SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/version.h>
  23 #include "compat.h"
  24 #include "hash.h"
  25 #include "crc32c.h"
  26 #include "ctree.h"
  27 #include "disk-io.h"
  28 #include "print-tree.h"
  29 #include "transaction.h"
  30 #include "volumes.h"
  31 #include "locking.h"
  32 #include "ref-cache.h"
  33 #include "compat.h"
  34
  35 #define PENDING_EXTENT_INSERT 0
  36 #define PENDING_EXTENT_DELETE 1
  37 #define PENDING_BACKREF_UPDATE 2
  38
  39 struct pending_extent_op {
  40         int type;
  41         u64 bytenr;
  42         u64 num_bytes;
  43         u64 parent;
  44         u64 orig_parent;
  45         u64 generation;
  46         u64 orig_generation;
  47         int level;
  48         struct list_head list;
  49         int del;
  50 };
  51
  52 static int finish_current_insert(struct btrfs_trans_handle *trans,
  53                                  struct btrfs_root *extent_root, int all);
  54 static int del_pending_extents(struct btrfs_trans_handle *trans,
  55                                struct btrfs_root *extent_root, int all);
  56 static int pin_down_bytes(struct btrfs_trans_handle *trans,
  57                           struct btrfs_root *root,
  58                           u64 bytenr, u64 num_bytes, int is_data);
  59 static int update_block_group(struct btrfs_trans_handle *trans,
  60                               struct btrfs_root *root,
  61                               u64 bytenr, u64 num_bytes, int alloc,
  62                               int mark_free);
  63
  64 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  65 {
  66         return (cache->flags & bits) == bits;
  67 }
  68
  69 /*
  70  * this adds the block group to the fs_info rb tree for the block group
  71  * cache
  72  */
  73 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  74                                 struct btrfs_block_group_cache *block_group)
  75 {
  76         struct rb_node **p;
  77         struct rb_node *parent = NULL;
  78         struct btrfs_block_group_cache *cache;
  79
  80         spin_lock(&info->block_group_cache_lock);
  81         p = &info->block_group_cache_tree.rb_node;
  82
  83         while (*p) {
  84                 parent = *p;
  85                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  86                                  cache_node);
  87                 if (block_group->key.objectid < cache->key.objectid) {
  88                         p = &(*p)->rb_left;
  89                 } else if (block_group->key.objectid > cache->key.objectid) {
  90                         p = &(*p)->rb_right;
  91                 } else {
  92                         spin_unlock(&info->block_group_cache_lock);
  93                         return -EEXIST;
  94                 }
  95         }
  96
  97         rb_link_node(&block_group->cache_node, parent, p);
  98         rb_insert_color(&block_group->cache_node,
  99                         &info->block_group_cache_tree);
 100         spin_unlock(&info->block_group_cache_lock);
 101
 102         return 0;
 103 }
 104
 105 /*
 106  * This will return the block group at or after bytenr if contains is 0, else
 107  * it will return the block group that contains the bytenr
 108  */
 109 static struct btrfs_block_group_cache *
 110 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 111                               int contains)
 112 {
 113         struct btrfs_block_group_cache *cache, *ret = NULL;
 114         struct rb_node *n;
 115         u64 end, start;
 116
 117         spin_lock(&info->block_group_cache_lock);
 118         n = info->block_group_cache_tree.rb_node;
 119
 120         while (n) {
 121                 cache = rb_entry(n, struct btrfs_block_group_cache,
 122                                  cache_node);
 123                 end = cache->key.objectid + cache->key.offset - 1;
 124                 start = cache->key.objectid;
 125
 126                 if (bytenr < start) {
 127                         if (!contains && (!ret || start < ret->key.objectid))
 128                                 ret = cache;
 129                         n = n->rb_left;
 130                 } else if (bytenr > start) {
 131                         if (contains && bytenr <= end) {
 132                                 ret = cache;
 133                                 break;
 134                         }
 135                         n = n->rb_right;
 136                 } else {
 137                         ret = cache;
 138                         break;
 139                 }
 140         }
 141         if (ret)
 142                 atomic_inc(&ret->count);
 143         spin_unlock(&info->block_group_cache_lock);
 144
 145         return ret;
 146 }
 147
 148 /*
 149  * this is only called by cache_block_group, since we could have freed extents
 150  * we need to check the pinned_extents for any extents that can't be used yet
 151  * since their free space will be released as soon as the transaction commits.
 152  */
 153 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 154                               struct btrfs_fs_info *info, u64 start, u64 end)
 155 {
 156         u64 extent_start, extent_end, size;
 157         int ret;
 158
 159         mutex_lock(&info->pinned_mutex);
 160         while (start < end) {
 161                 ret = find_first_extent_bit(&info->pinned_extents, start,
 162                                             &extent_start, &extent_end,
 163                                             EXTENT_DIRTY);
 164                 if (ret)
 165                         break;
 166
 167                 if (extent_start == start) {
 168                         start = extent_end + 1;
 169                 } else if (extent_start > start && extent_start < end) {
 170                         size = extent_start - start;
 171                         ret = btrfs_add_free_space(block_group, start,
 172                                                    size);
 173                         BUG_ON(ret);
 174                         start = extent_end + 1;
 175                 } else {
 176                         break;
 177                 }
 178         }
 179
 180         if (start < end) {
 181                 size = end - start;
 182                 ret = btrfs_add_free_space(block_group, start, size);
 183                 BUG_ON(ret);
 184         }
 185         mutex_unlock(&info->pinned_mutex);
 186
 187         return 0;
 188 }
 189
 190 static int remove_sb_from_cache(struct btrfs_root *root,
 191                                 struct btrfs_block_group_cache *cache)
 192 {
 193         u64 bytenr;
 194         u64 *logical;
 195         int stripe_len;
 196         int i, nr, ret;
 197
 198         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 199                 bytenr = btrfs_sb_offset(i);
 200                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 201                                        cache->key.objectid, bytenr, 0,
 202                                        &logical, &nr, &stripe_len);
 203                 BUG_ON(ret);
 204                 while (nr--) {
 205                         btrfs_remove_free_space(cache, logical[nr],
 206                                                 stripe_len);
 207                 }
 208                 kfree(logical);
 209         }
 210         return 0;
 211 }
 212
 213 static int cache_block_group(struct btrfs_root *root,
 214                              struct btrfs_block_group_cache *block_group)
 215 {
 216         struct btrfs_path *path;
 217         int ret = 0;
 218         struct btrfs_key key;
 219         struct extent_buffer *leaf;
 220         int slot;
 221         u64 last;
 222
 223         if (!block_group)
 224                 return 0;
 225
 226         root = root->fs_info->extent_root;
 227
 228         if (block_group->cached)
 229                 return 0;
 230
 231         path = btrfs_alloc_path();
 232         if (!path)
 233                 return -ENOMEM;
 234
 235         path->reada = 2;
 236         /*
 237          * we get into deadlocks with paths held by callers of this function.
