = rb_entry(n, struct btrfs_block_group_cache,
1740                                          cache_node);
1741                         if (entry->dirty) {
1742                                 cache = entry;
1743                                 break;
1744                         }
1745                 }
1746                 spin_unlock(&root->fs_info->block_group_cache_lock);
1747
1748                 if (!cache)
1749                         break;
1750
1751                 cache->dirty = 0;
1752                 last += cache->key.offset;
1753
1754                 err = write_one_cache_group(trans, root,
1755                                             path, cache);
1756                 /*
1757                  * if we fail to write the cache group, we want
1758                  * to keep it marked dirty in hopes that a later
1759                  * write will work
1760                  */
1761                 if (err) {
1762                         werr = err;
1763                         continue;
1764                 }
1765         }
1766         btrfs_free_path(path);
1767         return werr;
1768 }
1769
1770 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1771 {
1772         struct btrfs_block_group_cache *block_group;
1773         int readonly = 0;
1774
1775         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1776         if (!block_group || block_group->ro)
1777                 readonly = 1;
1778         if (block_group)
1779                 put_block_group(block_group);
1780         return readonly;
1781 }
1782
1783 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1784                              u64 total_bytes, u64 bytes_used,
1785                              struct btrfs_space_info **space_info)
1786 {
1787         struct btrfs_space_info *found;
1788
1789         found = __find_space_info(info, flags);
1790         if (found) {
1791                 spin_lock(&found->lock);
1792                 found->total_bytes += total_bytes;
1793                 found->bytes_used += bytes_used;
1794                 found->full = 0;
1795                 spin_unlock(&found->lock);
1796                 *space_info = found;
1797                 return 0;
1798         }
1799         found = kzalloc(sizeof(*found), GFP_NOFS);
1800         if (!found)
1801                 return -ENOMEM;
1802
1803         list_add(&found->list, &info->space_info);
1804         INIT_LIST_HEAD(&found->block_groups);
1805         init_rwsem(&found->groups_sem);
1806         spin_lock_init(&found->lock);
1807         found->flags = flags;
1808         found->total_bytes = total_bytes;
1809         found->bytes_used = bytes_used;
1810         found->bytes_pinned = 0;
1811         found->bytes_reserved = 0;
1812         found->bytes_readonly = 0;
1813         found->full = 0;
1814         found->force_alloc = 0;
1815         *space_info = found;
1816         return 0;
1817 }
1818
1819 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1820 {
1821         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1822                                    BTRFS_BLOCK_GROUP_RAID1 |
1823                                    BTRFS_BLOCK_GROUP_RAID10 |
1824                                    BTRFS_BLOCK_GROUP_DUP);
1825         if (extra_flags) {
1826                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1827                         fs_info->avail_data_alloc_bits |= extra_flags;
1828                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1829                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1830                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1831                         fs_info->avail_system_alloc_bits |= extra_flags;
1832         }
1833 }
1834
1835 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1836 {
1837         spin_lock(&cache->space_info->lock);
1838         spin_lock(&cache->lock);
1839         if (!cache->ro) {
1840                 cache->space_info->bytes_readonly += cache->key.offset -
1841                                         btrfs_block_group_used(&cache->item);
1842                 cache->ro = 1;
1843         }
1844         spin_unlock(&cache->lock);
1845         spin_unlock(&cache->space_info->lock);
1846 }
1847
1848 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1849 {
1850         u64 num_devices = root->fs_info->fs_devices->rw_devices;
1851
1852         if (num_devices == 1)
1853                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1854         if (num_devices < 4)
1855                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1856
1857         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1858             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1859                       BTRFS_BLOCK_GROUP_RAID10))) {
1860                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1861         }
1862
1863         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1864             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1865                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1866         }
1867
1868         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1869             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1870              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1871              (flags & BTRFS_BLOCK_GROUP_DUP)))
1872                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1873         return flags;
1874 }
1875
1876 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1877                           struct btrfs_root *extent_root, u64 alloc_bytes,
1878                           u64 flags, int force)
1879 {
1880         struct btrfs_space_info *space_info;
1881         u64 thresh;
1882         int ret = 0;
1883
1884         mutex_lock(&extent_root->fs_info->chunk_mutex);
1885
1886         flags = btrfs_reduce_alloc_profile(extent_root, flags);
1887
1888         space_info = __find_space_info(extent_root->fs_info, flags);
1889         if (!space_info) {
1890                 ret = update_space_info(extent_root->fs_info, flags,
1891                                         0, 0, &space_info);
1892                 BUG_ON(ret);
1893         }
1894         BUG_ON(!space_info);
1895
1896         spin_lock(&space_info->lock);
1897         if (space_info->force_alloc) {
1898                 force = 1;
1899                 space_info->force_alloc = 0;
1900         }
1901         if (space_info->full) {
1902                 spin_unlock(&space_info->lock);
1903                 goto out;
1904         }
1905
1906         thresh = space_info->total_bytes - space_info->bytes_readonly;
1907         thresh = div_factor(thresh, 6);
1908         if (!force &&
1909            (space_info->bytes_used + space_info->bytes_pinned +
1910             space_info->bytes_reserved + alloc_bytes) < thresh) {
1911                 spin_unlock(&space_info->lock);
1912                 goto out;
1913         }
1914         spin_unlock(&space_info->lock);
1915
1916         ret = btrfs_alloc_chunk(trans, extent_root, flags);
1917         if (ret) {
1918 printk("space info full %Lu\n", flags);
1919                 space_info->full = 1;
1920         }
1921 out:
1922         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1923         return ret;
1924 }
1925
1926 static int update_block_group(struct btrfs_trans_handle *trans,
1927                               struct btrfs_root *root,
1928                               u64 bytenr, u64 num_bytes, int alloc,
1929                               int mark_free)
1930 {
1931         struct btrfs_block_group_cache *cache;
1932         struct btrfs_fs_info *info = root->fs_info;
1933         u64 total = num_bytes;
1934         u64 old_val;
1935         u64 byte_in_group;
1936
1937         while(total) {
1938                 cache = btrfs_lookup_block_group(info, bytenr);
1939                 if (!cache)
1940                         return -1;
1941                 byte_in_group = bytenr - cache->key.objectid;
1942                 WARN_ON(byte_in_group > cache->key.offset);
1943
1944                 spin_lock(&cache->space_info->lock);
1945                 spin_lock(&cache->lock);
1946                 cache->dirty = 1;
1947                 old_val = btrfs_block_group_used(&cache->item);
1948                 num_bytes = min(total, cache->key.offset - byte_in_group);
1949                 if (alloc) {
1950                         old_val += num_bytes;
1951                         cache->space_info->bytes_used += num_bytes;
1952                         if (cache->ro)
1953                                 cache->space_info->bytes_readonly -= num_bytes;
1954                         btrfs_set_block_group_used(&cache->item, old_val);
1955                         spin_unlock(&cache->lock);
1956                         spin_unlock(&cache->space_info->lock);
1957                 } else {
1958                         old_val -= num_bytes;
1959                         cache->space_info->bytes_used -= num_bytes;
1960                         if (cache->ro)
1961                                 cache->space_info->bytes_readonly += num_bytes;
1962                         btrfs_set_block_group_used(&cache->item, old_val);
1963                         spin_unlock(&cache->lock);
1964                         spin_unlock(&cache->space_info->lock);
1965                         if (mark_free) {
1966                                 int ret;
1967                                 ret = btrfs_add_free_space(cache, bytenr,
1968                                                            num_bytes);
1969                                 WARN_ON(ret);
1970                         }
1971                 }
1972                 put_block_group(cache);
1973                 total -= num_bytes;
1974                 bytenr += num_bytes;
1975         }
1976         return 0;
1977 }
1978
1979 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1980 {
1981         struct btrfs_block_group_cache *cache;
1982         u64 bytenr;
1983
1984         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1985         if (!cache)
1986                 return 0;
1987
1988         bytenr = cache->key.objectid;
1989         put_block_group(cache);
1990
1991         return bytenr;
1992 }
1993
1994 int btrfs_update_pinned_extents(struct btrfs_root *root,
1995                                 u64 bytenr, u64 num, int pin)
1996 {
1997         u64 len;
1998         struct btrfs_block_group_cache *cache;
1999         struct btrfs_fs_info *fs_info = root->fs_info;
2000
2001         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2002         if (pin) {
2003                 set_extent_dirty(&fs_info->pinned_extents,
2004                                 bytenr, bytenr + num - 1, GFP_NOFS);
2005         } else {
2006                 clear_extent_dirty(&fs_info->pinned_extents,
2007                                 bytenr, bytenr + num - 1, GFP_NOFS);
2008         }
2009         while (num > 0) {
2010                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2011                 BUG_ON(!cache);
2012                 len = min(num, cache->key.offset -
2013                           (bytenr - cache->key.objectid));
2014                 if (pin) {
2015                         spin_lock(&cache->space_info->lock);
2016                         spin_lock(&cache->lock);
2017                         cache->pinned += len;
2018                         cache->space_info->bytes_pinned += len;
2019                         spin_unlock(&cache->lock);
2020                         spin_unlock(&cache->space_info->lock);
2021                         fs_info->total_pinned += len;
2022                 } else {
2023                         spin_lock(&cache->space_info->lock);
2024                         spin_lock(&cache->lock);
2025                         cache->pinned -= len;
2026                         cache->space_info->bytes_pinned -= len;
2027                         spin_unlock(&cache->lock);
2028                         spin_unlock(&cache->space_info->lock);
2029                         fs_info->total_pinned -= len;
2030                         if (cache->cached)
2031                                 btrfs_add_free_space(cache, bytenr, len);
2032                 }
2033                 put_block_group(cache);
2034                 bytenr += len;
2035                 num -= len;
2036         }
2037         return 0;
2038 }
2039
2040 static int update_reserved_extents(struct btrfs_root *root,
2041                                    u64 bytenr, u64 num, int reserve)
2042 {
2043         u64 len;
2044         struct btrfs_block_group_cache *cache;
2045         struct btrfs_fs_info *fs_info = root->fs_info;
2046
2047         while (num > 0) {
2048                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2049                 BUG_ON(!cache);
2050                 len = min(num, cache->key.offset -
2051                           (bytenr - cache->key.objectid));
2052
2053                 spin_lock(&cache->space_info->lock);
2054                 spin_lock(&cache->lock);
2055                 if (reserve) {
2056                         cache->reserved += len;
2057                         cache->space_info->bytes_reserved += len;
2058                 } else {
2059                         cache->reserved -= len;
2060                         cache->space_info->bytes_reserved -= len;
2061                 }
2062                 spin_unlock(&cache->lock);
2063                 spin_unlock(&cache->space_info->lock);
2064                 put_block_group(cache);
2065                 bytenr += len;
2066                 num -= len;
2067         }
2068         return 0;
2069 }
2070
2071 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2072 {
2073         u64 last = 0;
2074         u64 start;
2075         u64 end;
2076         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2077         int ret;
2078
2079         mutex_lock(&root->fs_info->pinned_mutex);
2080         while(1) {
2081                 ret = find_first_extent_bit(pinned_extents, last,
2082                                             &start, &end, EXTENT_DIRTY);
2083                 if (ret)
2084                         break;
2085                 set_extent_dirty(copy, start, end, GFP_NOFS);
2086                 last = end + 1;
2087         }
2088         mutex_unlock(&root->fs_info->pinned_mutex);
2089         return 0;
2090 }
2091
2092 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2093                                struct btrfs_root *root,
2094                                struct extent_io_tree *unpin)
2095 {
2096         u64 start;
2097         u64 end;
2098         int ret;
2099
2100         mutex_lock(&root->fs_info->pinned_mutex);
2101         while(1) {
2102                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2103                                             EXTENT_DIRTY);
2104                 if (ret)
2105                         break;
2106                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2107                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2108                 if (need_resched()) {
2109                         mutex_unlock(&root->fs_info->pinned_mutex);
2110                         cond_resched();
2111                         mutex_lock(&root->fs_info->pinned_mutex);
2112                 }
2113         }
2114         mutex_unlock(&root->fs_info->pinned_mutex);
2115         return 0;
2116 }
2117
2118 static int finish_current_insert(struct btrfs_trans_handle *trans,
2119                                  struct btrfs_root *extent_root, int all)
2120 {
2121         u64 start;
2122         u64 end;
2123         u64 priv;
2124         u64 search = 0;
2125         u64 skipped = 0;
2126         struct btrfs_fs_info *info = extent_root->fs_info;
2127         struct btrfs_path *path;
2128         struct pending_extent_op *extent_op, *tmp;
2129         struct list_head insert_list, update_list;
2130         int ret;
2131         int num_inserts = 0, max_inserts;
2132
2133         path = btrfs_alloc_path();
2134         INIT_LIST_HEAD(&insert_list);
2135         INIT_LIST_HEAD(&update_list);
2136
2137         max_inserts = extent_root->leafsize /
2138                 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2139                  sizeof(struct btrfs_extent_ref) +
2140                  sizeof(struct btrfs_extent_item));
2141 again:
2142         mutex_lock(&info->extent_ins_mutex);
2143         while (1) {
2144                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2145                                             &end, EXTENT_WRITEBACK);
2146                 if (ret) {
2147                         if (skipped && all && !num_inserts) {
2148                                 skipped = 0;
2149                                 search = 0;
2150                                 continue;
2151                         }
2152                         mutex_unlock(&info->extent_ins_mutex);
2153                         break;
2154                 }
2155
2156                 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2157                 if (!ret) {
2158                         skipped = 1;
2159                         search = end + 1;
2160                         if (need_resched()) {
2161                                 mutex_unlock(&info->extent_ins_mutex);
2162                                 cond_resched();
2163                                 mutex_lock(&info->extent_ins_mutex);
2164                         }
2165                         continue;
2166                 }
2167
2168                 ret = get_state_private(&info->extent_ins, start, &priv);
2169                 BUG_ON(ret);
2170                 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2171
2172                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2173                         num_inserts++;
2174                         list_add_tail(&extent_op->list, &insert_list);
2175                         search = end + 1;
2176                         if (num_inserts == max_inserts) {
2177                                 mutex_unlock(&info->extent_ins_mutex);
2178                                 break;
2179                         }
2180                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2181                         list_add_tail(&extent_op->list, &update_list);
2182                         search = end + 1;
2183                 } else {
2184                         BUG();
2185                 }
2186         }
2187
2188         /*
2189          * process the update list, clear the writeback bit for it, and if
2190          * somebody marked this thing for deletion then just unlock it and be
2191          * done, the free_extents will handle it
2192          */
2193         mutex_lock(&info->extent_ins_mutex);
2194         list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2195                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2196                                   extent_op->bytenr + extent_op->num_bytes - 1,
2197                                   EXTENT_WRITEBACK, GFP_NOFS);
2198                 if (extent_op->del) {
2199                         list_del_init(&extent_op->list);
2200                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2201                                       extent_op->bytenr + extent_op->num_bytes
2202                                       - 1, GFP_NOFS);
2203                         kfree(extent_op);
2204                 }
2205         }
2206         mutex_unlock(&info->extent_ins_mutex);
2207
2208         /*
2209          * still have things left on the update list, go ahead an update
2210          * everything
2211          */
2212         if (!list_empty(&update_list)) {
2213                 ret = update_backrefs(trans, extent_root, path, &update_list);
2214                 BUG_ON(ret);
2215         }
2216
2217         /*
2218          * if no inserts need to be done, but we skipped some extents and we
2219          * need to make sure everything is cleaned then reset everything and
2220          * go back to the beginning
2221          */
2222         if (!num_inserts && all && skipped) {
2223                 search = 0;
2224                 skipped = 0;
2225                 INIT_LIST_HEAD(&update_list);
2226                 INIT_LIST_HEAD(&insert_list);
2227                 goto again;
2228         } else if (!num_inserts) {
2229                 goto out;
2230         }
2231
2232         /*
2233          * process the insert extents list.  Again if we are deleting this
2234          * extent, then just unlock it, pin down the bytes if need be, and be
2235          * done with it.  Saves us from having to actually insert the extent
2236          * into the tree and then subsequently come along and delete it
2237          */
2238         mutex_lock(&info->extent_ins_mutex);
2239         list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2240                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2241                                   extent_op->bytenr + extent_op->num_bytes - 1,
2242                                   EXTENT_WRITEBACK, GFP_NOFS);
2243                 if (extent_op->del) {
2244                         list_del_init(&extent_op->list);
2245                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2246                                       extent_op->bytenr + extent_op->num_bytes
2247                                       - 1, GFP_NOFS);
2248
2249                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2250                         ret = pin_down_bytes(trans, extent_root,
2251                                              extent_op->bytenr,
2252                                              extent_op->num_bytes, 0);
2253                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2254
2255                         ret = update_block_group(trans, extent_root,
2256                                                  extent_op->bytenr,
2257                                                  extent_op->num_bytes,
2258                                                  0, ret > 0);
2259                         BUG_ON(ret);
2260                         kfree(extent_op);
2261                         num_inserts--;
2262                 }
2263         }
2264         mutex_unlock(&info->extent_ins_mutex);
2265
2266         ret = insert_extents(trans, extent_root, path, &insert_list,
2267                              num_inserts);
2268         BUG_ON(ret);
2269
2270         /*
2271          * if we broke out of the loop in order to insert stuff because we hit
2272          * the maximum number of inserts at a time we can handle, then loop
2273          * back and pick up where we left off
2274          */
2275         if (num_inserts == max_inserts) {
2276                 INIT_LIST_HEAD(&insert_list);
2277                 INIT_LIST_HEAD(&update_list);
2278                 num_inserts = 0;
2279                 goto again;
2280         }
2281
2282         /*
2283          * again, if we need to make absolutely sure there are no more pending
2284          * extent operations left and we know that we skipped some, go back to
2285          * the beginning and do it all again
2286          */
2287         if (all && skipped) {
2288                 INIT_LIST_HEAD(&insert_list);
2289                 INIT_LIST_HEAD(&update_list);
2290                 search = 0;
2291                 skipped = 0;
2292                 num_inserts = 0;
2293                 goto again;
2294         }
2295 out:
2296         btrfs_free_path(path);
2297         return 0;
2298 }
2299
2300 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2301                           struct btrfs_root *root,
2302                           u64 bytenr, u64 num_bytes, int is_data)
2303 {
2304         int err = 0;
2305         struct extent_buffer *buf;
2306
2307         if (is_data)
2308                 goto pinit;
2309
2310         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2311         if (!buf)
2312                 goto pinit;
2313
2314         /* we can reuse a block if it hasn't been written
2315          * and it is from this transaction.  We can't
2316          * reuse anything from the tree log root because
2317          * it has tiny sub-transactions.
