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