 238          * since the alloc_mutex is protecting things right now, just
 239          * skip the locking here
 240          */
 241         path->skip_locking = 1;
 242         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 243         key.objectid = last;
 244         key.offset = 0;
 245         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 246         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 247         if (ret < 0)
 248                 goto err;
 249
 250         while (1) {
 251                 leaf = path->nodes[0];
 252                 slot = path->slots[0];
 253                 if (slot >= btrfs_header_nritems(leaf)) {
 254                         ret = btrfs_next_leaf(root, path);
 255                         if (ret < 0)
 256                                 goto err;
 257                         if (ret == 0)
 258                                 continue;
 259                         else
 260                                 break;
 261                 }
 262                 btrfs_item_key_to_cpu(leaf, &key, slot);
 263                 if (key.objectid < block_group->key.objectid)
 264                         goto next;
 265
 266                 if (key.objectid >= block_group->key.objectid +
 267                     block_group->key.offset)
 268                         break;
 269
 270                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 271                         add_new_free_space(block_group, root->fs_info, last,
 272                                            key.objectid);
 273
 274                         last = key.objectid + key.offset;
 275                 }
 276 next:
 277                 path->slots[0]++;
 278         }
 279
 280         add_new_free_space(block_group, root->fs_info, last,
 281                            block_group->key.objectid +
 282                            block_group->key.offset);
 283
 284         remove_sb_from_cache(root, block_group);
 285         block_group->cached = 1;
 286         ret = 0;
 287 err:
 288         btrfs_free_path(path);
 289         return ret;
 290 }
 291
 292 /*
 293  * return the block group that starts at or after bytenr
 294  */
 295 static struct btrfs_block_group_cache *
 296 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 297 {
 298         struct btrfs_block_group_cache *cache;
 299
 300         cache = block_group_cache_tree_search(info, bytenr, 0);
 301
 302         return cache;
 303 }
 304
 305 /*
 306  * return the block group that contains teh given bytenr
 307  */
 308 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 309                                                  struct btrfs_fs_info *info,
 310                                                  u64 bytenr)
 311 {
 312         struct btrfs_block_group_cache *cache;
 313
 314         cache = block_group_cache_tree_search(info, bytenr, 1);
 315
 316         return cache;
 317 }
 318
 319 static inline void put_block_group(struct btrfs_block_group_cache *cache)
 320 {
 321         if (atomic_dec_and_test(&cache->count))
 322                 kfree(cache);
 323 }
 324
 325 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 326                                                   u64 flags)
 327 {
 328         struct list_head *head = &info->space_info;
 329         struct list_head *cur;
 330         struct btrfs_space_info *found;
 331         list_for_each(cur, head) {
 332                 found = list_entry(cur, struct btrfs_space_info, list);
 333                 if (found->flags == flags)
 334                         return found;
 335         }
 336         return NULL;
 337 }
 338
 339 static u64 div_factor(u64 num, int factor)
 340 {
 341         if (factor == 10)
 342                 return num;
 343         num *= factor;
 344         do_div(num, 10);
 345         return num;
 346 }
 347
 348 u64 btrfs_find_block_group(struct btrfs_root *root,
 349                            u64 search_start, u64 search_hint, int owner)
 350 {
 351         struct btrfs_block_group_cache *cache;
 352         u64 used;
 353         u64 last = max(search_hint, search_start);
 354         u64 group_start = 0;
 355         int full_search = 0;
 356         int factor = 9;
 357         int wrapped = 0;
 358 again:
 359         while (1) {
 360                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 361                 if (!cache)
 362                         break;
 363
 364                 spin_lock(&cache->lock);
 365                 last = cache->key.objectid + cache->key.offset;
 366                 used = btrfs_block_group_used(&cache->item);
 367
 368                 if ((full_search || !cache->ro) &&
 369                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 370                         if (used + cache->pinned + cache->reserved <
 371                             div_factor(cache->key.offset, factor)) {
 372                                 group_start = cache->key.objectid;
 373                                 spin_unlock(&cache->lock);
 374                                 put_block_group(cache);
 375                                 goto found;
 376                         }
 377                 }
 378                 spin_unlock(&cache->lock);
 379                 put_block_group(cache);
 380                 cond_resched();
 381         }
 382         if (!wrapped) {
 383                 last = search_start;
 384                 wrapped = 1;
 385                 goto again;
 386         }
 387         if (!full_search && factor < 10) {
 388                 last = search_start;
 389                 full_search = 1;
 390                 factor = 10;
 391                 goto again;
 392         }
 393 found:
 394         return group_start;
 395 }
 396
 397 /* simple helper to search for an existing extent at a given offset */
 398 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 399 {
 400         int ret;
 401         struct btrfs_key key;
 402         struct btrfs_path *path;
 403
 404         path = btrfs_alloc_path();
 405         BUG_ON(!path);
 406         key.objectid = start;
 407         key.offset = len;
 408         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 409         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 410                                 0, 0);
 411         btrfs_free_path(path);
 412         return ret;
 413 }
 414
 415 /*
 416  * Back reference rules.  Back refs have three main goals:
 417  *
 418  * 1) differentiate between all holders of references to an extent so that
 419  *    when a reference is dropped we can make sure it was a valid reference
 420  *    before freeing the extent.
 421  *
 422  * 2) Provide enough information to quickly find the holders of an extent
 423  *    if we notice a given block is corrupted or bad.
 424  *
 425  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 426  *    maintenance.  This is actually the same as #2, but with a slightly
 427  *    different use case.
 428  *
 429  * File extents can be referenced by:
 430  *
 431  * - multiple snapshots, subvolumes, or different generations in one subvol
 432  * - different files inside a single subvolume
 433  * - different offsets inside a file (bookend extents in file.c)
 434  *
 435  * The extent ref structure has fields for:
 436  *
 437  * - Objectid of the subvolume root
 438  * - Generation number of the tree holding the reference
 439  * - objectid of the file holding the reference
 440  * - number of references holding by parent node (alway 1 for tree blocks)
 441  *
 442  * Btree leaf may hold multiple references to a file extent. In most cases,
 443  * these references are from same file and the corresponding offsets inside
 444  * the file are close together.
 445  *
 446  * When a file extent is allocated the fields are filled in:
 447  *     (root_key.objectid, trans->transid, inode objectid, 1)
 448  *
 449  * When a leaf is cow'd new references are added for every file extent found
 450  * in the leaf.  It looks similar to the create case, but trans->transid will
 451  * be different when the block is cow'd.
 452  *
 453  *     (root_key.objectid, trans->transid, inode objectid,
 454  *      number of references in the leaf)
 455  *
 456  * When a file extent is removed either during snapshot deletion or
 457  * file truncation, we find the corresponding back reference and check
 458  * the following fields:
 459  *
 460  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 461  *      inode objectid)
 462  *
 463  * Btree extents can be referenced by:
 464  *
 465  * - Different subvolumes
 466  * - Different generations of the same subvolume
 467  *
 468  * When a tree block is created, back references are inserted:
 469  *
 470  * (root->root_key.objectid, trans->transid, level, 1)
 471  *
 472  * When a tree block is cow'd, new back references are added for all the
 473  * blocks it points to. If the tree block isn't in reference counted root,
 474  * the old back references are removed. These new back references are of
 475  * the form (trans->transid will have increased since creation):
 476  *
 477  * (root->root_key.objectid, trans->transid, level, 1)
 478  *
 479  * When a backref is in deleting, the following fields are checked:
 480  *
 481  * if backref was for a tree root:
 482  *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
 483  * else
 484  *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
 485  *
 486  * Back Reference Key composing:
 487  *
 488  * The key objectid corresponds to the first byte in the extent, the key
 489  * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
 490  * byte of parent extent. If a extent is tree root, the key offset is set
 491  * to the key objectid.