2318          */
2319         if (btrfs_buffer_uptodate(buf, 0) &&
2320             btrfs_try_tree_lock(buf)) {
2321                 u64 header_owner = btrfs_header_owner(buf);
2322                 u64 header_transid = btrfs_header_generation(buf);
2323                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2324                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2325                     header_transid == trans->transid &&
2326                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2327                         clean_tree_block(NULL, root, buf);
2328                         btrfs_tree_unlock(buf);
2329                         free_extent_buffer(buf);
2330                         return 1;
2331                 }
2332                 btrfs_tree_unlock(buf);
2333         }
2334         free_extent_buffer(buf);
2335 pinit:
2336         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2337
2338         BUG_ON(err < 0);
2339         return 0;
2340 }
2341
2342 /*
2343  * remove an extent from the root, returns 0 on success
2344  */
2345 static int __free_extent(struct btrfs_trans_handle *trans,
2346                          struct btrfs_root *root,
2347                          u64 bytenr, u64 num_bytes, u64 parent,
2348                          u64 root_objectid, u64 ref_generation,
2349                          u64 owner_objectid, int pin, int mark_free)
2350 {
2351         struct btrfs_path *path;
2352         struct btrfs_key key;
2353         struct btrfs_fs_info *info = root->fs_info;
2354         struct btrfs_root *extent_root = info->extent_root;
2355         struct extent_buffer *leaf;
2356         int ret;
2357         int extent_slot = 0;
2358         int found_extent = 0;
2359         int num_to_del = 1;
2360         struct btrfs_extent_item *ei;
2361         u32 refs;
2362
2363         key.objectid = bytenr;
2364         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2365         key.offset = num_bytes;
2366         path = btrfs_alloc_path();
2367         if (!path)
2368                 return -ENOMEM;
2369
2370         path->reada = 1;
2371         ret = lookup_extent_backref(trans, extent_root, path,
2372                                     bytenr, parent, root_objectid,
2373                                     ref_generation, owner_objectid, 1);
2374         if (ret == 0) {
2375                 struct btrfs_key found_key;
2376                 extent_slot = path->slots[0];
2377                 while(extent_slot > 0) {
2378                         extent_slot--;
2379                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2380                                               extent_slot);
2381                         if (found_key.objectid != bytenr)
2382                                 break;
2383                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2384                             found_key.offset == num_bytes) {
2385                                 found_extent = 1;
2386                                 break;
2387                         }
2388                         if (path->slots[0] - extent_slot > 5)
2389                                 break;
2390                 }
2391                 if (!found_extent) {
2392                         ret = remove_extent_backref(trans, extent_root, path);
2393                         BUG_ON(ret);
2394                         btrfs_release_path(extent_root, path);
2395                         ret = btrfs_search_slot(trans, extent_root,
2396                                                 &key, path, -1, 1);
2397                         if (ret) {
2398                                 printk(KERN_ERR "umm, got %d back from search"
2399                                        ", was looking for %Lu\n", ret,
2400                                        bytenr);
2401                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2402                         }
2403                         BUG_ON(ret);
2404                         extent_slot = path->slots[0];
2405                 }
2406         } else {
2407                 btrfs_print_leaf(extent_root, path->nodes[0]);
2408                 WARN_ON(1);
2409                 printk("Unable to find ref byte nr %Lu root %Lu "
2410                        "gen %Lu owner %Lu\n", bytenr,
2411                        root_objectid, ref_generation, owner_objectid);
2412         }
2413
2414         leaf = path->nodes[0];
2415         ei = btrfs_item_ptr(leaf, extent_slot,
2416                             struct btrfs_extent_item);
2417         refs = btrfs_extent_refs(leaf, ei);
2418         BUG_ON(refs == 0);
2419         refs -= 1;
2420         btrfs_set_extent_refs(leaf, ei, refs);
2421
2422         btrfs_mark_buffer_dirty(leaf);
2423
2424         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2425                 struct btrfs_extent_ref *ref;
2426                 ref = btrfs_item_ptr(leaf, path->slots[0],
2427                                      struct btrfs_extent_ref);
2428                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2429                 /* if the back ref and the extent are next to each other
2430                  * they get deleted below in one shot
2431                  */
2432                 path->slots[0] = extent_slot;
2433                 num_to_del = 2;
2434         } else if (found_extent) {
2435                 /* otherwise delete the extent back ref */
2436                 ret = remove_extent_backref(trans, extent_root, path);
2437                 BUG_ON(ret);
2438                 /* if refs are 0, we need to setup the path for deletion */
2439                 if (refs == 0) {
2440                         btrfs_release_path(extent_root, path);
2441                         ret = btrfs_search_slot(trans, extent_root, &key, path,
2442                                                 -1, 1);
2443                         BUG_ON(ret);
2444                 }
2445         }
2446
2447         if (refs == 0) {
2448                 u64 super_used;
2449                 u64 root_used;
2450 #ifdef BIO_RW_DISCARD
2451                 u64 map_length = num_bytes;
2452                 struct btrfs_multi_bio *multi = NULL;
2453 #endif
2454
2455                 if (pin) {
2456                         mutex_lock(&root->fs_info->pinned_mutex);
2457                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2458                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2459                         mutex_unlock(&root->fs_info->pinned_mutex);
2460                         if (ret > 0)
2461                                 mark_free = 1;
2462                         BUG_ON(ret < 0);
2463                 }
2464                 /* block accounting for super block */
2465                 spin_lock_irq(&info->delalloc_lock);
2466                 super_used = btrfs_super_bytes_used(&info->super_copy);
2467                 btrfs_set_super_bytes_used(&info->super_copy,
2468                                            super_used - num_bytes);
2469                 spin_unlock_irq(&info->delalloc_lock);
2470
2471                 /* block accounting for root item */
2472                 root_used = btrfs_root_used(&root->root_item);
2473                 btrfs_set_root_used(&root->root_item,
2474                                            root_used - num_bytes);
2475                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2476                                       num_to_del);
2477                 BUG_ON(ret);
2478                 btrfs_release_path(extent_root, path);
2479                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2480                                          mark_free);
2481                 BUG_ON(ret);
2482
2483                 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2484                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2485                         BUG_ON(ret);
2486                 }
2487
2488 #ifdef BIO_RW_DISCARD
2489                 /* Tell the block device(s) that the sectors can be discarded */
2490                 ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2491                                       bytenr, &map_length, &multi, 0);
2492                 if (!ret) {
2493                         struct btrfs_bio_stripe *stripe = multi->stripes;
2494                         int i;
2495
2496                         if (map_length > num_bytes)
2497                                 map_length = num_bytes;
2498
2499                         for (i = 0; i < multi->num_stripes; i++, stripe++) {
2500                                 btrfs_issue_discard(stripe->dev->bdev,
2501                                                     stripe->physical,
2502                                                      map_length);
2503                         }
2504                         kfree(multi);
2505                 }
2506 #endif
2507         }
2508         btrfs_free_path(path);
2509         finish_current_insert(trans, extent_root, 0);
2510         return ret;
2511 }
2512
2513 /*
2514  * find all the blocks marked as pending in the radix tree and remove
2515  * them from the extent map
2516  */
2517 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2518                                btrfs_root *extent_root, int all)
2519 {
2520         int ret;
2521         int err = 0;
2522         u64 start;
2523         u64 end;
2524         u64 priv;
2525         u64 search = 0;
2526         int nr = 0, skipped = 0;
2527         struct extent_io_tree *pending_del;
2528         struct extent_io_tree *extent_ins;
2529         struct pending_extent_op *extent_op;
2530         struct btrfs_fs_info *info = extent_root->fs_info;
2531         struct list_head delete_list;
2532
2533         INIT_LIST_HEAD(&delete_list);
2534         extent_ins = &extent_root->fs_info->extent_ins;
2535         pending_del = &extent_root->fs_info->pending_del;
2536
2537 again:
2538         mutex_lock(&info->extent_ins_mutex);
2539         while(1) {
2540                 ret = find_first_extent_bit(pending_del, search, &start, &end,
2541                                             EXTENT_WRITEBACK);
2542                 if (ret) {
2543                         if (all && skipped && !nr) {
2544                                 search = 0;
2545                                 continue;
2546                         }
2547                         mutex_unlock(&info->extent_ins_mutex);
2548                         break;
2549                 }
2550
2551                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2552                 if (!ret) {
2553                         search = end+1;
2554                         skipped = 1;
2555
2556                         if (need_resched()) {
2557                                 mutex_unlock(&info->extent_ins_mutex);
2558                                 cond_resched();
2559                                 mutex_lock(&info->extent_ins_mutex);
2560                         }
2561
2562                         continue;
2563                 }
2564                 BUG_ON(ret < 0);
2565
2566                 ret = get_state_private(pending_del, start, &priv);
2567                 BUG_ON(ret);
2568                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2569
2570                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2571                                   GFP_NOFS);
2572                 if (!test_range_bit(extent_ins, start, end,
2573                                     EXTENT_WRITEBACK, 0)) {
2574                         list_add_tail(&extent_op->list, &delete_list);
2575                         nr++;
2576                 } else {
2577                         kfree(extent_op);
2578
2579                         ret = get_state_private(&info->extent_ins, start,
2580                                                 &priv);
2581                         BUG_ON(ret);
2582                         extent_op = (struct pending_extent_op *)
2583                                                 (unsigned long)priv;
2584
2585                         clear_extent_bits(&info->extent_ins, start, end,
2586                                           EXTENT_WRITEBACK, GFP_NOFS);
2587
2588                         if (extent_op->type == PENDING_BACKREF_UPDATE) {
2589                                 list_add_tail(&extent_op->list, &delete_list);
2590                                 search = end + 1;
2591                                 nr++;
2592                                 continue;
2593                         }
2594
2595                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2596                         ret = pin_down_bytes(trans, extent_root, start,
2597                                              end + 1 - start, 0);
2598                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2599
2600                         ret = update_block_group(trans, extent_root, start,
2601                                                 end + 1 - start, 0, ret > 0);
2602
2603                         unlock_extent(extent_ins, start, end, GFP_NOFS);
2604                         BUG_ON(ret);
2605                         kfree(extent_op);
2606                 }
2607                 if (ret)
2608                         err = ret;
2609
2610                 search = end + 1;
2611
2612                 if (need_resched()) {
2613                         mutex_unlock(&info->extent_ins_mutex);
2614                         cond_resched();
2615                         mutex_lock(&info->extent_ins_mutex);
2616                 }
2617         }
2618
2619         if (nr) {
2620                 ret = free_extents(trans, extent_root, &delete_list);
2621                 BUG_ON(ret);
2622         }
2623
2624         if (all && skipped) {
2625                 INIT_LIST_HEAD(&delete_list);
2626                 search = 0;
2627                 nr = 0;
2628                 goto again;
2629         }
2630
2631         return err;
2632 }
2633
2634 /*
2635  * remove an extent from the root, returns 0 on success
2636  */
2637 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2638                                struct btrfs_root *root,
2639                                u64 bytenr, u64 num_bytes, u64 parent,
2640                                u64 root_objectid, u64 ref_generation,
2641                                u64 owner_objectid, int pin)
2642 {
2643         struct btrfs_root *extent_root = root->fs_info->extent_root;
2644         int pending_ret;
2645         int ret;
2646
2647         WARN_ON(num_bytes < root->sectorsize);
2648         if (root == extent_root) {
2649                 struct pending_extent_op *extent_op = NULL;
2650
2651                 mutex_lock(&root->fs_info->extent_ins_mutex);
2652                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2653                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2654                         u64 priv;
2655                         ret = get_state_private(&root->fs_info->extent_ins,
2656                                                 bytenr, &priv);
2657                         BUG_ON(ret);
2658                         extent_op = (struct pending_extent_op *)
2659                                                 (unsigned long)priv;
2660
2661                         extent_op->del = 1;
2662                         if (extent_op->type == PENDING_EXTENT_INSERT) {
2663                                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2664                                 return 0;
2665                         }
2666                 }
2667
2668                 if (extent_op) {
2669                         ref_generation = extent_op->orig_generation;
2670                         parent = extent_op->orig_parent;
2671                 }
2672
2673                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2674                 BUG_ON(!extent_op);
2675
2676                 extent_op->type = PENDING_EXTENT_DELETE;
2677                 extent_op->bytenr = bytenr;
2678                 extent_op->num_bytes = num_bytes;
2679                 extent_op->parent = parent;
2680                 extent_op->orig_parent = parent;
2681                 extent_op->generation = ref_generation;
2682                 extent_op->orig_generation = ref_generation;
2683                 extent_op->level = (int)owner_objectid;
2684                 INIT_LIST_HEAD(&extent_op->list);
2685                 extent_op->del = 0;
2686
2687                 set_extent_bits(&root->fs_info->pending_del,
2688                                 bytenr, bytenr + num_bytes - 1,
2689                                 EXTENT_WRITEBACK, GFP_NOFS);
2690                 set_state_private(&root->fs_info->pending_del,
2691                                   bytenr, (unsigned long)extent_op);
2692                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2693                 return 0;
2694         }
2695         /* if metadata always pin */
2696         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2697                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2698                         struct btrfs_block_group_cache *cache;
2699
2700                         /* btrfs_free_reserved_extent */
2701                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2702                         BUG_ON(!cache);
2703                         btrfs_add_free_space(cache, bytenr, num_bytes);
2704                         put_block_group(cache);
2705                         update_reserved_extents(root, bytenr, num_bytes, 0);
2706                         return 0;
2707                 }
2708                 pin = 1;
2709         }
2710
2711         /* if data pin when any transaction has committed this */
2712         if (ref_generation != trans->transid)
2713                 pin = 1;
2714
2715         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2716                             root_objectid, ref_generation,
2717                             owner_objectid, pin, pin == 0);
2718
2719         finish_current_insert(trans, root->fs_info->extent_root, 0);
2720         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2721         return ret ? ret : pending_ret;
2722 }
2723
2724 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2725                       struct btrfs_root *root,
2726                       u64 bytenr, u64 num_bytes, u64 parent,
2727                       u64 root_objectid, u64 ref_generation,
2728                       u64 owner_objectid, int pin)
2729 {
2730         int ret;
2731
2732         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2733                                   root_objectid, ref_generation,
2734                                   owner_objectid, pin);
2735         return ret;
2736 }
2737
2738 static u64 stripe_align(struct btrfs_root *root, u64 val)
2739 {
2740         u64 mask = ((u64)root->stripesize - 1);
2741         u64 ret = (val + mask) & ~mask;
2742         return ret;
2743 }
2744
2745 /*
2746  * walks the btree of allocated extents and find a hole of a given size.
2747  * The key ins is changed to record the hole:
2748  * ins->objectid == block start
2749  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2750  * ins->offset == number of blocks
2751  * Any available blocks before search_start are skipped.
