Pileus Git - ~andy/linux/blob - fs/btrfs/extent-tree.c

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