 492  */
 493
 494 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
 495                                           struct btrfs_root *root,
 496                                           struct btrfs_path *path,
 497                                           u64 bytenr, u64 parent,
 498                                           u64 ref_root, u64 ref_generation,
 499                                           u64 owner_objectid, int del)
 500 {
 501         struct btrfs_key key;
 502         struct btrfs_extent_ref *ref;
 503         struct extent_buffer *leaf;
 504         u64 ref_objectid;
 505         int ret;
 506
 507         key.objectid = bytenr;
 508         key.type = BTRFS_EXTENT_REF_KEY;
 509         key.offset = parent;
 510
 511         ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
 512         if (ret < 0)
 513                 goto out;
 514         if (ret > 0) {
 515                 ret = -ENOENT;
 516                 goto out;
 517         }
 518
 519         leaf = path->nodes[0];
 520         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 521         ref_objectid = btrfs_ref_objectid(leaf, ref);
 522         if (btrfs_ref_root(leaf, ref) != ref_root ||
 523             btrfs_ref_generation(leaf, ref) != ref_generation ||
 524             (ref_objectid != owner_objectid &&
 525              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 526                 ret = -EIO;
 527                 WARN_ON(1);
 528                 goto out;
 529         }
 530         ret = 0;
 531 out:
 532         return ret;
 533 }
 534
 535 /*
 536  * updates all the backrefs that are pending on update_list for the
 537  * extent_root
 538  */
 539 static noinline int update_backrefs(struct btrfs_trans_handle *trans,
 540                                     struct btrfs_root *extent_root,
 541                                     struct btrfs_path *path,
 542                                     struct list_head *update_list)
 543 {
 544         struct btrfs_key key;
 545         struct btrfs_extent_ref *ref;
 546         struct btrfs_fs_info *info = extent_root->fs_info;
 547         struct pending_extent_op *op;
 548         struct extent_buffer *leaf;
 549         int ret = 0;
 550         struct list_head *cur = update_list->next;
 551         u64 ref_objectid;
 552         u64 ref_root = extent_root->root_key.objectid;
 553
 554         op = list_entry(cur, struct pending_extent_op, list);
 555
 556 search:
 557         key.objectid = op->bytenr;
 558         key.type = BTRFS_EXTENT_REF_KEY;
 559         key.offset = op->orig_parent;
 560
 561         ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
 562         BUG_ON(ret);
 563
 564         leaf = path->nodes[0];
 565
 566 loop:
 567         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 568
 569         ref_objectid = btrfs_ref_objectid(leaf, ref);
 570
 571         if (btrfs_ref_root(leaf, ref) != ref_root ||
 572             btrfs_ref_generation(leaf, ref) != op->orig_generation ||
 573             (ref_objectid != op->level &&
 574              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 575                 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
 576                        "root %llu, owner %u\n",
 577                        (unsigned long long)op->bytenr,
 578                        (unsigned long long)op->orig_parent,
 579                        (unsigned long long)ref_root, op->level);
 580                 btrfs_print_leaf(extent_root, leaf);
 581                 BUG();
 582         }
 583
 584         key.objectid = op->bytenr;
 585         key.offset = op->parent;
 586         key.type = BTRFS_EXTENT_REF_KEY;
 587         ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
 588         BUG_ON(ret);
 589         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 590         btrfs_set_ref_generation(leaf, ref, op->generation);
 591
 592         cur = cur->next;
 593
 594         list_del_init(&op->list);
 595         unlock_extent(&info->extent_ins, op->bytenr,
 596                       op->bytenr + op->num_bytes - 1, GFP_NOFS);
 597         kfree(op);
 598
 599         if (cur == update_list) {
 600                 btrfs_mark_buffer_dirty(path->nodes[0]);
 601                 btrfs_release_path(extent_root, path);
 602                 goto out;
 603         }
 604
 605         op = list_entry(cur, struct pending_extent_op, list);
 606
 607         path->slots[0]++;
 608         while (path->slots[0] < btrfs_header_nritems(leaf)) {
 609                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 610                 if (key.objectid == op->bytenr &&
 611                     key.type == BTRFS_EXTENT_REF_KEY)
 612                         goto loop;
 613                 path->slots[0]++;
 614         }
 615
 616         btrfs_mark_buffer_dirty(path->nodes[0]);
 617         btrfs_release_path(extent_root, path);
 618         goto search;
 619
 620 out:
 621         return 0;
 622 }
 623
 624 static noinline int insert_extents(struct btrfs_trans_handle *trans,
 625                                    struct btrfs_root *extent_root,
 626                                    struct btrfs_path *path,
 627                                    struct list_head *insert_list, int nr)
 628 {
 629         struct btrfs_key *keys;
 630         u32 *data_size;
 631         struct pending_extent_op *op;
 632         struct extent_buffer *leaf;
 633         struct list_head *cur = insert_list->next;
 634         struct btrfs_fs_info *info = extent_root->fs_info;
 635         u64 ref_root = extent_root->root_key.objectid;
 636         int i = 0, last = 0, ret;
 637         int total = nr * 2;
 638
 639         if (!nr)
 640                 return 0;
 641
 642         keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
 643         if (!keys)
 644                 return -ENOMEM;
 645
 646         data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
 647         if (!data_size) {
 648                 kfree(keys);
 649                 return -ENOMEM;
 650         }
 651
 652         list_for_each_entry(op, insert_list, list) {
 653                 keys[i].objectid = op->bytenr;
 654                 keys[i].offset = op->num_bytes;
 655                 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
 656                 data_size[i] = sizeof(struct btrfs_extent_item);
 657                 i++;
 658
 659                 keys[i].objectid = op->bytenr;
 660                 keys[i].offset = op->parent;
 661                 keys[i].type = BTRFS_EXTENT_REF_KEY;
 662                 data_size[i] = sizeof(struct btrfs_extent_ref);
 663                 i++;
 664         }
 665
 666         op = list_entry(cur, struct pending_extent_op, list);
 667         i = 0;
 668         while (i < total) {
 669                 int c;
 670                 ret = btrfs_insert_some_items(trans, extent_root, path,
 671                                               keys+i, data_size+i, total-i);
 672                 BUG_ON(ret < 0);
 673
 674                 if (last && ret > 1)
 675                         BUG();
 676
 677                 leaf = path->nodes[0];
 678                 for (c = 0; c < ret; c++) {
 679                         int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
 680
 681                         /*
 682                          * if the first item we inserted was a backref, then
 683                          * the EXTENT_ITEM will be the odd c's, else it will
 684                          * be the even c's
 685                          */
 686                         if ((ref_first && (c % 2)) ||
 687                             (!ref_first && !(c % 2))) {
 688                                 struct btrfs_extent_item *itm;
 689
 690                                 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
 691                                                      struct btrfs_extent_item);
 692                                 btrfs_set_extent_refs(path->nodes[0], itm, 1);
 693                                 op->del++;
 694                         } else {
 695                                 struct btrfs_extent_ref *ref;
 696
 697                                 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
 698                                                      struct btrfs_extent_ref);
 699                                 btrfs_set_ref_root(leaf, ref, ref_root);
 700                                 btrfs_set_ref_generation(leaf, ref,
 701                                                          op->generation);
 702                                 btrfs_set_ref_objectid(leaf, ref, op->level);
 703                                 btrfs_set_ref_num_refs(leaf, ref, 1);
 704                                 op->del++;
 705                         }
 706
 707                         /*
 708                          * using del to see when its ok to free up the
 709                          * pending_extent_op.  In the case where we insert the
 710                          * last item on the list in order to help do batching
 711                          * we need to not free the extent op until we actually
 712                          * insert the extent_item
 713                          */
 714                         if (op->del == 2) {
 715                                 unlock_extent(&info->extent_ins, op->bytenr,
 716                                               op->bytenr + op->num_bytes - 1,
 717                                               GFP_NOFS);
 718                                 cur = cur->next;
 719                                 list_del_init(&op->list);
 720                                 kfree(op);
 721                                 if (cur != insert_list)
 722                                         op = list_entry(cur,
 723                                                 struct pending_extent_op,
 724                                                 list);
 725                         }
 726                 }
 727                 btrfs_mark_buffer_dirty(leaf);
 728                 btrfs_release_path(extent_root, path);
 729
 730                 /*
 731                  * Ok backref's and items usually go right next to eachother,
 732                  * but if we could only insert 1 item that means that we
 733                  * inserted on the end of a leaf, and we have no idea what may
 734                  * be on the next leaf so we just play it safe.  In order to
 735                  * try and help this case we insert the last thing on our
 736                  * insert list so hopefully it will end up being the last
 737                  * thing on the leaf and everything else will be before it,
 738                  * which will let us insert a whole bunch of items at the same
 739                  * time.