2752  */
2753 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2754                                      struct btrfs_root *orig_root,
2755                                      u64 num_bytes, u64 empty_size,
2756                                      u64 search_start, u64 search_end,
2757                                      u64 hint_byte, struct btrfs_key *ins,
2758                                      u64 exclude_start, u64 exclude_nr,
2759                                      int data)
2760 {
2761         int ret = 0;
2762         struct btrfs_root * root = orig_root->fs_info->extent_root;
2763         u64 total_needed = num_bytes;
2764         u64 *last_ptr = NULL;
2765         u64 last_wanted = 0;
2766         struct btrfs_block_group_cache *block_group = NULL;
2767         int chunk_alloc_done = 0;
2768         int empty_cluster = 2 * 1024 * 1024;
2769         int allowed_chunk_alloc = 0;
2770         struct list_head *head = NULL, *cur = NULL;
2771         int loop = 0;
2772         int extra_loop = 0;
2773         struct btrfs_space_info *space_info;
2774
2775         WARN_ON(num_bytes < root->sectorsize);
2776         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2777         ins->objectid = 0;
2778         ins->offset = 0;
2779
2780         if (orig_root->ref_cows || empty_size)
2781                 allowed_chunk_alloc = 1;
2782
2783         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2784                 last_ptr = &root->fs_info->last_alloc;
2785                 empty_cluster = 64 * 1024;
2786         }
2787
2788         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2789                 last_ptr = &root->fs_info->last_data_alloc;
2790
2791         if (last_ptr) {
2792                 if (*last_ptr) {
2793                         hint_byte = *last_ptr;
2794                         last_wanted = *last_ptr;
2795                 } else
2796                         empty_size += empty_cluster;
2797         } else {
2798                 empty_cluster = 0;
2799         }
2800         search_start = max(search_start, first_logical_byte(root, 0));
2801         search_start = max(search_start, hint_byte);
2802
2803         if (last_wanted && search_start != last_wanted) {
2804                 last_wanted = 0;
2805                 empty_size += empty_cluster;
2806         }
2807
2808         total_needed += empty_size;
2809         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2810         if (!block_group)
2811                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2812                                                              search_start);
2813         space_info = __find_space_info(root->fs_info, data);
2814
2815         down_read(&space_info->groups_sem);
2816         while (1) {
2817                 struct btrfs_free_space *free_space;
2818                 /*
2819                  * the only way this happens if our hint points to a block
2820                  * group thats not of the proper type, while looping this
2821                  * should never happen
2822                  */
2823                 if (empty_size)
2824                         extra_loop = 1;
2825
2826                 if (!block_group)
2827                         goto new_group_no_lock;
2828
2829                 if (unlikely(!block_group->cached)) {
2830                         mutex_lock(&block_group->cache_mutex);
2831                         ret = cache_block_group(root, block_group);
2832                         mutex_unlock(&block_group->cache_mutex);
2833                         if (ret)
2834                                 break;
2835                 }
2836
2837                 mutex_lock(&block_group->alloc_mutex);
2838                 if (unlikely(!block_group_bits(block_group, data)))
2839                         goto new_group;
2840
2841                 if (unlikely(block_group->ro))
2842                         goto new_group;
2843
2844                 free_space = btrfs_find_free_space(block_group, search_start,
2845                                                    total_needed);
2846                 if (free_space) {
2847                         u64 start = block_group->key.objectid;
2848                         u64 end = block_group->key.objectid +
2849                                 block_group->key.offset;
2850
2851                         search_start = stripe_align(root, free_space->offset);
2852
2853                         /* move on to the next group */
2854                         if (search_start + num_bytes >= search_end)
2855                                 goto new_group;
2856
2857                         /* move on to the next group */
2858                         if (search_start + num_bytes > end)
2859                                 goto new_group;
2860
2861                         if (last_wanted && search_start != last_wanted) {
2862                                 total_needed += empty_cluster;
2863                                 empty_size += empty_cluster;
2864                                 last_wanted = 0;
2865                                 /*
2866                                  * if search_start is still in this block group
2867                                  * then we just re-search this block group
2868                                  */
2869                                 if (search_start >= start &&
2870                                     search_start < end) {
2871                                         mutex_unlock(&block_group->alloc_mutex);
2872                                         continue;
2873                                 }
2874
2875                                 /* else we go to the next block group */
2876                                 goto new_group;
2877                         }
2878
2879                         if (exclude_nr > 0 &&
2880                             (search_start + num_bytes > exclude_start &&
2881                              search_start < exclude_start + exclude_nr)) {
2882                                 search_start = exclude_start + exclude_nr;
2883                                 /*
2884                                  * if search_start is still in this block group
2885                                  * then we just re-search this block group
2886                                  */
2887                                 if (search_start >= start &&
2888                                     search_start < end) {
2889                                         mutex_unlock(&block_group->alloc_mutex);
2890                                         last_wanted = 0;
2891                                         continue;
2892                                 }
2893
2894                                 /* else we go to the next block group */
2895                                 goto new_group;
2896                         }
2897
2898                         ins->objectid = search_start;
2899                         ins->offset = num_bytes;
2900
2901                         btrfs_remove_free_space_lock(block_group, search_start,
2902                                                      num_bytes);
2903                         /* we are all good, lets return */
2904                         mutex_unlock(&block_group->alloc_mutex);
2905                         break;
2906                 }
2907 new_group:
2908                 mutex_unlock(&block_group->alloc_mutex);
2909                 put_block_group(block_group);
2910                 block_group = NULL;
2911 new_group_no_lock:
2912                 /* don't try to compare new allocations against the
2913                  * last allocation any more
2914                  */
2915                 last_wanted = 0;
2916
2917                 /*
2918                  * Here's how this works.
2919                  * loop == 0: we were searching a block group via a hint
2920                  *              and didn't find anything, so we start at
2921                  *              the head of the block groups and keep searching
2922                  * loop == 1: we're searching through all of the block groups
2923                  *              if we hit the head again we have searched
2924                  *              all of the block groups for this space and we
2925                  *              need to try and allocate, if we cant error out.
2926                  * loop == 2: we allocated more space and are looping through
2927                  *              all of the block groups again.
2928                  */
2929                 if (loop == 0) {
2930                         head = &space_info->block_groups;
2931                         cur = head->next;
2932                         loop++;
2933                 } else if (loop == 1 && cur == head) {
2934                         int keep_going;
2935
2936                         /* at this point we give up on the empty_size
2937                          * allocations and just try to allocate the min
2938                          * space.
2939                          *
2940                          * The extra_loop field was set if an empty_size
2941                          * allocation was attempted above, and if this
2942                          * is try we need to try the loop again without
2943                          * the additional empty_size.
2944                          */
2945                         total_needed -= empty_size;
2946                         empty_size = 0;
2947                         keep_going = extra_loop;
2948                         loop++;
2949
2950                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2951                                 up_read(&space_info->groups_sem);
2952                                 ret = do_chunk_alloc(trans, root, num_bytes +
2953                                                      2 * 1024 * 1024, data, 1);
2954                                 down_read(&space_info->groups_sem);
2955                                 if (ret < 0)
2956                                         goto loop_check;
2957                                 head = &space_info->block_groups;
2958                                 /*
2959                                  * we've allocated a new chunk, keep
2960                                  * trying
2961                                  */
2962                                 keep_going = 1;
2963                                 chunk_alloc_done = 1;
2964                         } else if (!allowed_chunk_alloc) {
2965                                 space_info->force_alloc = 1;
2966                         }
2967 loop_check:
2968                         if (keep_going) {
2969                                 cur = head->next;
2970                                 extra_loop = 0;
2971                         } else {
2972                                 break;
2973                         }
2974                 } else if (cur == head) {
2975                         break;
2976                 }
2977
2978                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2979                                          list);
2980                 atomic_inc(&block_group->count);
2981
2982                 search_start = block_group->key.objectid;
2983                 cur = cur->next;
2984         }
2985
2986         /* we found what we needed */
2987         if (ins->objectid) {
2988                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2989                         trans->block_group = block_group->key.objectid;
2990
2991                 if (last_ptr)
2992                         *last_ptr = ins->objectid + ins->offset;
2993                 ret = 0;
2994         } else if (!ret) {
2995                 printk(KERN_ERR "we were searching for %Lu bytes, num_bytes %Lu,"
2996                        " loop %d, allowed_alloc %d\n", total_needed, num_bytes,
2997                        loop, allowed_chunk_alloc);
2998                 ret = -ENOSPC;
2999         }
3000         if (block_group)
3001                 put_block_group(block_group);
3002
3003         up_read(&space_info->groups_sem);
3004         return ret;
3005 }
3006
3007 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3008 {
3009         struct btrfs_block_group_cache *cache;
3010         struct list_head *l;
3011
3012         printk(KERN_INFO "space_info has %Lu free, is %sfull\n",
3013                info->total_bytes - info->bytes_used - info->bytes_pinned -
3014                info->bytes_reserved, (info->full) ? "" : "not ");
3015
3016         down_read(&info->groups_sem);
3017         list_for_each(l, &info->block_groups) {
3018                 cache = list_entry(l, struct btrfs_block_group_cache, list);
3019                 spin_lock(&cache->lock);
3020                 printk(KERN_INFO "block group %Lu has %Lu bytes, %Lu used "
3021                        "%Lu pinned %Lu reserved\n",
3022                        cache->key.objectid, cache->key.offset,
3023                        btrfs_block_group_used(&cache->item),
3024                        cache->pinned, cache->reserved);
3025                 btrfs_dump_free_space(cache, bytes);
3026                 spin_unlock(&cache->lock);
3027         }
3028         up_read(&info->groups_sem);
3029 }
3030
3031 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3032                                   struct btrfs_root *root,
3033                                   u64 num_bytes, u64 min_alloc_size,
3034                                   u64 empty_size, u64 hint_byte,
3035                                   u64 search_end, struct btrfs_key *ins,
3036                                   u64 data)
3037 {
3038         int ret;
3039         u64 search_start = 0;
3040         u64 alloc_profile;
3041         struct btrfs_fs_info *info = root->fs_info;
3042
3043         if (data) {
3044                 alloc_profile = info->avail_data_alloc_bits &
3045                                 info->data_alloc_profile;
3046                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3047         } else if (root == root->fs_info->chunk_root) {
3048                 alloc_profile = info->avail_system_alloc_bits &
3049                                 info->system_alloc_profile;
3050                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3051         } else {
3052                 alloc_profile = info->avail_metadata_alloc_bits &
3053                                 info->metadata_alloc_profile;
3054                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3055         }
3056 again:
3057         data = btrfs_reduce_alloc_profile(root, data);
3058         /*
3059          * the only place that sets empty_size is btrfs_realloc_node, which
3060          * is not called recursively on allocations
3061          */
3062         if (empty_size || root->ref_cows) {
3063                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3064                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3065                                      2 * 1024 * 1024,
3066                                      BTRFS_BLOCK_GROUP_METADATA |
3067                                      (info->metadata_alloc_profile &
3068                                       info->avail_metadata_alloc_bits), 0);
3069                 }
3070                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3071                                      num_bytes + 2 * 1024 * 1024, data, 0);
3072         }
3073
3074         WARN_ON(num_bytes < root->sectorsize);
3075         ret = find_free_extent(trans, root, num_bytes, empty_size,
3076                                search_start, search_end, hint_byte, ins,
3077                                trans->alloc_exclude_start,
3078                                trans->alloc_exclude_nr, data);
3079
3080         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3081                 num_bytes = num_bytes >> 1;
3082                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3083                 num_bytes = max(num_bytes, min_alloc_size);
3084                 do_chunk_alloc(trans, root->fs_info->extent_root,
3085                                num_bytes, data, 1);
3086                 goto again;
3087         }
3088         if (ret) {
3089                 struct btrfs_space_info *sinfo;
3090
3091                 sinfo = __find_space_info(root->fs_info, data);
3092                 printk("allocation failed flags %Lu, wanted %Lu\n",
3093                        data, num_bytes);
3094                 dump_space_info(sinfo, num_bytes);
3095                 BUG();
3096         }
3097
3098         return ret;
3099 }
3100
3101 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3102 {
3103         struct btrfs_block_group_cache *cache;
3104
3105         cache = btrfs_lookup_block_group(root->fs_info, start);
3106         if (!cache) {
3107                 printk(KERN_ERR "Unable to find block group for %Lu\n", start);
3108                 return -ENOSPC;
3109         }
3110         btrfs_add_free_space(cache, start, len);
3111         put_block_group(cache);
3112         update_reserved_extents(root, start, len, 0);
3113         return 0;
3114 }
3115
3116 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3117                                   struct btrfs_root *root,
3118                                   u64 num_bytes, u64 min_alloc_size,
3119                                   u64 empty_size, u64 hint_byte,
3120                                   u64 search_end, struct btrfs_key *ins,
3121                                   u64 data)
3122 {
3123         int ret;
3124         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3125                                      empty_size, hint_byte, search_end, ins,
3126                                      data);
3127         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3128         return ret;
3129 }
3130
3131 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3132                                          struct btrfs_root *root, u64 parent,
3133                                          u64 root_objectid, u64 ref_generation,
3134                                          u64 owner, struct btrfs_key *ins)
3135 {
3136         int ret;
3137         int pending_ret;
3138         u64 super_used;
3139         u64 root_used;
3140         u64 num_bytes = ins->offset;
3141         u32 sizes[2];
3142         struct btrfs_fs_info *info = root->fs_info;
3143         struct btrfs_root *extent_root = info->extent_root;
3144         struct btrfs_extent_item *extent_item;
3145         struct btrfs_extent_ref *ref;
3146         struct btrfs_path *path;
3147         struct btrfs_key keys[2];
3148
3149         if (parent == 0)
3150                 parent = ins->objectid;
3151
3152         /* block accounting for super block */
3153         spin_lock_irq(&info->delalloc_lock);
3154         super_used = btrfs_super_bytes_used(&info->super_copy);
3155         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3156         spin_unlock_irq(&info->delalloc_lock);
3157
3158         /* block accounting for root item */
3159         root_used = btrfs_root_used(&root->root_item);
3160         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3161
3162         if (root == extent_root) {
3163                 struct pending_extent_op *extent_op;
3164
3165                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3166                 BUG_ON(!extent_op);
3167
3168                 extent_op->type = PENDING_EXTENT_INSERT;
3169                 extent_op->bytenr = ins->objectid;
3170                 extent_op->num_bytes = ins->offset;
3171                 extent_op->parent = parent;
3172                 extent_op->orig_parent = 0;
3173                 extent_op->generation = ref_generation;
3174                 extent_op->orig_generation = 0;
3175                 extent_op->level = (int)owner;
3176                 INIT_LIST_HEAD(&extent_op->list);
3177                 extent_op->del = 0;
3178
3179                 mutex_lock(&root->fs_info->extent_ins_mutex);
3180                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3181                                 ins->objectid + ins->offset - 1,
3182                                 EXTENT_WRITEBACK, GFP_NOFS);
3183                 set_state_private(&root->fs_info->extent_ins,
3184                                   ins->objectid, (unsigned long)extent_op);
3185                 mutex_unlock(&root->fs_info->extent_ins_mutex);
3186                 goto update_block;
3187         }
3188
3189         memcpy(&keys[0], ins, sizeof(*ins));
3190         keys[1].objectid = ins->objectid;
3191         keys[1].type = BTRFS_EXTENT_REF_KEY;
3192         keys[1].offset = parent;
3193         sizes[0] = sizeof(*extent_item);
3194         sizes[1] = sizeof(*ref);
3195
3196         path = btrfs_alloc_path();
3197         BUG_ON(!path);
3198
3199         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3200                                        sizes, 2);
3201         BUG_ON(ret);
3202
3203         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3204                                      struct btrfs_extent_item);
3205         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3206         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3207                              struct btrfs_extent_ref);
3208
3209         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3210         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3211         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3212         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3213
3214         btrfs_mark_buffer_dirty(path->nodes[0]);
3215
3216         trans->alloc_exclude_start = 0;
3217         trans->alloc_exclude_nr = 0;
3218         btrfs_free_path(path);
3219         finish_current_insert(trans, extent_root, 0);
3220         pending_ret = del_pending_extents(trans, extent_root, 0);
3221
3222         if (ret)
3223                 goto out;
3224         if (pending_ret) {
3225                 ret = pending_ret;
3226                 goto out;
3227         }
3228
3229 update_block:
3230         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
3231         if (ret) {
3232                 printk("update block group failed for %Lu %Lu\n",
3233                        ins->objectid, ins->offset);
3234                 BUG();
3235         }
3236 out:
3237         return ret;
3238 }
3239
3240 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3241                                 struct btrfs_root *root, u64 parent,
3242                                 u64 root_objectid, u64 ref_generation,
3243                                 u64 owner, struct btrfs_key *ins)
3244 {
3245         int ret;
3246
3247         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3248                 return 0;
3249         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3250                                             ref_generation, owner, ins);
3251         update_reserved_extents(root, ins->objectid, ins->offset, 0);
3252         return ret;
3253 }
3254
3255 /*
3256  * this is used by the tree logging recovery code.  It records that
3257  * an extent has been allocated and makes sure to clear the free
3258  * space cache bits as well
3259  */
3260 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3261                                 struct btrfs_root *root, u64 parent,
3262                                 u64 root_objectid, u64 ref_generation,
3263                                 u64 owner, struct btrfs_key *ins)
3264 {
3265         int ret;
3266         struct btrfs_block_group_cache *block_group;
3267
3268         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3269         mutex_lock(&block_group->cache_mutex);
3270         cache_block_group(root, block_group);
3271         mutex_unlock(&block_group->cache_mutex);
3272
3273         ret = btrfs_remove_free_space(block_group, ins->objectid,
3274                                       ins->offset);
3275         BUG_ON(ret);
3276         put_block_group(block_group);
3277         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3278                                             ref_generation, owner, ins);
3279         return ret;
3280 }
3281
3282 /*
3283  * finds a free extent and does all the dirty work required for allocation
3284  * returns the key for the extent through ins, and a tree buffer for
3285  * the first block of the extent through buf.
3286  *
3287  * returns 0 if everything worked, non-zero otherwise.
3288  */
3289 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3290                        struct btrfs_root *root,
3291                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3292                        u64 root_objectid, u64 ref_generation,
3293                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3294                        u64 search_end, struct btrfs_key *ins, u64 data)
3295 {
3296         int ret;
3297
3298         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3299                                      min_alloc_size, empty_size, hint_byte,
3300                                      search_end, ins, data);
3301         BUG_ON(ret);
3302         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3303                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3304                                         root_objectid, ref_generation,
3305                                         owner_objectid, ins);
3306                 BUG_ON(ret);
3307
3308         } else {
3309                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3310         }
3311         return ret;
3312 }
3313
3314 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3315                                             struct btrfs_root *root,
3316                                             u64 bytenr, u32 blocksize)
3317 {
3318         struct extent_buffer *buf;
3319
3320         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3321         if (!buf)
3322                 return ERR_PTR(-ENOMEM);
3323         btrfs_set_header_generation(buf, trans->transid);
3324         btrfs_tree_lock(buf);
3325         clean_tree_block(trans, root, buf);
3326         btrfs_set_buffer_uptodate(buf);
3327         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3328                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3329                          buf->start + buf->len - 1, GFP_NOFS);
3330         } else {
3331                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3332                          buf->start + buf->len - 1, GFP_NOFS);
3333         }
3334         trans->blocks_used++;
3335         return buf;
3336 }
3337
3338 /*
3339  * helper function to allocate a block for a given tree
3340  * returns the tree buffer or NULL.
3341  */
3342 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3343                                              struct btrfs_root *root,
3344                                              u32 blocksize, u64 parent,
3345                                              u64 root_objectid,
3346                                              u64 ref_generation,
3347                                              int level,
3348                                              u64 hint,
3349                                              u64 empty_size)
3350 {
3351         struct btrfs_key ins;
3352         int ret;
3353         struct extent_buffer *buf;
3354
3355         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3356                                  root_objectid, ref_generation, level,
3357                                  empty_size, hint, (u64)-1, &ins, 0);
3358         if (ret) {
3359                 BUG_ON(ret > 0);
3360                 return ERR_PTR(ret);
3361         }
3362
3363         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3364         return buf;
3365 }
3366
3367 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3368                         struct btrfs_root *root, struct extent_buffer *leaf)
3369 {
3370         u64 leaf_owner;
3371         u64 leaf_generation;
3372         struct btrfs_key key;
3373         struct btrfs_file_extent_item *fi;
3374         int i;
3375         int nritems;
3376         int ret;
3377
3378         BUG_ON(!btrfs_is_leaf(leaf));
3379         nritems = btrfs_header_nritems(leaf);
3380         leaf_owner = btrfs_header_owner(leaf);
3381         leaf_generation = btrfs_header_generation(leaf);
3382
3383         for (i = 0; i < nritems; i++) {
3384                 u64 disk_bytenr;
3385                 cond_resched();
3386
3387                 btrfs_item_key_to_cpu(leaf, &key, i);
3388                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3389                         continue;
3390                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3391                 if (btrfs_file_extent_type(leaf, fi) ==
3392                     BTRFS_FILE_EXTENT_INLINE)
3393                         continue;
3394                 /*
3395                  * FIXME make sure to insert a trans record that
3396                  * repeats the snapshot del on crash
3397                  */
3398                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3399                 if (disk_bytenr == 0)
3400                         continue;
3401
3402                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3403                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3404                                 leaf->start, leaf_owner, leaf_generation,
3405                                 key.objectid, 0);
3406                 BUG_ON(ret);
3407
3408                 atomic_inc(&root->fs_info->throttle_gen);
3409                 wake_up(&root->fs_info->transaction_throttle);
3410                 cond_resched();
3411         }
3412         return 0;
3413 }
3414
3415 static int noinline cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3416                                         struct btrfs_root *root,
3417                                         struct btrfs_leaf_ref *ref)
3418 {
3419         int i;
3420         int ret;
3421         struct btrfs_extent_info *info = ref->extents;
3422
3423         for (i = 0; i < ref->nritems; i++) {
3424                 ret = __btrfs_free_extent(trans, root, info->bytenr,
3425                                           info->num_bytes, ref->bytenr,
3426                                           ref->owner, ref->generation,
3427                                           info->objectid, 0);
3428
3429                 atomic_inc(&root->fs_info->throttle_gen);
3430                 wake_up(&root->fs_info->transaction_throttle);
3431                 cond_resched();
3432
3433                 BUG_ON(ret);
3434                 info++;
3435         }
3436
3437         return 0;
3438 }
3439
3440 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, u64 len,
3441                               u32 *refs)
3442 {
3443         int ret;
3444
3445         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3446         BUG_ON(ret);
3447
3448 #if 0 // some debugging code in case we see problems here
3449         /* if the refs count is one, it won't get increased again.  But
3450          * if the ref count is > 1, someone may be decreasing it at
3451          * the same time we are.
3452          */
3453         if (*refs != 1) {
3454                 struct extent_buffer *eb = NULL;
3455                 eb = btrfs_find_create_tree_block(root, start, len);
3456                 if (eb)
3457                         btrfs_tree_lock(eb);
3458
3459                 mutex_lock(&root->fs_info->alloc_mutex);
3460                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3461                 BUG_ON(ret);
3462                 mutex_unlock(&root->fs_info->alloc_mutex);
3463
3464                 if (eb) {
3465                         btrfs_tree_unlock(eb);
3466                         free_extent_buffer(eb);
3467                 }
3468                 if (*refs == 1) {
3469                         printk("block %llu went down to one during drop_snap\n",
3470                                (unsigned long long)start);
3471                 }
3472
3473         }
3474 #endif
3475
3476         cond_resched();
3477         return ret;
3478 }
3479
3480 /*
3481  * helper function for drop_snapshot, this walks down the tree dropping ref
3482  * counts as it goes.
3483  */
3484 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
3485                                    struct btrfs_root *root,
3486                                    struct btrfs_path *path, int *level)
3487 {
3488         u64 root_owner;
3489         u64 root_gen;
3490         u64 bytenr;
3491         u64 ptr_gen;
3492         struct extent_buffer *next;
3493         struct extent_buffer *cur;
3494         struct extent_buffer *parent;
3495         struct btrfs_leaf_ref *ref;
3496         u32 blocksize;
3497         int ret;
3498         u32 refs;
3499
3500         WARN_ON(*level < 0);
3501         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3502         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3503                                 path->nodes[*level]->len, &refs);
3504         BUG_ON(ret);
3505         if (refs > 1)
3506                 goto out;
3507
3508         /*
3509          * walk down to the last node level and free all the leaves
3510          */
3511         while(*level >= 0) {
3512                 WARN_ON(*level < 0);
3513                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3514                 cur = path->nodes[*level];
3515
3516                 if (btrfs_header_level(cur) != *level)
3517                         WARN_ON(1);
3518
3519                 if (path->slots[*level] >=
3520                     btrfs_header_nritems(cur))
3521                         break;
3522                 if (*level == 0) {
3523                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3524                         BUG_ON(ret);
3525                         break;
3526                 }
3527                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3528                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3529                 blocksize = btrfs_level_size(root, *level - 1);
3530
3531                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3532                 BUG_ON(ret);
3533                 if (refs != 1) {
3534                         parent = path->nodes[*level];
3535                         root_owner = btrfs_header_owner(parent);
3536                         root_gen = btrfs_header_generation(parent);
3537                         path->slots[*level]++;
3538
3539                         ret = __btrfs_free_extent(trans, root, bytenr,
3540                                                 blocksize, parent->start,
3541                                                 root_owner, root_gen,
3542                                                 *level - 1, 1);
3543                         BUG_ON(ret);
3544
3545                         atomic_inc(&root->fs_info->throttle_gen);
3546                         wake_up(&root->fs_info->transaction_throttle);
3547                         cond_resched();
3548
3549                         continue;
3550                 }
3551                 /*
3552                  * at this point, we have a single ref, and since the
3553                  * only place referencing this extent is a dead root
3554                  * the reference count should never go higher.
3555                  * So, we don't need to check it again
3556                  */
3557                 if (*level == 1) {
3558                         ref = btrfs_lookup_leaf_ref(root, bytenr);
3559                         if (ref && ref->generation != ptr_gen) {
3560                                 btrfs_free_leaf_ref(root, ref);
3561                                 ref = NULL;
3562                         }
3563                         if (ref) {
3564                                 ret = cache_drop_leaf_ref(trans, root, ref);
3565                                 BUG_ON(ret);
3566                                 btrfs_remove_leaf_ref(root, ref);
3567                                 btrfs_free_leaf_ref(root, ref);
3568                                 *level = 0;
3569                                 break;
3570                         }
3571                         if (printk_ratelimit()) {
3572                                 printk("leaf ref miss for bytenr %llu\n",
3573                                        (unsigned long long)bytenr);
3574                         }
3575                 }
3576                 next = btrfs_find_tree_block(root, bytenr, blocksize);
3577                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3578                         free_extent_buffer(next);
3579
3580                         next = read_tree_block(root, bytenr, blocksize,
3581                                                ptr_gen);
3582                         cond_resched();
3583 #if 0
3584                         /*
3585                          * this is a debugging check and can go away
3586                          * the ref should never go all the way down to 1
3587                          * at this point
3588                          */
3589                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3590                                                 &refs);
3591                         BUG_ON(ret);
3592                         WARN_ON(refs != 1);
3593 #endif
3594                 }
3595                 WARN_ON(*level <= 0);
3596                 if (path->nodes[*level-1])
3597                         free_extent_buffer(path->nodes[*level-1]);
3598                 path->nodes[*level-1] = next;
3599                 *level = btrfs_header_level(next);
3600                 path->slots[*level] = 0;
3601                 cond_resched();
3602         }
3603 out:
3604         WARN_ON(*level < 0);
3605         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3606
3607         if (path->nodes[*level] == root->node) {
3608                 parent = path->nodes[*level];
3609                 bytenr = path->nodes[*level]->start;
3610         } else {
3611                 parent = path->nodes[*level + 1];
3612                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3613         }
3614
3615         blocksize = btrfs_level_size(root, *level);
3616         root_owner = btrfs_header_owner(parent);
3617         root_gen = btrfs_header_generation(parent);
3618
3619         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3620                                   parent->start, root_owner, root_gen,
3621                                   *level, 1);
3622         free_extent_buffer(path->nodes[*level]);
3623         path->nodes[*level] = NULL;
3624         *level += 1;
3625         BUG_ON(ret);
3626
3627         cond_resched();
3628         return 0;
3629 }
3630
3631 /*
3632  * helper function for drop_subtree, this function is similar to
3633  * walk_down_tree. The main difference is that it checks reference
3634  * counts while tree blocks are locked.
3635  */
3636 static int noinline walk_down_subtree(struct btrfs_trans_handle *trans,
3637                                       struct btrfs_root *root,
3638                                       struct btrfs_path *path, int *level)
3639 {
3640         struct extent_buffer *next;
3641         struct extent_buffer *cur;
3642         struct extent_buffer *parent;
3643         u64 bytenr;
3644         u64 ptr_gen;
3645         u32 blocksize;
3646         u32 refs;
3647         int ret;
3648
3649         cur = path->nodes[*level];
3650         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3651                                       &refs);
3652         BUG_ON(ret);
3653         if (refs > 1)
3654                 goto out;
3655
3656         while (*level >= 0) {
3657                 cur = path->nodes[*level];
3658                 if (*level == 0) {
3659                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3660                         BUG_ON(ret);
3661                         clean_tree_block(trans, root, cur);
3662                         break;
3663                 }
3664                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3665                         clean_tree_block(trans, root, cur);
3666                         break;
3667                 }
3668
3669                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3670                 blocksize = btrfs_level_size(root, *level - 1);
3671                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3672
3673                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3674                 btrfs_tree_lock(next);
3675
3676                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3677                                               &refs);
3678                 BUG_ON(ret);
3679                 if (refs > 1) {
3680                         parent = path->nodes[*level];
3681                         ret = btrfs_free_extent(trans, root, bytenr,
3682                                         blocksize, parent->start,
3683                                         btrfs_header_owner(parent),
3684                                         btrfs_header_generation(parent),
3685                                         *level - 1, 1);
3686                         BUG_ON(ret);
3687                         path->slots[*level]++;
3688                         btrfs_tree_unlock(next);
3689                         free_extent_buffer(next);
3690                         continue;
3691                 }
3692
3693                 *level = btrfs_header_level(next);
3694                 path->nodes[*level] = next;
3695                 path->slots[*level] = 0;
3696                 path->locks[*level] = 1;
3697                 cond_resched();
3698         }
3699 out:
3700         parent = path->nodes[*level + 1];
3701         bytenr = path->nodes[*level]->start;
3702         blocksize = path->nodes[*level]->len;
3703
3704         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3705                         parent->start, btrfs_header_owner(parent),
3706                         btrfs_header_generation(parent), *level, 1);
3707         BUG_ON(ret);
3708
3709         if (path->locks[*level]) {
3710                 btrfs_tree_unlock(path->nodes[*level]);
3711                 path->locks[*level] = 0;
3712         }
3713         free_extent_buffer(path->nodes[*level]);
3714         path->nodes[*level] = NULL;
3715         *level += 1;
3716         cond_resched();
3717         return 0;
3718 }
3719
3720 /*
3721  * helper for dropping snapshots.  This walks back up the tree in the path
3722  * to find the first node higher up where we haven't yet gone through
3723  * all the slots
3724  */
3725 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3726                                  struct btrfs_root *root,
3727                                  struct btrfs_path *path,
3728                                  int *level, int max_level)
3729 {
3730         u64 root_owner;
3731         u64 root_gen;
3732         struct btrfs_root_item *root_item = &root->root_item;
3733         int i;
3734         int slot;
3735         int ret;
3736
3737         for (i = *level; i < max_level && path->nodes[i]; i++) {
3738                 slot = path->slots[i];
3739                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3740                         struct extent_buffer *node;
3741                         struct btrfs_disk_key disk_key;
3742                         node = path->nodes[i];
3743                         path->slots[i]++;
3744                         *level = i;
3745                         WARN_ON(*level == 0);
3746                         btrfs_node_key(node, &disk_key, path->slots[i]);
3747                         memcpy(&root_item->drop_progress,
3748                                &disk_key, sizeof(disk_key));
3749                         root_item->drop_level = i;
3750                         return 0;
3751                 } else {
3752                         struct extent_buffer *parent;
3753                         if (path->nodes[*level] == root->node)
3754                                 parent = path->nodes[*level];
3755                         else
3756                                 parent = path->nodes[*level + 1];
3757
3758                         root_owner = btrfs_header_owner(parent);
3759                         root_gen = btrfs_header_generation(parent);
3760
3761                         clean_tree_block(trans, root, path->nodes[*level]);
3762                         ret = btrfs_free_extent(trans, root,
3763                                                 path->nodes[*level]->start,
3764                                                 path->nodes[*level]->len,
3765                                                 parent->start, root_owner,
3766                                                 root_gen, *level, 1);
3767                         BUG_ON(ret);
3768                         if (path->locks[*level]) {
3769                                 btrfs_tree_unlock(path->nodes[*level]);
3770                                 path->locks[*level] = 0;
3771                         }
3772                         free_extent_buffer(path->nodes[*level]);
3773                         path->nodes[*level] = NULL;
3774                         *level = i + 1;
3775                 }
3776         }
3777         return 1;
3778 }
3779
3780 /*
3781  * drop the reference count on the tree rooted at 'snap'.  This traverses
3782  * the tree freeing any blocks that have a ref count of zero after being
3783  * decremented.