 740                  */
 741                 if (ret == 1 && !last && (i + ret < total)) {
 742                         /*
 743                          * last: where we will pick up the next time around
 744                          * i: our current key to insert, will be total - 1
 745                          * cur: the current op we are screwing with
 746                          * op: duh
 747                          */
 748                         last = i + ret;
 749                         i = total - 1;
 750                         cur = insert_list->prev;
 751                         op = list_entry(cur, struct pending_extent_op, list);
 752                 } else if (last) {
 753                         /*
 754                          * ok we successfully inserted the last item on the
 755                          * list, lets reset everything
 756                          *
 757                          * i: our current key to insert, so where we left off
 758                          *    last time
 759                          * last: done with this
 760                          * cur: the op we are messing with
 761                          * op: duh
 762                          * total: since we inserted the last key, we need to
 763                          *        decrement total so we dont overflow
 764                          */
 765                         i = last;
 766                         last = 0;
 767                         total--;
 768                         if (i < total) {
 769                                 cur = insert_list->next;
 770                                 op = list_entry(cur, struct pending_extent_op,
 771                                                 list);
 772                         }
 773                 } else {
 774                         i += ret;
 775                 }
 776
 777                 cond_resched();
 778         }
 779         ret = 0;
 780         kfree(keys);
 781         kfree(data_size);
 782         return ret;
 783 }
 784
 785 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
 786                                           struct btrfs_root *root,
 787                                           struct btrfs_path *path,
 788                                           u64 bytenr, u64 parent,
 789                                           u64 ref_root, u64 ref_generation,
 790                                           u64 owner_objectid)
 791 {
 792         struct btrfs_key key;
 793         struct extent_buffer *leaf;
 794         struct btrfs_extent_ref *ref;
 795         u32 num_refs;
 796         int ret;
 797
 798         key.objectid = bytenr;
 799         key.type = BTRFS_EXTENT_REF_KEY;
 800         key.offset = parent;
 801
 802         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
 803         if (ret == 0) {
 804                 leaf = path->nodes[0];
 805                 ref = btrfs_item_ptr(leaf, path->slots[0],
 806                                      struct btrfs_extent_ref);
 807                 btrfs_set_ref_root(leaf, ref, ref_root);
 808                 btrfs_set_ref_generation(leaf, ref, ref_generation);
 809                 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
 810                 btrfs_set_ref_num_refs(leaf, ref, 1);
 811         } else if (ret == -EEXIST) {
 812                 u64 existing_owner;
 813                 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
 814                 leaf = path->nodes[0];
 815                 ref = btrfs_item_ptr(leaf, path->slots[0],
 816                                      struct btrfs_extent_ref);
 817                 if (btrfs_ref_root(leaf, ref) != ref_root ||
 818                     btrfs_ref_generation(leaf, ref) != ref_generation) {
 819                         ret = -EIO;
 820                         WARN_ON(1);
 821                         goto out;
 822                 }
 823
 824                 num_refs = btrfs_ref_num_refs(leaf, ref);
 825                 BUG_ON(num_refs == 0);
 826                 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
 827
 828                 existing_owner = btrfs_ref_objectid(leaf, ref);
 829                 if (existing_owner != owner_objectid &&
 830                     existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
 831                         btrfs_set_ref_objectid(leaf, ref,
 832                                         BTRFS_MULTIPLE_OBJECTIDS);
 833                 }
 834                 ret = 0;
 835         } else {
 836                 goto out;
 837         }
 838         btrfs_mark_buffer_dirty(path->nodes[0]);
 839 out:
 840         btrfs_release_path(root, path);
 841         return ret;
 842 }
 843
 844 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
 845                                           struct btrfs_root *root,
 846                                           struct btrfs_path *path)
 847 {
 848         struct extent_buffer *leaf;
 849         struct btrfs_extent_ref *ref;
 850         u32 num_refs;
 851         int ret = 0;
 852
 853         leaf = path->nodes[0];
 854         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 855         num_refs = btrfs_ref_num_refs(leaf, ref);
 856         BUG_ON(num_refs == 0);
 857         num_refs -= 1;
 858         if (num_refs == 0) {
 859                 ret = btrfs_del_item(trans, root, path);
 860         } else {
 861                 btrfs_set_ref_num_refs(leaf, ref, num_refs);
 862                 btrfs_mark_buffer_dirty(leaf);
 863         }
 864         btrfs_release_path(root, path);
 865         return ret;
 866 }
 867
 868 #ifdef BIO_RW_DISCARD
 869 static void btrfs_issue_discard(struct block_device *bdev,
 870                                 u64 start, u64 len)
 871 {
 872 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 28)
 873         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
 874 #else
 875         blkdev_issue_discard(bdev, start >> 9, len >> 9);
 876 #endif
 877 }
 878 #endif
 879
 880 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
 881                                 u64 num_bytes)
 882 {
 883 #ifdef BIO_RW_DISCARD
 884         int ret;
 885         u64 map_length = num_bytes;
 886         struct btrfs_multi_bio *multi = NULL;
 887
 888         /* Tell the block device(s) that the sectors can be discarded */
 889         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
 890                               bytenr, &map_length, &multi, 0);
 891         if (!ret) {
 892                 struct btrfs_bio_stripe *stripe = multi->stripes;
 893                 int i;
 894
 895                 if (map_length > num_bytes)
 896                         map_length = num_bytes;
 897
 898                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
 899                         btrfs_issue_discard(stripe->dev->bdev,
 900                                             stripe->physical,
 901                                             map_length);
 902                 }
 903                 kfree(multi);
 904         }
 905
 906         return ret;
 907 #else
 908         return 0;
 909 #endif
 910 }
 911
 912 static noinline int free_extents(struct btrfs_trans_handle *trans,
 913                                  struct btrfs_root *extent_root,
 914                                  struct list_head *del_list)
 915 {
 916         struct btrfs_fs_info *info = extent_root->fs_info;
 917         struct btrfs_path *path;
 918         struct btrfs_key key, found_key;
 919         struct extent_buffer *leaf;
 920         struct list_head *cur;
 921         struct pending_extent_op *op;
 922         struct btrfs_extent_item *ei;
 923         int ret, num_to_del, extent_slot = 0, found_extent = 0;
 924         u32 refs;
 925         u64 bytes_freed = 0;
 926
 927         path = btrfs_alloc_path();
 928         if (!path)
 929                 return -ENOMEM;
 930         path->reada = 1;
 931
 932 search:
 933         /* search for the backref for the current ref we want to delete */
 934         cur = del_list->next;
 935         op = list_entry(cur, struct pending_extent_op, list);
 936         ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
 937                                     op->orig_parent,
 938                                     extent_root->root_key.objectid,
 939                                     op->orig_generation, op->level, 1);
 940         if (ret) {
 941                 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
 942                        "root %llu gen %llu owner %u\n",
 943                        (unsigned long long)op->bytenr,
 944                        (unsigned long long)extent_root->root_key.objectid,
 945                        (unsigned long long)op->orig_generation, op->level);
 946                 btrfs_print_leaf(extent_root, path->nodes[0]);
 947                 WARN_ON(1);
 948                 goto out;
 949         }
 950
 951         extent_slot = path->slots[0];
 952         num_to_del = 1;
 953         found_extent = 0;
 954
 955         /*
 956          * if we aren't the first item on the leaf we can move back one and see
 957          * if our ref is right next to our extent item
 958          */
 959         if (likely(extent_slot)) {
 960                 extent_slot--;
 961                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 962                                       extent_slot);
 963                 if (found_key.objectid == op->bytenr &&
 964                     found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 965                     found_key.offset == op->num_bytes) {
 966                         num_to_del++;
 967                         found_extent = 1;
 968                 }
 969         }
 970
 971         /*
 972          * if we didn't find the extent we need to delete the backref and then
 973          * search for the extent item key so we can update its ref count