3784  */
3785 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3786                         *root)
3787 {
3788         int ret = 0;
3789         int wret;
3790         int level;
3791         struct btrfs_path *path;
3792         int i;
3793         int orig_level;
3794         struct btrfs_root_item *root_item = &root->root_item;
3795
3796         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3797         path = btrfs_alloc_path();
3798         BUG_ON(!path);
3799
3800         level = btrfs_header_level(root->node);
3801         orig_level = level;
3802         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3803                 path->nodes[level] = root->node;
3804                 extent_buffer_get(root->node);
3805                 path->slots[level] = 0;
3806         } else {
3807                 struct btrfs_key key;
3808                 struct btrfs_disk_key found_key;
3809                 struct extent_buffer *node;
3810
3811                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3812                 level = root_item->drop_level;
3813                 path->lowest_level = level;
3814                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3815                 if (wret < 0) {
3816                         ret = wret;
3817                         goto out;
3818                 }
3819                 node = path->nodes[level];
3820                 btrfs_node_key(node, &found_key, path->slots[level]);
3821                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3822                                sizeof(found_key)));
3823                 /*
3824                  * unlock our path, this is safe because only this
3825                  * function is allowed to delete this snapshot
3826                  */
3827                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3828                         if (path->nodes[i] && path->locks[i]) {
3829                                 path->locks[i] = 0;
3830                                 btrfs_tree_unlock(path->nodes[i]);
3831                         }
3832                 }
3833         }
3834         while(1) {
3835                 wret = walk_down_tree(trans, root, path, &level);
3836                 if (wret > 0)
3837                         break;
3838                 if (wret < 0)
3839                         ret = wret;
3840
3841                 wret = walk_up_tree(trans, root, path, &level,
3842                                     BTRFS_MAX_LEVEL);
3843                 if (wret > 0)
3844                         break;
3845                 if (wret < 0)
3846                         ret = wret;
3847                 if (trans->transaction->in_commit) {
3848                         ret = -EAGAIN;
3849                         break;
3850                 }
3851                 atomic_inc(&root->fs_info->throttle_gen);
3852                 wake_up(&root->fs_info->transaction_throttle);
3853         }
3854         for (i = 0; i <= orig_level; i++) {
3855                 if (path->nodes[i]) {
3856                         free_extent_buffer(path->nodes[i]);
3857                         path->nodes[i] = NULL;
3858                 }
3859         }
3860 out:
3861         btrfs_free_path(path);
3862         return ret;
3863 }
3864
3865 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3866                         struct btrfs_root *root,
3867                         struct extent_buffer *node,
3868                         struct extent_buffer *parent)
3869 {
3870         struct btrfs_path *path;
3871         int level;
3872         int parent_level;
3873         int ret = 0;
3874         int wret;
3875
3876         path = btrfs_alloc_path();
3877         BUG_ON(!path);
3878
3879         BUG_ON(!btrfs_tree_locked(parent));
3880         parent_level = btrfs_header_level(parent);
3881         extent_buffer_get(parent);
3882         path->nodes[parent_level] = parent;
3883         path->slots[parent_level] = btrfs_header_nritems(parent);
3884
3885         BUG_ON(!btrfs_tree_locked(node));
3886         level = btrfs_header_level(node);
3887         extent_buffer_get(node);
3888         path->nodes[level] = node;
3889         path->slots[level] = 0;
3890
3891         while (1) {
3892                 wret = walk_down_subtree(trans, root, path, &level);
3893                 if (wret < 0)
3894                         ret = wret;
3895                 if (wret != 0)
3896                         break;
3897
3898                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3899                 if (wret < 0)
3900                         ret = wret;
3901                 if (wret != 0)
3902                         break;
3903         }
3904
3905         btrfs_free_path(path);
3906         return ret;
3907 }
3908
3909 static unsigned long calc_ra(unsigned long start, unsigned long last,
3910                              unsigned long nr)
3911 {
3912         return min(last, start + nr - 1);
3913 }
3914
3915 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
3916                                          u64 len)
3917 {
3918         u64 page_start;
3919         u64 page_end;
3920         unsigned long first_index;
3921         unsigned long last_index;
3922         unsigned long i;
3923         struct page *page;
3924         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3925         struct file_ra_state *ra;
3926         struct btrfs_ordered_extent *ordered;
3927         unsigned int total_read = 0;
3928         unsigned int total_dirty = 0;
3929         int ret = 0;
3930
3931         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3932
3933         mutex_lock(&inode->i_mutex);
3934         first_index = start >> PAGE_CACHE_SHIFT;
3935         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3936
3937         /* make sure the dirty trick played by the caller work */
3938         ret = invalidate_inode_pages2_range(inode->i_mapping,
3939                                             first_index, last_index);
3940         if (ret)
3941                 goto out_unlock;
3942
3943         file_ra_state_init(ra, inode->i_mapping);
3944
3945         for (i = first_index ; i <= last_index; i++) {
3946                 if (total_read % ra->ra_pages == 0) {
3947                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3948                                        calc_ra(i, last_index, ra->ra_pages));
3949                 }
3950                 total_read++;
3951 again:
3952                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3953                         BUG_ON(1);
3954                 page = grab_cache_page(inode->i_mapping, i);
3955                 if (!page) {
3956                         ret = -ENOMEM;
3957                         goto out_unlock;
3958                 }
3959                 if (!PageUptodate(page)) {
3960                         btrfs_readpage(NULL, page);
3961                         lock_page(page);
3962                         if (!PageUptodate(page)) {
3963                                 unlock_page(page);
3964                                 page_cache_release(page);
3965                                 ret = -EIO;
3966                                 goto out_unlock;
3967                         }
3968                 }
3969                 wait_on_page_writeback(page);
3970
3971                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3972                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3973                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3974
3975                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3976                 if (ordered) {
3977                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3978                         unlock_page(page);
3979                         page_cache_release(page);
3980                         btrfs_start_ordered_extent(inode, ordered, 1);
3981                         btrfs_put_ordered_extent(ordered);
3982                         goto again;
3983                 }
3984                 set_page_extent_mapped(page);
3985
3986                 btrfs_set_extent_delalloc(inode, page_start, page_end);
3987                 if (i == first_index)
3988                         set_extent_bits(io_tree, page_start, page_end,
3989                                         EXTENT_BOUNDARY, GFP_NOFS);
3990
3991                 set_page_dirty(page);
3992                 total_dirty++;
3993
3994                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3995                 unlock_page(page);
3996                 page_cache_release(page);
3997         }
3998
3999 out_unlock:
4000         kfree(ra);
4001         mutex_unlock(&inode->i_mutex);
4002         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4003         return ret;
4004 }
4005
4006 static int noinline relocate_data_extent(struct inode *reloc_inode,
4007                                          struct btrfs_key *extent_key,
4008                                          u64 offset)
4009 {
4010         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4011         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4012         struct extent_map *em;
4013         u64 start = extent_key->objectid - offset;
4014         u64 end = start + extent_key->offset - 1;
4015
4016         em = alloc_extent_map(GFP_NOFS);
4017         BUG_ON(!em || IS_ERR(em));
4018
4019         em->start = start;
4020         em->len = extent_key->offset;
4021         em->block_len = extent_key->offset;
4022         em->block_start = extent_key->objectid;
4023         em->bdev = root->fs_info->fs_devices->latest_bdev;
4024         set_bit(EXTENT_FLAG_PINNED, &em->flags);
4025
4026         /* setup extent map to cheat btrfs_readpage */
4027         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4028         while (1) {
4029                 int ret;
4030                 spin_lock(&em_tree->lock);
4031                 ret = add_extent_mapping(em_tree, em);
4032                 spin_unlock(&em_tree->lock);
4033                 if (ret != -EEXIST) {
4034                         free_extent_map(em);
4035                         break;
4036                 }
4037                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4038         }
4039         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4040
4041         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4042 }
4043
4044 struct btrfs_ref_path {
4045         u64 extent_start;
4046         u64 nodes[BTRFS_MAX_LEVEL];
4047         u64 root_objectid;
4048         u64 root_generation;
4049         u64 owner_objectid;
4050         u32 num_refs;
4051         int lowest_level;
4052         int current_level;
4053         int shared_level;
4054
4055         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4056         u64 new_nodes[BTRFS_MAX_LEVEL];
4057 };
4058
4059 struct disk_extent {
4060         u64 ram_bytes;
4061         u64 disk_bytenr;
4062         u64 disk_num_bytes;
4063         u64 offset;
4064         u64 num_bytes;
4065         u8 compression;
4066         u8 encryption;
4067         u16 other_encoding;
4068 };
4069
4070 static int is_cowonly_root(u64 root_objectid)
4071 {
4072         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4073             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4074             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4075             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4076             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4077             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4078                 return 1;
4079         return 0;
4080 }
4081
4082 static int noinline __next_ref_path(struct btrfs_trans_handle *trans,
4083                                     struct btrfs_root *extent_root,
4084                                     struct btrfs_ref_path *ref_path,
4085                                     int first_time)
4086 {
4087         struct extent_buffer *leaf;
4088         struct btrfs_path *path;
4089         struct btrfs_extent_ref *ref;
4090         struct btrfs_key key;
4091         struct btrfs_key found_key;
4092         u64 bytenr;
4093         u32 nritems;
4094         int level;
4095         int ret = 1;
4096
4097         path = btrfs_alloc_path();
4098         if (!path)
4099                 return -ENOMEM;
4100
4101         if (first_time) {
4102                 ref_path->lowest_level = -1;
4103                 ref_path->current_level = -1;
4104                 ref_path->shared_level = -1;
4105                 goto walk_up;
4106         }
4107 walk_down:
4108         level = ref_path->current_level - 1;
4109         while (level >= -1) {
4110                 u64 parent;
4111                 if (level < ref_path->lowest_level)
4112                         break;
4113
4114                 if (level >= 0) {
4115                         bytenr = ref_path->nodes[level];
4116                 } else {
4117                         bytenr = ref_path->extent_start;
4118                 }
4119                 BUG_ON(bytenr == 0);
4120
4121                 parent = ref_path->nodes[level + 1];
4122                 ref_path->nodes[level + 1] = 0;
4123                 ref_path->current_level = level;
4124                 BUG_ON(parent == 0);
4125
4126                 key.objectid = bytenr;
4127                 key.offset = parent + 1;
4128                 key.type = BTRFS_EXTENT_REF_KEY;
4129
4130                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4131                 if (ret < 0)
4132                         goto out;
4133                 BUG_ON(ret == 0);
4134
4135                 leaf = path->nodes[0];
4136                 nritems = btrfs_header_nritems(leaf);
4137                 if (path->slots[0] >= nritems) {
4138                         ret = btrfs_next_leaf(extent_root, path);
4139                         if (ret < 0)
4140                                 goto out;
4141                         if (ret > 0)
4142                                 goto next;
4143                         leaf = path->nodes[0];
4144                 }
4145
4146                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4147                 if (found_key.objectid == bytenr &&
4148                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4149                         if (level < ref_path->shared_level)
4150                                 ref_path->shared_level = level;
4151                         goto found;
4152                 }
4153 next:
4154                 level--;
4155                 btrfs_release_path(extent_root, path);
4156                 cond_resched();
4157         }
4158         /* reached lowest level */
4159         ret = 1;
4160         goto out;
4161 walk_up:
4162         level = ref_path->current_level;
4163         while (level < BTRFS_MAX_LEVEL - 1) {
4164                 u64 ref_objectid;
4165                 if (level >= 0) {
4166                         bytenr = ref_path->nodes[level];
4167                 } else {
4168                         bytenr = ref_path->extent_start;
4169                 }
4170                 BUG_ON(bytenr == 0);
4171
4172                 key.objectid = bytenr;
4173                 key.offset = 0;
4174                 key.type = BTRFS_EXTENT_REF_KEY;
4175
4176                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4177                 if (ret < 0)
4178                         goto out;
4179
4180                 leaf = path->nodes[0];
4181                 nritems = btrfs_header_nritems(leaf);
4182                 if (path->slots[0] >= nritems) {
4183                         ret = btrfs_next_leaf(extent_root, path);
4184                         if (ret < 0)
4185                                 goto out;
4186                         if (ret > 0) {
4187                                 /* the extent was freed by someone */
4188                                 if (ref_path->lowest_level == level)
4189                                         goto out;
4190                                 btrfs_release_path(extent_root, path);
4191                                 goto walk_down;
4192                         }
4193                         leaf = path->nodes[0];
4194                 }
4195
4196                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4197                 if (found_key.objectid != bytenr ||
4198                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4199                         /* the extent was freed by someone */
4200                         if (ref_path->lowest_level == level) {
4201                                 ret = 1;
4202                                 goto out;
4203                         }
4204                         btrfs_release_path(extent_root, path);
4205                         goto walk_down;
4206                 }
4207 found:
4208                 ref = btrfs_item_ptr(leaf, path->slots[0],
4209                                 struct btrfs_extent_ref);
4210                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4211                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4212                         if (first_time) {
4213                                 level = (int)ref_objectid;
4214                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4215                                 ref_path->lowest_level = level;
4216                                 ref_path->current_level = level;
4217                                 ref_path->nodes[level] = bytenr;
4218                         } else {
4219                                 WARN_ON(ref_objectid != level);
4220                         }
4221                 } else {
4222                         WARN_ON(level != -1);
4223                 }
4224                 first_time = 0;
4225
4226                 if (ref_path->lowest_level == level) {
4227                         ref_path->owner_objectid = ref_objectid;
4228                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4229                 }
4230
4231                 /*
4232                  * the block is tree root or the block isn't in reference
4233                  * counted tree.
4234                  */
4235                 if (found_key.objectid == found_key.offset ||
4236                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4237                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4238                         ref_path->root_generation =
4239                                 btrfs_ref_generation(leaf, ref);
4240                         if (level < 0) {
4241                                 /* special reference from the tree log */
4242                                 ref_path->nodes[0] = found_key.offset;
4243                                 ref_path->current_level = 0;
4244                         }
4245                         ret = 0;
4246                         goto out;
4247                 }
4248
4249                 level++;
4250                 BUG_ON(ref_path->nodes[level] != 0);
4251                 ref_path->nodes[level] = found_key.offset;
4252                 ref_path->current_level = level;
4253
4254                 /*
4255                  * the reference was created in the running transaction,
4256                  * no need to continue walking up.
4257                  */
4258                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4259                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4260                         ref_path->root_generation =
4261                                 btrfs_ref_generation(leaf, ref);
4262                         ret = 0;
4263                         goto out;
4264                 }
4265
4266                 btrfs_release_path(extent_root, path);
4267                 cond_resched();
4268         }
4269         /* reached max tree level, but no tree root found. */
4270         BUG();
4271 out:
4272         btrfs_free_path(path);
4273         return ret;
4274 }
4275
4276 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4277                                 struct btrfs_root *extent_root,
4278                                 struct btrfs_ref_path *ref_path,
4279                                 u64 extent_start)
4280 {
4281         memset(ref_path, 0, sizeof(*ref_path));
4282         ref_path->extent_start = extent_start;
4283
4284         return __next_ref_path(trans, extent_root, ref_path, 1);
4285 }
4286
4287 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4288                                struct btrfs_root *extent_root,
4289                                struct btrfs_ref_path *ref_path)
4290 {
4291         return __next_ref_path(trans, extent_root, ref_path, 0);
4292 }
4293
4294 static int noinline get_new_locations(struct inode *reloc_inode,
4295                                       struct btrfs_key *extent_key,
4296                                       u64 offset, int no_fragment,
4297                                       struct disk_extent **extents,
4298                                       int *nr_extents)
4299 {
4300         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4301         struct btrfs_path *path;
4302         struct btrfs_file_extent_item *fi;
4303         struct extent_buffer *leaf;
4304         struct disk_extent *exts = *extents;
4305         struct btrfs_key found_key;
4306         u64 cur_pos;
4307         u64 last_byte;
4308         u32 nritems;
4309         int nr = 0;
4310         int max = *nr_extents;
4311         int ret;
4312
4313         WARN_ON(!no_fragment && *extents);
4314         if (!exts) {
4315                 max = 1;
4316                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4317                 if (!exts)
4318                         return -ENOMEM;
4319         }
4320
4321         path = btrfs_alloc_path();
4322         BUG_ON(!path);
4323
4324         cur_pos = extent_key->objectid - offset;
4325         last_byte = extent_key->objectid + extent_key->offset;
4326         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4327                                        cur_pos, 0);
4328         if (ret < 0)
4329                 goto out;
4330         if (ret > 0) {
4331                 ret = -ENOENT;
4332                 goto out;
4333         }
4334
4335         while (1) {
4336                 leaf = path->nodes[0];
4337                 nritems = btrfs_header_nritems(leaf);
4338                 if (path->slots[0] >= nritems) {
4339                         ret = btrfs_next_leaf(root, path);
4340                         if (ret < 0)
4341                                 goto out;
4342                         if (ret > 0)
4343                                 break;
4344                         leaf = path->nodes[0];
4345                 }
4346
4347                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4348                 if (found_key.offset != cur_pos ||
4349                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4350                     found_key.objectid != reloc_inode->i_ino)
4351                         break;
4352
4353                 fi = btrfs_item_ptr(leaf, path->slots[0],
4354                                     struct btrfs_file_extent_item);
4355                 if (btrfs_file_extent_type(leaf, fi) !=
4356                     BTRFS_FILE_EXTENT_REG ||
4357                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4358                         break;
4359
4360                 if (nr == max) {
4361                         struct disk_extent *old = exts;
4362                         max *= 2;
4363                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4364                         memcpy(exts, old, sizeof(*exts) * nr);
4365                         if (old != *extents)
4366                                 kfree(old);
4367                 }
4368
4369                 exts[nr].disk_bytenr =
4370                         btrfs_file_extent_disk_bytenr(leaf, fi);
4371                 exts[nr].disk_num_bytes =
4372                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4373                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4374                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4375                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4376                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4377                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4378                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4379                                                                            fi);
4380                 BUG_ON(exts[nr].offset > 0);
4381                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4382                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4383
4384                 cur_pos += exts[nr].num_bytes;
4385                 nr++;
4386
4387                 if (cur_pos + offset >= last_byte)
4388                         break;
4389
4390                 if (no_fragment) {
4391                         ret = 1;
4392                         goto out;
4393                 }
4394                 path->slots[0]++;
4395         }
4396
4397         WARN_ON(cur_pos + offset > last_byte);
4398         if (cur_pos + offset < last_byte) {
4399                 ret = -ENOENT;
4400                 goto out;
4401         }
4402         ret = 0;
4403 out:
4404         btrfs_free_path(path);
4405         if (ret) {
4406                 if (exts != *extents)
4407                         kfree(exts);
4408         } else {
4409                 *extents = exts;
4410                 *nr_extents = nr;
4411         }
4412         return ret;
4413 }
4414
4415 static int noinline replace_one_extent(struct btrfs_trans_handle *trans,
4416                                         struct btrfs_root *root,
4417                                         struct btrfs_path *path,
4418                                         struct btrfs_key *extent_key,
4419                                         struct btrfs_key *leaf_key,
4420                                         struct btrfs_ref_path *ref_path,
4421                                         struct disk_extent *new_extents,
4422                                         int nr_extents)
4423 {
4424         struct extent_buffer *leaf;
4425         struct btrfs_file_extent_item *fi;
4426         struct inode *inode = NULL;
4427         struct btrfs_key key;
4428         u64 lock_start = 0;
4429         u64 lock_end = 0;
4430         u64 num_bytes;
4431         u64 ext_offset;
4432         u64 first_pos;
4433         u32 nritems;
4434         int nr_scaned = 0;
4435         int extent_locked = 0;
4436         int extent_type;
4437         int ret;
4438
4439         memcpy(&key, leaf_key, sizeof(key));
4440         first_pos = INT_LIMIT(loff_t) - extent_key->offset;
4441         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4442                 if (key.objectid < ref_path->owner_objectid ||
4443                     (key.objectid == ref_path->owner_objectid &&
4444                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4445                         key.objectid = ref_path->owner_objectid;
4446                         key.type = BTRFS_EXTENT_DATA_KEY;
4447                         key.offset = 0;
4448                 }
4449         }
4450
4451         while (1) {
4452                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4453                 if (ret < 0)
4454                         goto out;
4455
4456                 leaf = path->nodes[0];
4457                 nritems = btrfs_header_nritems(leaf);
4458 next:
4459                 if (extent_locked && ret > 0) {
4460                         /*
4461                          * the file extent item was modified by someone
4462                          * before the extent got locked.
4463                          */
4464                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4465                                       lock_end, GFP_NOFS);
4466                         extent_locked = 0;
4467                 }
4468
4469                 if (path->slots[0] >= nritems) {
4470                         if (++nr_scaned > 2)
4471                                 break;
4472
4473                         BUG_ON(extent_locked);
4474                         ret = btrfs_next_leaf(root, path);
4475                         if (ret < 0)
4476                                 goto out;
4477                         if (ret > 0)
4478                                 break;
4479                         leaf = path->nodes[0];
4480                         nritems = btrfs_header_nritems(leaf);
4481                 }
4482
4483                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4484
4485                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4486                         if ((key.objectid > ref_path->owner_objectid) ||
4487                             (key.objectid == ref_path->owner_objectid &&
4488                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4489                             (key.offset >= first_pos + extent_key->offset))
4490                                 break;
4491                 }
4492
4493                 if (inode && key.objectid != inode->i_ino) {
4494                         BUG_ON(extent_locked);
4495                         btrfs_release_path(root, path);
4496                         mutex_unlock(&inode->i_mutex);
4497                         iput(inode);
4498                         inode = NULL;
4499                         continue;
4500                 }
4501
4502                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4503                         path->slots[0]++;
4504                         ret = 1;
4505                         goto next;
4506                 }
4507                 fi = btrfs_item_ptr(leaf, path->slots[0],
4508                                     struct btrfs_file_extent_item);
4509                 extent_type = btrfs_file_extent_type(leaf, fi);
4510                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4511                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4512                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4513                      extent_key->objectid)) {
4514                         path->slots[0]++;
4515                         ret = 1;
4516                         goto next;
4517                 }
4518
4519                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4520                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4521
4522                 if (first_pos > key.offset - ext_offset)
4523                         first_pos = key.offset - ext_offset;
4524
4525                 if (!extent_locked) {
4526                         lock_start = key.offset;
4527                         lock_end = lock_start + num_bytes - 1;
4528                 } else {
4529                         if (lock_start > key.offset ||
4530                             lock_end + 1 < key.offset + num_bytes) {
4531                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4532                                               lock_start, lock_end, GFP_NOFS);
4533                                 extent_locked = 0;
4534                         }
4535                 }
4536
4537                 if (!inode) {
4538                         btrfs_release_path(root, path);
4539
4540                         inode = btrfs_iget_locked(root->fs_info->sb,
4541                                                   key.objectid, root);
4542                         if (inode->i_state & I_NEW) {
4543                                 BTRFS_I(inode)->root = root;
4544                                 BTRFS_I(inode)->location.objectid =
4545                                         key.objectid;
4546                                 BTRFS_I(inode)->location.type =
4547                                         BTRFS_INODE_ITEM_KEY;
4548                                 BTRFS_I(inode)->location.offset = 0;
4549                                 btrfs_read_locked_inode(inode);
4550                                 unlock_new_inode(inode);
4551                         }
4552                         /*
4553                          * some code call btrfs_commit_transaction while
4554                          * holding the i_mutex, so we can't use mutex_lock
4555                          * here.
4556                          */
4557                         if (is_bad_inode(inode) ||
4558                             !mutex_trylock(&inode->i_mutex)) {
4559                                 iput(inode);
4560                                 inode = NULL;
4561                                 key.offset = (u64)-1;
4562                                 goto skip;
4563                         }
4564                 }
4565
4566                 if (!extent_locked) {
4567                         struct btrfs_ordered_extent *ordered;
4568
4569                         btrfs_release_path(root, path);
4570
4571                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4572                                     lock_end, GFP_NOFS);
4573                         ordered = btrfs_lookup_first_ordered_extent(inode,
4574                                                                     lock_end);
4575                         if (ordered &&
4576                             ordered->file_offset <= lock_end &&
4577                             ordered->file_offset + ordered->len > lock_start) {
4578                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4579                                               lock_start, lock_end, GFP_NOFS);
4580                                 btrfs_start_ordered_extent(inode, ordered, 1);
4581                                 btrfs_put_ordered_extent(ordered);
4582                                 key.offset += num_bytes;
4583                                 goto skip;
4584                         }
4585                         if (ordered)
4586                                 btrfs_put_ordered_extent(ordered);
4587
4588                         extent_locked = 1;
4589                         continue;
4590                 }
4591
4592                 if (nr_extents == 1) {
4593                         /* update extent pointer in place */
4594                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4595                                                 new_extents[0].disk_bytenr);
4596                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4597                                                 new_extents[0].disk_num_bytes);
4598                         btrfs_mark_buffer_dirty(leaf);
4599
4600                         btrfs_drop_extent_cache(inode, key.offset,
4601                                                 key.offset + num_bytes - 1, 0);
4602
4603                         ret = btrfs_inc_extent_ref(trans, root,
4604                                                 new_extents[0].disk_bytenr,
4605                                                 new_extents[0].disk_num_bytes,
4606                                                 leaf->start,
4607                                                 root->root_key.objectid,
4608                                                 trans->transid,
4609                                                 key.objectid);
4610                         BUG_ON(ret);
4611
4612                         ret = btrfs_free_extent(trans, root,
4613                                                 extent_key->objectid,
4614                                                 extent_key->offset,
4615                                                 leaf->start,
4616                                                 btrfs_header_owner(leaf),
4617                                                 btrfs_header_generation(leaf),
4618                                                 key.objectid, 0);
4619                         BUG_ON(ret);
4620
4621                         btrfs_release_path(root, path);
4622                         key.offset += num_bytes;
4623                 } else {
4624                         BUG_ON(1);
4625 #if 0
4626                         u64 alloc_hint;
4627                         u64 extent_len;
4628                         int i;
4629                         /*
4630                          * drop old extent pointer at first, then insert the
4631                          * new pointers one bye one
4632                          */
4633                         btrfs_release_path(root, path);
4634                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4635                                                  key.offset + num_bytes,
4636                                                  key.offset, &alloc_hint);
4637                         BUG_ON(ret);
4638
4639                         for (i = 0; i < nr_extents; i++) {
4640                                 if (ext_offset >= new_extents[i].num_bytes) {
4641                                         ext_offset -= new_extents[i].num_bytes;
4642                                         continue;
4643                                 }
4644                                 extent_len = min(new_extents[i].num_bytes -
4645                                                  ext_offset, num_bytes);
4646
4647                                 ret = btrfs_insert_empty_item(trans, root,
4648                                                               path, &key,
4649                                                               sizeof(*fi));
4650                                 BUG_ON(ret);
4651
4652                                 leaf = path->nodes[0];
4653                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4654                                                 struct btrfs_file_extent_item);
4655                                 btrfs_set_file_extent_generation(leaf, fi,
4656                                                         trans->transid);
4657                                 btrfs_set_file_extent_type(leaf, fi,
4658                                                         BTRFS_FILE_EXTENT_REG);
4659                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4660                                                 new_extents[i].disk_bytenr);
4661                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4662                                                 new_extents[i].disk_num_bytes);
4663                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4664                                                 new_extents[i].ram_bytes);
4665
4666                                 btrfs_set_file_extent_compression(leaf, fi,
4667                                                 new_extents[i].compression);
4668                                 btrfs_set_file_extent_encryption(leaf, fi,
4669                                                 new_extents[i].encryption);
4670                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4671                                                 new_extents[i].other_encoding);
4672
4673                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4674                                                         extent_len);
4675                                 ext_offset += new_extents[i].offset;
4676                                 btrfs_set_file_extent_offset(leaf, fi,
4677                                                         ext_offset);
4678                                 btrfs_mark_buffer_dirty(leaf);
4679
4680                                 btrfs_drop_extent_cache(inode, key.offset,
4681                                                 key.offset + extent_len - 1, 0);
4682
4683                                 ret = btrfs_inc_extent_ref(trans, root,
4684                                                 new_extents[i].disk_bytenr,
4685                                                 new_extents[i].disk_num_bytes,
4686                                                 leaf->start,
4687                                                 root->root_key.objectid,
4688                                                 trans->transid, key.objectid);
4689                                 BUG_ON(ret);
4690                                 btrfs_release_path(root, path);
4691
4692                                 inode_add_bytes(inode, extent_len);
4693
4694                                 ext_offset = 0;
4695                                 num_bytes -= extent_len;
4696                                 key.offset += extent_len;
4697
4698                                 if (num_bytes == 0)
4699                                         break;
4700                         }
4701                         BUG_ON(i >= nr_extents);
4702 #endif
4703                 }
4704
4705                 if (extent_locked) {
4706                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4707                                       lock_end, GFP_NOFS);
4708                         extent_locked = 0;
4709                 }
4710 skip:
4711                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4712                     key.offset >= first_pos + extent_key->offset)
4713                         break;
4714
4715                 cond_resched();
4716         }
4717         ret = 0;
4718 out:
4719         btrfs_release_path(root, path);
4720         if (inode) {
4721                 mutex_unlock(&inode->i_mutex);
4722                 if (extent_locked) {
4723                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4724                                       lock_end, GFP_NOFS);
4725                 }
4726                 iput(inode);
4727         }
4728         return ret;
4729 }
4730
4731 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4732                                struct btrfs_root *root,
4733                                struct extent_buffer *buf, u64 orig_start)
4734 {
4735         int level;
4736         int ret;
4737
4738         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4739         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4740
4741         level = btrfs_header_level(buf);
4742         if (level == 0) {
4743                 struct btrfs_leaf_ref *ref;
4744                 struct btrfs_leaf_ref *orig_ref;
4745
4746                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4747                 if (!orig_ref)
4748                         return -ENOENT;
4749
4750                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4751                 if (!ref) {
4752                         btrfs_free_leaf_ref(root, orig_ref);
4753                         return -ENOMEM;
4754                 }
4755
4756                 ref->nritems = orig_ref->nritems;
4757                 memcpy(ref->extents, orig_ref->extents,
4758                         sizeof(ref->extents[0]) * ref->nritems);
4759
4760                 btrfs_free_leaf_ref(root, orig_ref);
4761
4762                 ref->root_gen = trans->transid;
4763                 ref->bytenr = buf->start;
4764                 ref->owner = btrfs_header_owner(buf);
4765                 ref->generation = btrfs_header_generation(buf);
4766                 ret = btrfs_add_leaf_ref(root, ref, 0);
4767                 WARN_ON(ret);
4768                 btrfs_free_leaf_ref(root, ref);
4769         }
4770         return 0;
4771 }
4772
4773 static int noinline invalidate_extent_cache(struct btrfs_root *root,
4774                                         struct extent_buffer *leaf,
4775                                         struct btrfs_block_group_cache *group,
4776                                         struct btrfs_root *target_root)
4777 {
4778         struct btrfs_key key;
4779         struct inode *inode = NULL;
4780         struct btrfs_file_extent_item *fi;
4781         u64 num_bytes;
4782         u64 skip_objectid = 0;
4783         u32 nritems;
4784         u32 i;
4785
4786         nritems = btrfs_header_nritems(leaf);
4787         for (i = 0; i < nritems; i++) {