 974          */
 975         if (!found_extent) {
 976                 key.objectid = op->bytenr;
 977                 key.type = BTRFS_EXTENT_ITEM_KEY;
 978                 key.offset = op->num_bytes;
 979
 980                 ret = remove_extent_backref(trans, extent_root, path);
 981                 BUG_ON(ret);
 982                 btrfs_release_path(extent_root, path);
 983                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
 984                 BUG_ON(ret);
 985                 extent_slot = path->slots[0];
 986         }
 987
 988         /* this is where we update the ref count for the extent */
 989         leaf = path->nodes[0];
 990         ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
 991         refs = btrfs_extent_refs(leaf, ei);
 992         BUG_ON(refs == 0);
 993         refs--;
 994         btrfs_set_extent_refs(leaf, ei, refs);
 995
 996         btrfs_mark_buffer_dirty(leaf);
 997
 998         /*
 999          * This extent needs deleting.  The reason cur_slot is extent_slot +
1000          * num_to_del is because extent_slot points to the slot where the extent
1001          * is, and if the backref was not right next to the extent we will be
1002          * deleting at least 1 item, and will want to start searching at the
1003          * slot directly next to extent_slot.  However if we did find the
1004          * backref next to the extent item them we will be deleting at least 2
1005          * items and will want to start searching directly after the ref slot
1006          */
1007         if (!refs) {
1008                 struct list_head *pos, *n, *end;
1009                 int cur_slot = extent_slot+num_to_del;
1010                 u64 super_used;
1011                 u64 root_used;
1012
1013                 path->slots[0] = extent_slot;
1014                 bytes_freed = op->num_bytes;
1015
1016                 mutex_lock(&info->pinned_mutex);
1017                 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1018                                      op->num_bytes, op->level >=
1019                                      BTRFS_FIRST_FREE_OBJECTID);
1020                 mutex_unlock(&info->pinned_mutex);
1021                 BUG_ON(ret < 0);
1022                 op->del = ret;
1023
1024                 /*
1025                  * we need to see if we can delete multiple things at once, so
1026                  * start looping through the list of extents we are wanting to
1027                  * delete and see if their extent/backref's are right next to
1028                  * eachother and the extents only have 1 ref
1029                  */
1030                 for (pos = cur->next; pos != del_list; pos = pos->next) {
1031                         struct pending_extent_op *tmp;
1032
1033                         tmp = list_entry(pos, struct pending_extent_op, list);
1034
1035                         /* we only want to delete extent+ref at this stage */
1036                         if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1037                                 break;
1038
1039                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1040                         if (found_key.objectid != tmp->bytenr ||
1041                             found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1042                             found_key.offset != tmp->num_bytes)
1043                                 break;
1044
1045                         /* check to make sure this extent only has one ref */
1046                         ei = btrfs_item_ptr(leaf, cur_slot,
1047                                             struct btrfs_extent_item);
1048                         if (btrfs_extent_refs(leaf, ei) != 1)
1049                                 break;
1050
1051                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1052                         if (found_key.objectid != tmp->bytenr ||
1053                             found_key.type != BTRFS_EXTENT_REF_KEY ||
1054                             found_key.offset != tmp->orig_parent)
1055                                 break;
1056
1057                         /*
1058                          * the ref is right next to the extent, we can set the
1059                          * ref count to 0 since we will delete them both now
1060                          */
1061                         btrfs_set_extent_refs(leaf, ei, 0);
1062
1063                         /* pin down the bytes for this extent */
1064                         mutex_lock(&info->pinned_mutex);
1065                         ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1066                                              tmp->num_bytes, tmp->level >=
1067                                              BTRFS_FIRST_FREE_OBJECTID);
1068                         mutex_unlock(&info->pinned_mutex);
1069                         BUG_ON(ret < 0);
1070
1071                         /*
1072                          * use the del field to tell if we need to go ahead and
1073                          * free up the extent when we delete the item or not.
1074                          */
1075                         tmp->del = ret;
1076                         bytes_freed += tmp->num_bytes;
1077
1078                         num_to_del += 2;
1079                         cur_slot += 2;
1080                 }
1081                 end = pos;
1082
1083                 /* update the free space counters */
1084                 spin_lock(&info->delalloc_lock);
1085                 super_used = btrfs_super_bytes_used(&info->super_copy);
1086                 btrfs_set_super_bytes_used(&info->super_copy,
1087                                            super_used - bytes_freed);
1088
1089                 root_used = btrfs_root_used(&extent_root->root_item);
1090                 btrfs_set_root_used(&extent_root->root_item,
1091                                     root_used - bytes_freed);
1092                 spin_unlock(&info->delalloc_lock);
1093
1094                 /* delete the items */
1095                 ret = btrfs_del_items(trans, extent_root, path,
1096                                       path->slots[0], num_to_del);
1097                 BUG_ON(ret);
1098
1099                 /*
1100                  * loop through the extents we deleted and do the cleanup work
1101                  * on them
1102                  */
1103                 for (pos = cur, n = pos->next; pos != end;
1104                      pos = n, n = pos->next) {
1105                         struct pending_extent_op *tmp;
1106                         tmp = list_entry(pos, struct pending_extent_op, list);
1107
1108                         /*
1109                          * remember tmp->del tells us wether or not we pinned
1110                          * down the extent
1111                          */
1112                         ret = update_block_group(trans, extent_root,
1113                                                  tmp->bytenr, tmp->num_bytes, 0,
1114                                                  tmp->del);
1115                         BUG_ON(ret);
1116
1117                         list_del_init(&tmp->list);
1118                         unlock_extent(&info->extent_ins, tmp->bytenr,
1119                                       tmp->bytenr + tmp->num_bytes - 1,
1120                                       GFP_NOFS);
1121                         kfree(tmp);
1122                 }
1123         } else if (refs && found_extent) {
1124                 /*
1125                  * the ref and extent were right next to eachother, but the
1126                  * extent still has a ref, so just free the backref and keep
1127                  * going
1128                  */
1129                 ret = remove_extent_backref(trans, extent_root, path);
1130                 BUG_ON(ret);
1131
1132                 list_del_init(&op->list);
1133                 unlock_extent(&info->extent_ins, op->bytenr,
1134                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1135                 kfree(op);
1136         } else {
1137                 /*
1138                  * the extent has multiple refs and the backref we were looking
1139                  * for was not right next to it, so just unlock and go next,
1140                  * we're good to go
1141                  */
1142                 list_del_init(&op->list);
1143                 unlock_extent(&info->extent_ins, op->bytenr,
1144                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1145                 kfree(op);
1146         }
1147
1148         btrfs_release_path(extent_root, path);
1149         if (!list_empty(del_list))
1150                 goto search;
1151
1152 out:
1153         btrfs_free_path(path);
1154         return ret;
1155 }
1156
1157 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1158                                      struct btrfs_root *root, u64 bytenr,
1159                                      u64 orig_parent, u64 parent,
1160                                      u64 orig_root, u64 ref_root,
1161                                      u64 orig_generation, u64 ref_generation,
1162                                      u64 owner_objectid)
1163 {
1164         int ret;
1165         struct btrfs_root *extent_root = root->fs_info->extent_root;
1166         struct btrfs_path *path;
1167
1168         if (root == root->fs_info->extent_root) {
1169                 struct pending_extent_op *extent_op;
1170                 u64 num_bytes;
1171
1172                 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1173                 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1174                 mutex_lock(&root->fs_info->extent_ins_mutex);
1175                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1176                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1177                         u64 priv;
1178                         ret = get_state_private(&root->fs_info->extent_ins,
1179                                                 bytenr, &priv);
1180                         BUG_ON(ret);
1181                         extent_op = (struct pending_extent_op *)
1182                                                         (unsigned long)priv;
1183                         BUG_ON(extent_op->parent != orig_parent);
1184                         BUG_ON(extent_op->generation != orig_generation);
1185
1186                         extent_op->parent = parent;
1187                         extent_op->generation = ref_generation;
1188                 } else {
1189                         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1190                         BUG_ON(!extent_op);
1191
1192                         extent_op->type = PENDING_BACKREF_UPDATE;
1193                         extent_op->bytenr = bytenr;
1194                         extent_op->num_bytes = num_bytes;
1195                         extent_op->parent = parent;
1196                         extent_op->orig_parent = orig_parent;
1197                         extent_op->generation = ref_generation;
1198                         extent_op->orig_generation = orig_generation;
1199                         extent_op->level = (int)owner_objectid;
1200                         INIT_LIST_HEAD(&extent_op->list);
1201                         extent_op->del = 0;
1202
1203                         set_extent_bits(&root->fs_info->extent_ins,
1204                                         bytenr, bytenr + num_bytes - 1,
1205                                         EXTENT_WRITEBACK, GFP_NOFS);
1206                         set_state_private(&root->fs_info->extent_ins,
1207                                           bytenr, (unsigned long)extent_op);
1208                 }
1209                 mutex_unlock(&root->fs_info->extent_ins_mutex);
1210                 return 0;
1211         }
1212
1213         path = btrfs_alloc_path();
1214         if (!path)
1215                 return -ENOMEM;
1216         ret = lookup_extent_backref(trans, extent_root, path,
1217                                     bytenr, orig_parent, orig_root,
1218                                     orig_generation, owner_objectid, 1);
1219         if (ret)
1220                 goto out;
1221         ret = remove_extent_backref(trans, extent_root, path);
1222         if (ret)
1223                 goto out;
1224         ret = insert_extent_backref(trans, extent_root, path, bytenr,
1225                                     parent, ref_root, ref_generation,
1226                                     owner_objectid);
1227         BUG_ON(ret);
1228         finish_current_insert(trans, extent_root, 0);
1229         del_pending_extents(trans, extent_root, 0);
1230 out:
1231         btrfs_free_path(path);
1232         return ret;
1233 }
1234
1235 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1236                             struct btrfs_root *root, u64 bytenr,
1237                             u64 orig_parent, u64 parent,
1238                             u64 ref_root, u64 ref_generation,
1239                             u64 owner_objectid)
1240 {
1241         int ret;
1242         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1243             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1244                 return 0;
1245         ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1246                                         parent, ref_root, ref_root,
1247                                         ref_generation, ref_generation,
1248                                         owner_objectid);
1249         return ret;
1250 }
1251
1252 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1253                                   struct btrfs_root *root, u64 bytenr,
1254                                   u64 orig_parent, u64 parent,
1255                                   u64 orig_root, u64 ref_root,
1256                                   u64 orig_generation, u64 ref_generation,
1257                                   u64 owner_objectid)
1258 {
1259         struct btrfs_path *path;
1260         int ret;
1261         struct btrfs_key key;
1262         struct extent_buffer *l;
1263         struct btrfs_extent_item *item;
1264         u32 refs;
1265
1266         path = btrfs_alloc_path();
1267         if (!path)
1268                 return -ENOMEM;
1269
1270         path->reada = 1;
1271         key.objectid = bytenr;
1272         key.type = BTRFS_EXTENT_ITEM_KEY;
1273         key.offset = (u64)-1;
1274
1275         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1276                                 0, 1);
1277         if (ret < 0)
1278                 return ret;
1279         BUG_ON(ret == 0 || path->slots[0] == 0);
1280
1281         path->slots[0]--;
1282         l = path->nodes[0];
1283
1284         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1285         if (key.objectid != bytenr) {
1286                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1287                 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
1288                        (unsigned long long)bytenr,
1289                        (unsigned long long)key.objectid);
1290                 BUG();
1291         }
1292         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1293
1294         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1295         refs = btrfs_extent_refs(l, item);
1296         btrfs_set_extent_refs(l, item, refs + 1);
1297         btrfs_mark_buffer_dirty(path->nodes[0]);
1298
1299         btrfs_release_path(root->fs_info->extent_root, path);
1300
1301         path->reada = 1;
1302         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1303                                     path, bytenr, parent,
1304                                     ref_root, ref_generation,
1305                                     owner_objectid);
1306         BUG_ON(ret);
1307         finish_current_insert(trans, root->fs_info->extent_root, 0);
1308         del_pending_extents(trans, root->fs_info->extent_root, 0);
1309
1310         btrfs_free_path(path);
1311         return 0;
1312 }
1313
1314 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1315                          struct btrfs_root *root,
1316                          u64 bytenr, u64 num_bytes, u64 parent,
1317                          u64 ref_root, u64 ref_generation,
1318                          u64 owner_objectid)
1319 {
1320         int ret;
1321         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1322             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1323                 return 0;
1324         ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1325                                      0, ref_root, 0, ref_generation,
1326                                      owner_objectid);
1327         return ret;
1328 }
1329
1330 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1331                          struct btrfs_root *root)
1332 {
1333         finish_current_insert(trans, root->fs_info->extent_root, 1);
1334         del_pending_extents(trans, root->fs_info->extent_root, 1);
1335         return 0;
1336 }
1337
1338 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1339                             struct btrfs_root *root, u64 bytenr,
1340                             u64 num_bytes, u32 *refs)
1341 {
1342         struct btrfs_path *path;
1343         int ret;
1344         struct btrfs_key key;
1345         struct extent_buffer *l;
1346         struct btrfs_extent_item *item;
1347
1348         WARN_ON(num_bytes < root->sectorsize);
1349         path = btrfs_alloc_path();
1350         path->reada = 1;
1351         key.objectid = bytenr;
1352         key.offset = num_bytes;
1353         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1354         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1355                                 0, 0);
1356         if (ret < 0)
1357                 goto out;
1358         if (ret != 0) {
1359                 btrfs_print_leaf(root, path->nodes[0]);
1360                 printk(KERN_INFO "btrfs failed to find block number %llu\n",
1361                        (unsigned long long)bytenr);
1362                 BUG();
1363         }
1364         l = path->nodes[0];
1365         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1366         *refs = btrfs_extent_refs(l, item);
1367 out:
1368         btrfs_free_path(path);
1369         return 0;
1370 }
1371
1372 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1373                           struct btrfs_root *root, u64 objectid, u64 bytenr)
1374 {
1375         struct btrfs_root *extent_root = root->fs_info->extent_root;
1376         struct btrfs_path *path;
1377         struct extent_buffer *leaf;
1378         struct btrfs_extent_ref *ref_item;
1379         struct btrfs_key key;
1380         struct btrfs_key found_key;
1381         u64 ref_root;
1382         u64 last_snapshot;
1383         u32 nritems;
1384         int ret;
1385
1386         key.objectid = bytenr;
1387         key.offset = (u64)-1;
1388         key.type = BTRFS_EXTENT_ITEM_KEY;
1389
1390         path = btrfs_alloc_path();
1391         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1392         if (ret < 0)
1393                 goto out;
1394         BUG_ON(ret == 0);
1395
1396         ret = -ENOENT;
1397         if (path->slots[0] == 0)
1398                 goto out;
1399
1400         path->slots[0]--;
1401         leaf = path->nodes[0];
1402         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1403
1404         if (found_key.objectid != bytenr ||
1405             found_key.type != BTRFS_EXTENT_ITEM_KEY)
1406                 goto out;
1407
1408         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1409         while (1) {
1410                 leaf = path->nodes[0];
1411                 nritems = btrfs_header_nritems(leaf);