4788                 btrfs_item_key_to_cpu(leaf, &key, i);
4789                 if (key.objectid == skip_objectid ||
4790                     key.type != BTRFS_EXTENT_DATA_KEY)
4791                         continue;
4792                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4793                 if (btrfs_file_extent_type(leaf, fi) ==
4794                     BTRFS_FILE_EXTENT_INLINE)
4795                         continue;
4796                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4797                         continue;
4798                 if (!inode || inode->i_ino != key.objectid) {
4799                         iput(inode);
4800                         inode = btrfs_ilookup(target_root->fs_info->sb,
4801                                               key.objectid, target_root, 1);
4802                 }
4803                 if (!inode) {
4804                         skip_objectid = key.objectid;
4805                         continue;
4806                 }
4807                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4808
4809                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4810                             key.offset + num_bytes - 1, GFP_NOFS);
4811                 btrfs_drop_extent_cache(inode, key.offset,
4812                                         key.offset + num_bytes - 1, 1);
4813                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4814                               key.offset + num_bytes - 1, GFP_NOFS);
4815                 cond_resched();
4816         }
4817         iput(inode);
4818         return 0;
4819 }
4820
4821 static int noinline replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4822                                         struct btrfs_root *root,
4823                                         struct extent_buffer *leaf,
4824                                         struct btrfs_block_group_cache *group,
4825                                         struct inode *reloc_inode)
4826 {
4827         struct btrfs_key key;
4828         struct btrfs_key extent_key;
4829         struct btrfs_file_extent_item *fi;
4830         struct btrfs_leaf_ref *ref;
4831         struct disk_extent *new_extent;
4832         u64 bytenr;
4833         u64 num_bytes;
4834         u32 nritems;
4835         u32 i;
4836         int ext_index;
4837         int nr_extent;
4838         int ret;
4839
4840         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4841         BUG_ON(!new_extent);
4842
4843         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4844         BUG_ON(!ref);
4845
4846         ext_index = -1;
4847         nritems = btrfs_header_nritems(leaf);
4848         for (i = 0; i < nritems; i++) {
4849                 btrfs_item_key_to_cpu(leaf, &key, i);
4850                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4851                         continue;
4852                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4853                 if (btrfs_file_extent_type(leaf, fi) ==
4854                     BTRFS_FILE_EXTENT_INLINE)
4855                         continue;
4856                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4857                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4858                 if (bytenr == 0)
4859                         continue;
4860
4861                 ext_index++;
4862                 if (bytenr >= group->key.objectid + group->key.offset ||
4863                     bytenr + num_bytes <= group->key.objectid)
4864                         continue;
4865
4866                 extent_key.objectid = bytenr;
4867                 extent_key.offset = num_bytes;
4868                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4869                 nr_extent = 1;
4870                 ret = get_new_locations(reloc_inode, &extent_key,
4871                                         group->key.objectid, 1,
4872                                         &new_extent, &nr_extent);
4873                 if (ret > 0)
4874                         continue;
4875                 BUG_ON(ret < 0);
4876
4877                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4878                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4879                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4880                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4881
4882                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4883                                                 new_extent->disk_bytenr);
4884                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4885                                                 new_extent->disk_num_bytes);
4886                 btrfs_mark_buffer_dirty(leaf);
4887
4888                 ret = btrfs_inc_extent_ref(trans, root,
4889                                         new_extent->disk_bytenr,
4890                                         new_extent->disk_num_bytes,
4891                                         leaf->start,
4892                                         root->root_key.objectid,
4893                                         trans->transid, key.objectid);
4894                 BUG_ON(ret);
4895                 ret = btrfs_free_extent(trans, root,
4896                                         bytenr, num_bytes, leaf->start,
4897                                         btrfs_header_owner(leaf),
4898                                         btrfs_header_generation(leaf),
4899                                         key.objectid, 0);
4900                 BUG_ON(ret);
4901                 cond_resched();
4902         }
4903         kfree(new_extent);
4904         BUG_ON(ext_index + 1 != ref->nritems);
4905         btrfs_free_leaf_ref(root, ref);
4906         return 0;
4907 }
4908
4909 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4910                           struct btrfs_root *root)
4911 {
4912         struct btrfs_root *reloc_root;
4913         int ret;
4914
4915         if (root->reloc_root) {
4916                 reloc_root = root->reloc_root;
4917                 root->reloc_root = NULL;
4918                 list_add(&reloc_root->dead_list,
4919                          &root->fs_info->dead_reloc_roots);
4920
4921                 btrfs_set_root_bytenr(&reloc_root->root_item,
4922                                       reloc_root->node->start);
4923                 btrfs_set_root_level(&root->root_item,
4924                                      btrfs_header_level(reloc_root->node));
4925                 memset(&reloc_root->root_item.drop_progress, 0,
4926                         sizeof(struct btrfs_disk_key));
4927                 reloc_root->root_item.drop_level = 0;
4928
4929                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4930                                         &reloc_root->root_key,
4931                                         &reloc_root->root_item);
4932                 BUG_ON(ret);
4933         }
4934         return 0;
4935 }
4936
4937 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4938 {
4939         struct btrfs_trans_handle *trans;
4940         struct btrfs_root *reloc_root;
4941         struct btrfs_root *prev_root = NULL;
4942         struct list_head dead_roots;
4943         int ret;
4944         unsigned long nr;
4945
4946         INIT_LIST_HEAD(&dead_roots);
4947         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4948
4949         while (!list_empty(&dead_roots)) {
4950                 reloc_root = list_entry(dead_roots.prev,
4951                                         struct btrfs_root, dead_list);
4952                 list_del_init(&reloc_root->dead_list);
4953
4954                 BUG_ON(reloc_root->commit_root != NULL);
4955                 while (1) {
4956                         trans = btrfs_join_transaction(root, 1);
4957                         BUG_ON(!trans);
4958
4959                         mutex_lock(&root->fs_info->drop_mutex);
4960                         ret = btrfs_drop_snapshot(trans, reloc_root);
4961                         if (ret != -EAGAIN)
4962                                 break;
4963                         mutex_unlock(&root->fs_info->drop_mutex);
4964
4965                         nr = trans->blocks_used;
4966                         ret = btrfs_end_transaction(trans, root);
4967                         BUG_ON(ret);
4968                         btrfs_btree_balance_dirty(root, nr);
4969                 }
4970
4971                 free_extent_buffer(reloc_root->node);
4972
4973                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4974                                      &reloc_root->root_key);
4975                 BUG_ON(ret);
4976                 mutex_unlock(&root->fs_info->drop_mutex);
4977
4978                 nr = trans->blocks_used;
4979                 ret = btrfs_end_transaction(trans, root);
4980                 BUG_ON(ret);
4981                 btrfs_btree_balance_dirty(root, nr);
4982
4983                 kfree(prev_root);
4984                 prev_root = reloc_root;
4985         }
4986         if (prev_root) {
4987                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4988                 kfree(prev_root);
4989         }
4990         return 0;
4991 }
4992
4993 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
4994 {
4995         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
4996         return 0;
4997 }
4998
4999 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5000 {
5001         struct btrfs_root *reloc_root;
5002         struct btrfs_trans_handle *trans;
5003         struct btrfs_key location;
5004         int found;
5005         int ret;
5006
5007         mutex_lock(&root->fs_info->tree_reloc_mutex);
5008         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5009         BUG_ON(ret);
5010         found = !list_empty(&root->fs_info->dead_reloc_roots);
5011         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5012
5013         if (found) {
5014                 trans = btrfs_start_transaction(root, 1);
5015                 BUG_ON(!trans);
5016                 ret = btrfs_commit_transaction(trans, root);
5017                 BUG_ON(ret);
5018         }
5019
5020         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5021         location.offset = (u64)-1;
5022         location.type = BTRFS_ROOT_ITEM_KEY;
5023
5024         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5025         BUG_ON(!reloc_root);
5026         btrfs_orphan_cleanup(reloc_root);
5027         return 0;
5028 }
5029
5030 static int noinline init_reloc_tree(struct btrfs_trans_handle *trans,
5031                                     struct btrfs_root *root)
5032 {
5033         struct btrfs_root *reloc_root;
5034         struct extent_buffer *eb;
5035         struct btrfs_root_item *root_item;
5036         struct btrfs_key root_key;
5037         int ret;
5038
5039         BUG_ON(!root->ref_cows);
5040         if (root->reloc_root)
5041                 return 0;
5042
5043         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5044         BUG_ON(!root_item);
5045
5046         ret = btrfs_copy_root(trans, root, root->commit_root,
5047                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5048         BUG_ON(ret);
5049
5050         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5051         root_key.offset = root->root_key.objectid;
5052         root_key.type = BTRFS_ROOT_ITEM_KEY;
5053
5054         memcpy(root_item, &root->root_item, sizeof(root_item));
5055         btrfs_set_root_refs(root_item, 0);
5056         btrfs_set_root_bytenr(root_item, eb->start);
5057         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5058         btrfs_set_root_generation(root_item, trans->transid);
5059
5060         btrfs_tree_unlock(eb);
5061         free_extent_buffer(eb);
5062
5063         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5064                                 &root_key, root_item);
5065         BUG_ON(ret);
5066         kfree(root_item);
5067
5068         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5069                                                  &root_key);
5070         BUG_ON(!reloc_root);
5071         reloc_root->last_trans = trans->transid;
5072         reloc_root->commit_root = NULL;
5073         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5074
5075         root->reloc_root = reloc_root;
5076         return 0;
5077 }
5078
5079 /*
5080  * Core function of space balance.
5081  *
5082  * The idea is using reloc trees to relocate tree blocks in reference
5083  * counted roots. There is one reloc tree for each subvol, and all
5084  * reloc trees share same root key objectid. Reloc trees are snapshots
5085  * of the latest committed roots of subvols (root->commit_root).
5086  *
5087  * To relocate a tree block referenced by a subvol, there are two steps.
5088  * COW the block through subvol's reloc tree, then update block pointer
5089  * in the subvol to point to the new block. Since all reloc trees share
5090  * same root key objectid, doing special handing for tree blocks owned
5091  * by them is easy. Once a tree block has been COWed in one reloc tree,
5092  * we can use the resulting new block directly when the same block is
5093  * required to COW again through other reloc trees. By this way, relocated
5094  * tree blocks are shared between reloc trees, so they are also shared
5095  * between subvols.
5096  */
5097 static int noinline relocate_one_path(struct btrfs_trans_handle *trans,
5098                                       struct btrfs_root *root,
5099                                       struct btrfs_path *path,
5100                                       struct btrfs_key *first_key,
5101                                       struct btrfs_ref_path *ref_path,
5102                                       struct btrfs_block_group_cache *group,
5103                                       struct inode *reloc_inode)
5104 {
5105         struct btrfs_root *reloc_root;
5106         struct extent_buffer *eb = NULL;
5107         struct btrfs_key *keys;
5108         u64 *nodes;
5109         int level;
5110         int shared_level;
5111         int lowest_level = 0;
5112         int ret;
5113
5114         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5115                 lowest_level = ref_path->owner_objectid;
5116
5117         if (!root->ref_cows) {
5118                 path->lowest_level = lowest_level;
5119                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5120                 BUG_ON(ret < 0);
5121                 path->lowest_level = 0;
5122                 btrfs_release_path(root, path);
5123                 return 0;
5124         }
5125
5126         mutex_lock(&root->fs_info->tree_reloc_mutex);
5127         ret = init_reloc_tree(trans, root);
5128         BUG_ON(ret);
5129         reloc_root = root->reloc_root;
5130
5131         shared_level = ref_path->shared_level;
5132         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5133
5134         keys = ref_path->node_keys;
5135         nodes = ref_path->new_nodes;
5136         memset(&keys[shared_level + 1], 0,
5137                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5138         memset(&nodes[shared_level + 1], 0,
5139                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5140
5141         if (nodes[lowest_level] == 0) {
5142                 path->lowest_level = lowest_level;
5143                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5144                                         0, 1);
5145                 BUG_ON(ret);
5146                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5147                         eb = path->nodes[level];
5148                         if (!eb || eb == reloc_root->node)
5149                                 break;
5150                         nodes[level] = eb->start;
5151                         if (level == 0)
5152                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5153                         else
5154                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5155                 }
5156                 if (nodes[0] &&
5157                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5158                         eb = path->nodes[0];
5159                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5160                                                       group, reloc_inode);
5161                         BUG_ON(ret);
5162                 }
5163                 btrfs_release_path(reloc_root, path);
5164         } else {
5165                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5166                                        lowest_level);
5167                 BUG_ON(ret);
5168         }
5169
5170         /*
5171          * replace tree blocks in the fs tree with tree blocks in
5172          * the reloc tree.
5173          */
5174         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5175         BUG_ON(ret < 0);
5176
5177         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5178                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5179                                         0, 0);
5180                 BUG_ON(ret);
5181                 extent_buffer_get(path->nodes[0]);
5182                 eb = path->nodes[0];
5183                 btrfs_release_path(reloc_root, path);
5184                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5185                 BUG_ON(ret);
5186                 free_extent_buffer(eb);
5187         }
5188
5189         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5190         path->lowest_level = 0;
5191         return 0;
5192 }
5193
5194 static int noinline relocate_tree_block(struct btrfs_trans_handle *trans,
5195                                         struct btrfs_root *root,
5196                                         struct btrfs_path *path,
5197                                         struct btrfs_key *first_key,
5198                                         struct btrfs_ref_path *ref_path)
5199 {
5200         int ret;
5201
5202         ret = relocate_one_path(trans, root, path, first_key,
5203                                 ref_path, NULL, NULL);
5204         BUG_ON(ret);
5205
5206         if (root == root->fs_info->extent_root)
5207                 btrfs_extent_post_op(trans, root);
5208
5209         return 0;
5210 }
5211
5212 static int noinline del_extent_zero(struct btrfs_trans_handle *trans,
5213                                     struct btrfs_root *extent_root,
5214                                     struct btrfs_path *path,
5215                                     struct btrfs_key *extent_key)
5216 {
5217         int ret;
5218
5219         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5220         if (ret)
5221                 goto out;
5222         ret = btrfs_del_item(trans, extent_root, path);
5223 out:
5224         btrfs_release_path(extent_root, path);
5225         return ret;
5226 }
5227
5228 static struct btrfs_root noinline *read_ref_root(struct btrfs_fs_info *fs_info,
5229                                                 struct btrfs_ref_path *ref_path)
5230 {
5231         struct btrfs_key root_key;
5232
5233         root_key.objectid = ref_path->root_objectid;
5234         root_key.type = BTRFS_ROOT_ITEM_KEY;
5235         if (is_cowonly_root(ref_path->root_objectid))
5236                 root_key.offset = 0;
5237         else
5238                 root_key.offset = (u64)-1;
5239
5240         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5241 }
5242
5243 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
5244                                         struct btrfs_path *path,
5245                                         struct btrfs_key *extent_key,
5246                                         struct btrfs_block_group_cache *group,
5247                                         struct inode *reloc_inode, int pass)
5248 {
5249         struct btrfs_trans_handle *trans;
5250         struct btrfs_root *found_root;
5251         struct btrfs_ref_path *ref_path = NULL;
5252         struct disk_extent *new_extents = NULL;
5253         int nr_extents = 0;
5254         int loops;
5255         int ret;
5256         int level;
5257         struct btrfs_key first_key;
5258         u64 prev_block = 0;
5259
5260
5261         trans = btrfs_start_transaction(extent_root, 1);
5262         BUG_ON(!trans);
5263
5264         if (extent_key->objectid == 0) {
5265                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5266                 goto out;
5267         }
5268
5269         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5270         if (!ref_path) {
5271                ret = -ENOMEM;
5272                goto out;
5273         }
5274
5275         for (loops = 0; ; loops++) {
5276                 if (loops == 0) {
5277                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5278                                                    extent_key->objectid);
5279                 } else {
5280                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5281                 }
5282                 if (ret < 0)
5283                         goto out;
5284                 if (ret > 0)
5285                         break;
5286
5287                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5288                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5289                         continue;
5290
5291                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5292                 BUG_ON(!found_root);
5293                 /*
5294                  * for reference counted tree, only process reference paths
5295                  * rooted at the latest committed root.
5296                  */
5297                 if (found_root->ref_cows &&
5298                     ref_path->root_generation != found_root->root_key.offset)
5299                         continue;
5300
5301                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5302                         if (pass == 0) {
5303                                 /*
5304                                  * copy data extents to new locations
5305                                  */
5306                                 u64 group_start = group->key.objectid;
5307                                 ret = relocate_data_extent(reloc_inode,
5308                                                            extent_key,
5309                                                            group_start);
5310                                 if (ret < 0)
5311                                         goto out;
5312                                 break;
5313                         }
5314                         level = 0;
5315                 } else {
5316                         level = ref_path->owner_objectid;
5317                 }
5318
5319                 if (prev_block != ref_path->nodes[level]) {
5320                         struct extent_buffer *eb;
5321                         u64 block_start = ref_path->nodes[level];
5322                         u64 block_size = btrfs_level_size(found_root, level);
5323
5324                         eb = read_tree_block(found_root, block_start,
5325                                              block_size, 0);
5326                         btrfs_tree_lock(eb);
5327                         BUG_ON(level != btrfs_header_level(eb));
5328
5329                         if (level == 0)
5330                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5331                         else
5332                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5333
5334                         btrfs_tree_unlock(eb);
5335                         free_extent_buffer(eb);
5336                         prev_block = block_start;
5337                 }
5338
5339                 btrfs_record_root_in_trans(found_root);
5340                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5341                         /*
5342                          * try to update data extent references while
5343                          * keeping metadata shared between snapshots.
5344                          */
5345                         if (pass == 1) {
5346                                 ret = relocate_one_path(trans, found_root,
5347                                                 path, &first_key, ref_path,
5348                                                 group, reloc_inode);
5349                                 if (ret < 0)
5350                                         goto out;
5351                                 continue;
5352                         }
5353                         /*
5354                          * use fallback method to process the remaining
5355                          * references.