1412                 if (path->slots[0] >= nritems) {
1413                         ret = btrfs_next_leaf(extent_root, path);
1414                         if (ret < 0)
1415                                 goto out;
1416                         if (ret == 0)
1417                                 continue;
1418                         break;
1419                 }
1420                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1421                 if (found_key.objectid != bytenr)
1422                         break;
1423
1424                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1425                         path->slots[0]++;
1426                         continue;
1427                 }
1428
1429                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1430                                           struct btrfs_extent_ref);
1431                 ref_root = btrfs_ref_root(leaf, ref_item);
1432                 if ((ref_root != root->root_key.objectid &&
1433                      ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1434                      objectid != btrfs_ref_objectid(leaf, ref_item)) {
1435                         ret = 1;
1436                         goto out;
1437                 }
1438                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1439                         ret = 1;
1440                         goto out;
1441                 }
1442
1443                 path->slots[0]++;
1444         }
1445         ret = 0;
1446 out:
1447         btrfs_free_path(path);
1448         return ret;
1449 }
1450
1451 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1452                     struct extent_buffer *buf, u32 nr_extents)
1453 {
1454         struct btrfs_key key;
1455         struct btrfs_file_extent_item *fi;
1456         u64 root_gen;
1457         u32 nritems;
1458         int i;
1459         int level;
1460         int ret = 0;
1461         int shared = 0;
1462
1463         if (!root->ref_cows)
1464                 return 0;
1465
1466         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1467                 shared = 0;
1468                 root_gen = root->root_key.offset;
1469         } else {
1470                 shared = 1;
1471                 root_gen = trans->transid - 1;
1472         }
1473
1474         level = btrfs_header_level(buf);
1475         nritems = btrfs_header_nritems(buf);
1476
1477         if (level == 0) {
1478                 struct btrfs_leaf_ref *ref;
1479                 struct btrfs_extent_info *info;
1480
1481                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1482                 if (!ref) {
1483                         ret = -ENOMEM;
1484                         goto out;
1485                 }
1486
1487                 ref->root_gen = root_gen;
1488                 ref->bytenr = buf->start;
1489                 ref->owner = btrfs_header_owner(buf);
1490                 ref->generation = btrfs_header_generation(buf);
1491                 ref->nritems = nr_extents;
1492                 info = ref->extents;
1493
1494                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1495                         u64 disk_bytenr;
1496                         btrfs_item_key_to_cpu(buf, &key, i);
1497                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1498                                 continue;
1499                         fi = btrfs_item_ptr(buf, i,
1500                                             struct btrfs_file_extent_item);
1501                         if (btrfs_file_extent_type(buf, fi) ==
1502                             BTRFS_FILE_EXTENT_INLINE)
1503                                 continue;
1504                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1505                         if (disk_bytenr == 0)
1506                                 continue;
1507
1508                         info->bytenr = disk_bytenr;
1509                         info->num_bytes =
1510                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1511                         info->objectid = key.objectid;
1512                         info->offset = key.offset;
1513                         info++;
1514                 }
1515
1516                 ret = btrfs_add_leaf_ref(root, ref, shared);
1517                 if (ret == -EEXIST && shared) {
1518                         struct btrfs_leaf_ref *old;
1519                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1520                         BUG_ON(!old);
1521                         btrfs_remove_leaf_ref(root, old);
1522                         btrfs_free_leaf_ref(root, old);
1523                         ret = btrfs_add_leaf_ref(root, ref, shared);
1524                 }
1525                 WARN_ON(ret);
1526                 btrfs_free_leaf_ref(root, ref);
1527         }
1528 out:
1529         return ret;
1530 }
1531
1532 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1533                   struct extent_buffer *orig_buf, struct extent_buffer *buf,
1534                   u32 *nr_extents)
1535 {
1536         u64 bytenr;
1537         u64 ref_root;
1538         u64 orig_root;
1539         u64 ref_generation;
1540         u64 orig_generation;
1541         u32 nritems;
1542         u32 nr_file_extents = 0;
1543         struct btrfs_key key;
1544         struct btrfs_file_extent_item *fi;
1545         int i;
1546         int level;
1547         int ret = 0;
1548         int faili = 0;
1549         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1550                             u64, u64, u64, u64, u64, u64, u64, u64);
1551
1552         ref_root = btrfs_header_owner(buf);
1553         ref_generation = btrfs_header_generation(buf);
1554         orig_root = btrfs_header_owner(orig_buf);
1555         orig_generation = btrfs_header_generation(orig_buf);
1556
1557         nritems = btrfs_header_nritems(buf);
1558         level = btrfs_header_level(buf);
1559
1560         if (root->ref_cows) {
1561                 process_func = __btrfs_inc_extent_ref;
1562         } else {
1563                 if (level == 0 &&
1564                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1565                         goto out;
1566                 if (level != 0 &&
1567                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1568                         goto out;
1569                 process_func = __btrfs_update_extent_ref;
1570         }
1571
1572         for (i = 0; i < nritems; i++) {
1573                 cond_resched();
1574                 if (level == 0) {
1575                         btrfs_item_key_to_cpu(buf, &key, i);
1576                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1577                                 continue;
1578                         fi = btrfs_item_ptr(buf, i,
1579                                             struct btrfs_file_extent_item);
1580                         if (btrfs_file_extent_type(buf, fi) ==
1581                             BTRFS_FILE_EXTENT_INLINE)
1582                                 continue;
1583                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1584                         if (bytenr == 0)
1585                                 continue;
1586
1587                         nr_file_extents++;
1588
1589                         ret = process_func(trans, root, bytenr,
1590                                            orig_buf->start, buf->start,
1591                                            orig_root, ref_root,
1592                                            orig_generation, ref_generation,
1593                                            key.objectid);
1594
1595                         if (ret) {
1596                                 faili = i;
1597                                 WARN_ON(1);
1598                                 goto fail;
1599                         }
1600                 } else {
1601                         bytenr = btrfs_node_blockptr(buf, i);
1602                         ret = process_func(trans, root, bytenr,
1603                                            orig_buf->start, buf->start,
1604                                            orig_root, ref_root,
1605                                            orig_generation, ref_generation,
1606                                            level - 1);
1607                         if (ret) {
1608                                 faili = i;
1609                                 WARN_ON(1);
1610                                 goto fail;
1611                         }
1612                 }
1613         }
1614 out:
1615         if (nr_extents) {
1616                 if (level == 0)
1617                         *nr_extents = nr_file_extents;
1618                 else
1619                         *nr_extents = nritems;
1620         }
1621         return 0;
1622 fail:
1623         WARN_ON(1);
1624         return ret;
1625 }
1626
1627 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1628                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1629                      struct extent_buffer *buf, int start_slot, int nr)
1630
1631 {
1632         u64 bytenr;
1633         u64 ref_root;
1634         u64 orig_root;
1635         u64 ref_generation;
1636         u64 orig_generation;
1637         struct btrfs_key key;
1638         struct btrfs_file_extent_item *fi;
1639         int i;
1640         int ret;
1641         int slot;
1642         int level;
1643
1644         BUG_ON(start_slot < 0);
1645         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1646
1647         ref_root = btrfs_header_owner(buf);
1648         ref_generation = btrfs_header_generation(buf);
1649         orig_root = btrfs_header_owner(orig_buf);
1650         orig_generation = btrfs_header_generation(orig_buf);
1651         level = btrfs_header_level(buf);
1652
1653         if (!root->ref_cows) {
1654                 if (level == 0 &&