5356                          */
5357                         if (!new_extents) {
5358                                 u64 group_start = group->key.objectid;
5359                                 new_extents = kmalloc(sizeof(*new_extents),
5360                                                       GFP_NOFS);
5361                                 nr_extents = 1;
5362                                 ret = get_new_locations(reloc_inode,
5363                                                         extent_key,
5364                                                         group_start, 1,
5365                                                         &new_extents,
5366                                                         &nr_extents);
5367                                 if (ret)
5368                                         goto out;
5369                         }
5370                         btrfs_record_root_in_trans(found_root);
5371                         ret = replace_one_extent(trans, found_root,
5372                                                 path, extent_key,
5373                                                 &first_key, ref_path,
5374                                                 new_extents, nr_extents);
5375                 } else {
5376                         ret = relocate_tree_block(trans, found_root, path,
5377                                                   &first_key, ref_path);
5378                 }
5379                 if (ret < 0)
5380                         goto out;
5381         }
5382         ret = 0;
5383 out:
5384         btrfs_end_transaction(trans, extent_root);
5385         kfree(new_extents);
5386         kfree(ref_path);
5387         return ret;
5388 }
5389
5390 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5391 {
5392         u64 num_devices;
5393         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5394                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5395
5396         num_devices = root->fs_info->fs_devices->rw_devices;
5397         if (num_devices == 1) {
5398                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5399                 stripped = flags & ~stripped;
5400
5401                 /* turn raid0 into single device chunks */
5402                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5403                         return stripped;
5404
5405                 /* turn mirroring into duplication */
5406                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5407                              BTRFS_BLOCK_GROUP_RAID10))
5408                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5409                 return flags;
5410         } else {
5411                 /* they already had raid on here, just return */
5412                 if (flags & stripped)
5413                         return flags;
5414
5415                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5416                 stripped = flags & ~stripped;
5417
5418                 /* switch duplicated blocks with raid1 */
5419                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5420                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5421
5422                 /* turn single device chunks into raid0 */
5423                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5424         }
5425         return flags;
5426 }
5427
5428 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5429                      struct btrfs_block_group_cache *shrink_block_group,
5430                      int force)
5431 {
5432         struct btrfs_trans_handle *trans;
5433         u64 new_alloc_flags;
5434         u64 calc;
5435
5436         spin_lock(&shrink_block_group->lock);
5437         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5438                 spin_unlock(&shrink_block_group->lock);
5439
5440                 trans = btrfs_start_transaction(root, 1);
5441                 spin_lock(&shrink_block_group->lock);
5442
5443                 new_alloc_flags = update_block_group_flags(root,
5444                                                    shrink_block_group->flags);
5445                 if (new_alloc_flags != shrink_block_group->flags) {
5446                         calc =
5447                              btrfs_block_group_used(&shrink_block_group->item);
5448                 } else {
5449                         calc = shrink_block_group->key.offset;
5450                 }
5451                 spin_unlock(&shrink_block_group->lock);
5452
5453                 do_chunk_alloc(trans, root->fs_info->extent_root,
5454                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5455
5456                 btrfs_end_transaction(trans, root);
5457         } else
5458                 spin_unlock(&shrink_block_group->lock);
5459         return 0;
5460 }
5461
5462 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5463                                  struct btrfs_root *root,
5464                                  u64 objectid, u64 size)
5465 {
5466         struct btrfs_path *path;
5467         struct btrfs_inode_item *item;
5468         struct extent_buffer *leaf;
5469         int ret;
5470
5471         path = btrfs_alloc_path();
5472         if (!path)
5473                 return -ENOMEM;
5474
5475         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5476         if (ret)
5477                 goto out;
5478
5479         leaf = path->nodes[0];
5480         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5481         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5482         btrfs_set_inode_generation(leaf, item, 1);
5483         btrfs_set_inode_size(leaf, item, size);
5484         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5485         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5486         btrfs_mark_buffer_dirty(leaf);
5487         btrfs_release_path(root, path);
5488 out:
5489         btrfs_free_path(path);
5490         return ret;
5491 }
5492
5493 static struct inode noinline *create_reloc_inode(struct btrfs_fs_info *fs_info,
5494                                         struct btrfs_block_group_cache *group)
5495 {
5496         struct inode *inode = NULL;
5497         struct btrfs_trans_handle *trans;
5498         struct btrfs_root *root;
5499         struct btrfs_key root_key;
5500         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5501         int err = 0;
5502
5503         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5504         root_key.type = BTRFS_ROOT_ITEM_KEY;
5505         root_key.offset = (u64)-1;
5506         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5507         if (IS_ERR(root))
5508                 return ERR_CAST(root);
5509
5510         trans = btrfs_start_transaction(root, 1);
5511         BUG_ON(!trans);
5512
5513         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5514         if (err)
5515                 goto out;
5516
5517         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5518         BUG_ON(err);
5519
5520         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5521                                        group->key.offset, 0, group->key.offset,
5522                                        0, 0, 0);
5523         BUG_ON(err);
5524
5525         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5526         if (inode->i_state & I_NEW) {
5527                 BTRFS_I(inode)->root = root;
5528                 BTRFS_I(inode)->location.objectid = objectid;
5529                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5530                 BTRFS_I(inode)->location.offset = 0;
5531                 btrfs_read_locked_inode(inode);
5532                 unlock_new_inode(inode);
5533                 BUG_ON(is_bad_inode(inode));
5534         } else {
5535                 BUG_ON(1);
5536         }
5537
5538         err = btrfs_orphan_add(trans, inode);
5539 out:
5540         btrfs_end_transaction(trans, root);
5541         if (err) {
5542                 if (inode)
5543                         iput(inode);
5544                 inode = ERR_PTR(err);
5545         }
5546         return inode;
5547 }
5548
5549 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5550 {
5551         struct btrfs_trans_handle *trans;
5552         struct btrfs_path *path;
5553         struct btrfs_fs_info *info = root->fs_info;
5554         struct extent_buffer *leaf;
5555         struct inode *reloc_inode;
5556         struct btrfs_block_group_cache *block_group;
5557         struct btrfs_key key;
5558         u64 skipped;
5559         u64 cur_byte;
5560         u64 total_found;
5561         u32 nritems;
5562         int ret;
5563         int progress;
5564         int pass = 0;
5565
5566         root = root->fs_info->extent_root;
5567
5568         block_group = btrfs_lookup_block_group(info, group_start);
5569         BUG_ON(!block_group);
5570
5571         printk("btrfs relocating block group %llu flags %llu\n",
5572                (unsigned long long)block_group->key.objectid,
5573                (unsigned long long)block_group->flags);
5574
5575         path = btrfs_alloc_path();
5576         BUG_ON(!path);
5577
5578         reloc_inode = create_reloc_inode(info, block_group);
5579         BUG_ON(IS_ERR(reloc_inode));
5580
5581         __alloc_chunk_for_shrink(root, block_group, 1);
5582         set_block_group_readonly(block_group);
5583
5584         btrfs_start_delalloc_inodes(info->tree_root);
5585         btrfs_wait_ordered_extents(info->tree_root, 0);
5586 again:
5587         skipped = 0;
5588         total_found = 0;
5589         progress = 0;
5590         key.objectid = block_group->key.objectid;
5591         key.offset = 0;
5592         key.type = 0;
5593         cur_byte = key.objectid;
5594
5595         trans = btrfs_start_transaction(info->tree_root, 1);
5596         btrfs_commit_transaction(trans, info->tree_root);
5597
5598         mutex_lock(&root->fs_info->cleaner_mutex);
5599         btrfs_clean_old_snapshots(info->tree_root);
5600         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5601         mutex_unlock(&root->fs_info->cleaner_mutex);
5602
5603         while(1) {
5604                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5605                 if (ret < 0)
5606                         goto out;
5607 next:
5608                 leaf = path->nodes[0];
5609                 nritems = btrfs_header_nritems(leaf);
5610                 if (path->slots[0] >= nritems) {
5611                         ret = btrfs_next_leaf(root, path);
5612                         if (ret < 0)
5613                                 goto out;
5614                         if (ret == 1) {
5615                                 ret = 0;
5616                                 break;
5617                         }
5618                         leaf = path->nodes[0];
5619                         nritems = btrfs_header_nritems(leaf);
5620                 }
5621
5622                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5623
5624                 if (key.objectid >= block_group->key.objectid +
5625                     block_group->key.offset)
5626                         break;
5627
5628                 if (progress && need_resched()) {
5629                         btrfs_release_path(root, path);
5630                         cond_resched();
5631                         progress = 0;
5632                         continue;
5633                 }
5634                 progress = 1;
5635
5636                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5637                     key.objectid + key.offset <= cur_byte) {
5638                         path->slots[0]++;
5639                         goto next;
5640                 }
5641
5642                 total_found++;
5643                 cur_byte = key.objectid + key.offset;
5644                 btrfs_release_path(root, path);
5645
5646                 __alloc_chunk_for_shrink(root, block_group, 0);
5647                 ret = relocate_one_extent(root, path, &key, block_group,
5648                                           reloc_inode, pass);
5649                 BUG_ON(ret < 0);
5650                 if (ret > 0)
5651                         skipped++;
5652
5653                 key.objectid = cur_byte;
5654                 key.type = 0;
5655                 key.offset = 0;
5656         }
5657
5658         btrfs_release_path(root, path);
5659
5660         if (pass == 0) {
5661                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5662                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5663                 WARN_ON(reloc_inode->i_mapping->nrpages);
5664         }
5665
5666         if (total_found > 0) {
5667                 printk("btrfs found %llu extents in pass %d\n",
5668                        (unsigned long long)total_found, pass);
5669                 pass++;
5670                 if (total_found == skipped && pass > 2) {
5671                         iput(reloc_inode);
5672                         reloc_inode = create_reloc_inode(info, block_group);
5673                         pass = 0;
5674                 }
5675                 goto again;
5676         }
5677
5678         /* delete reloc_inode */
5679         iput(reloc_inode);
5680
5681         /* unpin extents in this range */
5682         trans = btrfs_start_transaction(info->tree_root, 1);
5683         btrfs_commit_transaction(trans, info->tree_root);
5684
5685         spin_lock(&block_group->lock);
5686         WARN_ON(block_group->pinned > 0);
5687         WARN_ON(block_group->reserved > 0);
5688         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5689         spin_unlock(&block_group->lock);
5690         put_block_group(block_group);
5691         ret = 0;
5692 out:
5693         btrfs_free_path(path);
5694         return ret;
5695 }
5696
5697 static int find_first_block_group(struct btrfs_root *root,
5698                 struct btrfs_path *path, struct btrfs_key *key)
5699 {
5700         int ret = 0;
5701         struct btrfs_key found_key;
5702         struct extent_buffer *leaf;
5703         int slot;
5704
5705         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5706         if (ret < 0)
5707                 goto out;
5708
5709         while(1) {
5710                 slot = path->slots[0];
5711                 leaf = path->nodes[0];
5712                 if (slot >= btrfs_header_nritems(leaf)) {
5713                         ret = btrfs_next_leaf(root, path);
5714                         if (ret == 0)
5715                                 continue;
5716                         if (ret < 0)
5717                                 goto out;
5718                         break;
5719                 }
5720                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5721
5722                 if (found_key.objectid >= key->objectid &&
5723                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5724                         ret = 0;
5725                         goto out;
5726                 }
5727                 path->slots[0]++;
5728         }
5729         ret = -ENOENT;
5730 out:
5731         return ret;
5732 }
5733
5734 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5735 {
5736         struct btrfs_block_group_cache *block_group;
5737         struct rb_node *n;
5738
5739         spin_lock(&info->block_group_cache_lock);
5740         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5741                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5742                                        cache_node);
5743                 rb_erase(&block_group->cache_node,
5744                          &info->block_group_cache_tree);
5745                 spin_unlock(&info->block_group_cache_lock);
5746
5747                 btrfs_remove_free_space_cache(block_group);
5748                 down_write(&block_group->space_info->groups_sem);
5749                 list_del(&block_group->list);
5750                 up_write(&block_group->space_info->groups_sem);
5751
5752                 WARN_ON(atomic_read(&block_group->count) != 1);
5753                 kfree(block_group);
5754
5755                 spin_lock(&info->block_group_cache_lock);
5756         }
5757         spin_unlock(&info->block_group_cache_lock);
5758         return 0;
5759 }
5760
5761 int btrfs_read_block_groups(struct btrfs_root *root)
5762 {
5763         struct btrfs_path *path;
5764         int ret;
5765         struct btrfs_block_group_cache *cache;
5766         struct btrfs_fs_info *info = root->fs_info;
5767         struct btrfs_space_info *space_info;
5768         struct btrfs_key key;
5769         struct btrfs_key found_key;
5770         struct extent_buffer *leaf;
5771
5772         root = info->extent_root;
5773         key.objectid = 0;
5774         key.offset = 0;
5775         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5776         path = btrfs_alloc_path();
5777         if (!path)
5778                 return -ENOMEM;
5779
5780         while(1) {
5781                 ret = find_first_block_group(root, path, &key);
5782                 if (ret > 0) {
5783                         ret = 0;
5784                         goto error;
5785                 }
5786                 if (ret != 0)
5787                         goto error;
5788
5789                 leaf = path->nodes[0];
5790                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5791                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5792                 if (!cache) {
5793                         ret = -ENOMEM;
5794                         break;
5795                 }
5796
5797                 atomic_set(&cache->count, 1);
5798                 spin_lock_init(&cache->lock);
5799                 mutex_init(&cache->alloc_mutex);
5800                 mutex_init(&cache->cache_mutex);
5801                 INIT_LIST_HEAD(&cache->list);
5802                 read_extent_buffer(leaf, &cache->item,
5803                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5804                                    sizeof(cache->item));
5805                 memcpy(&cache->key, &found_key, sizeof(found_key));
5806
5807                 key.objectid = found_key.objectid + found_key.offset;
5808                 btrfs_release_path(root, path);
5809                 cache->flags = btrfs_block_group_flags(&cache->item);
5810
5811                 ret = update_space_info(info, cache->flags, found_key.offset,
5812                                         btrfs_block_group_used(&cache->item),
5813                                         &space_info);
5814                 BUG_ON(ret);
5815                 cache->space_info = space_info;
5816                 down_write(&space_info->groups_sem);
5817                 list_add_tail(&cache->list, &space_info->block_groups);
5818                 up_write(&space_info->groups_sem);
5819
5820                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5821                 BUG_ON(ret);
5822
5823                 set_avail_alloc_bits(root->fs_info, cache->flags);
5824                 if (btrfs_chunk_readonly(root, cache->key.objectid))
5825                         set_block_group_readonly(cache);
5826         }
5827         ret = 0;
5828 error:
5829         btrfs_free_path(path);
5830         return ret;
5831 }
5832
5833 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5834                            struct btrfs_root *root, u64 bytes_used,
5835                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5836                            u64 size)
5837 {
5838         int ret;
5839         struct btrfs_root *extent_root;
5840         struct btrfs_block_group_cache *cache;
5841
5842         extent_root = root->fs_info->extent_root;
5843
5844         root->fs_info->last_trans_new_blockgroup = trans->transid;
5845
5846         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5847         if (!cache)
5848                 return -ENOMEM;
5849
5850         cache->key.objectid = chunk_offset;
5851         cache->key.offset = size;
5852         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5853         atomic_set(&cache->count, 1);
5854         spin_lock_init(&cache->lock);
5855         mutex_init(&cache->alloc_mutex);
5856         mutex_init(&cache->cache_mutex);
5857         INIT_LIST_HEAD(&cache->list);
5858
5859         btrfs_set_block_group_used(&cache->item, bytes_used);
5860         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5861         cache->flags = type;
5862         btrfs_set_block_group_flags(&cache->item, type);
5863
5864         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5865                                 &cache->space_info);
5866         BUG_ON(ret);
5867         down_write(&cache->space_info->groups_sem);
5868         list_add_tail(&cache->list, &cache->space_info->block_groups);
5869         up_write(&cache->space_info->groups_sem);
5870
5871         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5872         BUG_ON(ret);
5873
5874         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5875                                 sizeof(cache->item));
5876         BUG_ON(ret);
5877
5878         finish_current_insert(trans, extent_root, 0);
5879         ret = del_pending_extents(trans, extent_root, 0);
5880         BUG_ON(ret);
5881         set_avail_alloc_bits(extent_root->fs_info, type);
5882
5883         return 0;
5884 }
5885
5886 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5887                              struct btrfs_root *root, u64 group_start)
5888 {
5889         struct btrfs_path *path;
5890         struct btrfs_block_group_cache *block_group;
5891         struct btrfs_key key;
5892         int ret;
5893
5894         root = root->fs_info->extent_root;
5895
5896         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5897         BUG_ON(!block_group);
5898         BUG_ON(!block_group->ro);
5899
5900         memcpy(&key, &block_group->key, sizeof(key));
5901
5902         path = btrfs_alloc_path();
5903         BUG_ON(!path);
5904
5905         btrfs_remove_free_space_cache(block_group);
5906         rb_erase(&block_group->cache_node,
5907                  &root->fs_info->block_group_cache_tree);
5908         down_write(&block_group->space_info->groups_sem);
5909         list_del(&block_group->list);
5910         up_write(&block_group->space_info->groups_sem);
5911
5912         spin_lock(&block_group->space_info->lock);
5913         block_group->space_info->total_bytes -= block_group->key.offset;
5914         block_group->space_info->bytes_readonly -= block_group->key.offset;
5915         spin_unlock(&block_group->space_info->lock);
5916         block_group->space_info->full = 0;
5917
5918         put_block_group(block_group);
5919         put_block_group(block_group);
5920
5921         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5922         if (ret > 0)
5923                 ret = -EIO;
5924         if (ret < 0)
5925                 goto out;
5926
5927         ret = btrfs_del_item(trans, root, path);
5928 out:
5929         btrfs_free_path(path);
5930         return ret;
5931 }