1655                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1656                         return 0;
1657                 if (level != 0 &&
1658                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1659                         return 0;
1660         }
1661
1662         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1663                 cond_resched();
1664                 if (level == 0) {
1665                         btrfs_item_key_to_cpu(buf, &key, slot);
1666                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1667                                 continue;
1668                         fi = btrfs_item_ptr(buf, slot,
1669                                             struct btrfs_file_extent_item);
1670                         if (btrfs_file_extent_type(buf, fi) ==
1671                             BTRFS_FILE_EXTENT_INLINE)
1672                                 continue;
1673                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1674                         if (bytenr == 0)
1675                                 continue;
1676                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1677                                             orig_buf->start, buf->start,
1678                                             orig_root, ref_root,
1679                                             orig_generation, ref_generation,
1680                                             key.objectid);
1681                         if (ret)
1682                                 goto fail;
1683                 } else {
1684                         bytenr = btrfs_node_blockptr(buf, slot);
1685                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1686                                             orig_buf->start, buf->start,
1687                                             orig_root, ref_root,
1688                                             orig_generation, ref_generation,
1689                                             level - 1);
1690                         if (ret)
1691                                 goto fail;
1692                 }
1693         }
1694         return 0;
1695 fail:
1696         WARN_ON(1);
1697         return -1;
1698 }
1699
1700 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1701                                  struct btrfs_root *root,
1702                                  struct btrfs_path *path,
1703                                  struct btrfs_block_group_cache *cache)
1704 {
1705         int ret;
1706         int pending_ret;
1707         struct btrfs_root *extent_root = root->fs_info->extent_root;
1708         unsigned long bi;
1709         struct extent_buffer *leaf;
1710
1711         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1712         if (ret < 0)
1713                 goto fail;
1714         BUG_ON(ret);
1715
1716         leaf = path->nodes[0];
1717         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1718         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1719         btrfs_mark_buffer_dirty(leaf);
1720         btrfs_release_path(extent_root, path);
1721 fail:
1722         finish_current_insert(trans, extent_root, 0);
1723         pending_ret = del_pending_extents(trans, extent_root, 0);
1724         if (ret)
1725                 return ret;
1726         if (pending_ret)
1727                 return pending_ret;
1728         return 0;
1729
1730 }
1731
1732 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1733                                    struct btrfs_root *root)
1734 {
1735         struct btrfs_block_group_cache *cache, *entry;
1736         struct rb_node *n;
1737         int err = 0;
1738         int werr = 0;
1739         struct btrfs_path *path;
1740         u64 last = 0;
1741
1742         path = btrfs_alloc_path();
1743         if (!path)
1744                 return -ENOMEM;
1745
1746         while (1) {
1747                 cache = NULL;
1748                 spin_lock(&root->fs_info->block_group_cache_lock);
1749                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1750                      n; n = rb_next(n)) {
1751                         entry = rb_entry(n, struct btrfs_block_group_cache,
1752                                          cache_node);
1753                         if (entry->dirty) {
1754                                 cache = entry;
1755                                 break;
1756                         }
1757                 }
1758                 spin_unlock(&root->fs_info->block_group_cache_lock);
1759
1760                 if (!cache)
1761                         break;
1762
1763                 cache->dirty = 0;
1764                 last += cache->key.offset;
1765
1766                 err = write_one_cache_group(trans, root,
1767                                             path, cache);
1768                 /*
1769                  * if we fail to write the cache group, we want
1770                  * to keep it marked dirty in hopes that a later
1771                  * write will work
1772                  */
1773                 if (err) {
1774                         werr = err;
1775                         continue;
1776                 }
1777         }
1778         btrfs_free_path(path);
1779         return werr;
1780 }
1781
1782 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1783 {
1784         struct btrfs_block_group_cache *block_group;
1785         int readonly = 0;
1786
1787         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1788         if (!block_group || block_group->ro)
1789                 readonly = 1;
1790         if (block_group)
1791                 put_block_group(block_group);
1792         return readonly;
1793 }
1794
1795 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1796                              u64 total_bytes, u64 bytes_used,
1797                              struct btrfs_space_info **space_info)
1798 {
1799         struct btrfs_space_info *found;
1800
1801         found = __find_space_info(info, flags);
1802         if (found) {
1803                 spin_lock(&found->lock);
1804                 found->total_bytes += total_bytes;
1805                 found->bytes_used += bytes_used;
1806                 found->full = 0;
1807                 spin_unlock(&found->lock);
1808                 *space_info = found;
1809                 return 0;
1810         }
1811         found = kzalloc(sizeof(*found), GFP_NOFS);
1812         if (!found)
1813                 return -ENOMEM;
1814
1815         list_add(&found->list, &info->space_info);
1816         INIT_LIST_HEAD(&found->block_groups);
1817         init_rwsem(&found->groups_sem);
1818         spin_lock_init(&found->lock);
1819         found->flags = flags;
1820         found->total_bytes = total_bytes;
1821         found->bytes_used = bytes_used;
1822         found->bytes_pinned = 0;
1823         found->bytes_reserved = 0;
1824         found->bytes_readonly = 0;
1825         found->full = 0;
1826         found->force_alloc = 0;
1827         *space_info = found;
1828         return 0;
1829 }
1830
1831 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1832 {
1833         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1834                                    BTRFS_BLOCK_GROUP_RAID1 |
1835                                    BTRFS_BLOCK_GROUP_RAID10 |
1836                                    BTRFS_BLOCK_GROUP_DUP);
1837         if (extra_flags) {
1838                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1839                         fs_info->avail_data_alloc_bits |= extra_flags;
1840                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1841                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1842                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1843                         fs_info->avail_system_alloc_bits |= extra_flags;
1844         }
1845 }
1846
1847 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1848 {
1849         spin_lock(&cache->space_info->lock);
1850         spin_lock(&cache->lock);
1851         if (!cache->ro) {
1852                 cache->space_info->bytes_readonly += cache->key.offset -
1853                                         btrfs_block_group_used(&cache->item);
1854                 cache->ro = 1;
1855         }
1856         spin_unlock(&cache->lock);
1857         spin_unlock(&cache->space_info->lock);
1858 }
1859
1860 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1861 {
1862         u64 num_devices = root->fs_info->fs_devices->rw_devices;
1863
1864         if (num_devices == 1)
1865                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1866         if (num_devices < 4)
1867                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1868
1869         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1870             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1871                       BTRFS_BLOCK_GROUP_RAID10))) {
1872                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1873         }
1874
1875         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1876             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1877                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1878         }
1879
1880         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1881             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1882              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1883              (flags & BTRFS_BLOCK_GROUP_DUP)))
1884                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1885         return flags;
1886 }
1887
1888 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1889                           struct btrfs_root *extent_root, u64 alloc_bytes,
1890                           u64 flags, int force)
1891 {
1892         struct btrfs_space_info *space_info;
1893         u64 thresh;
1894         int ret = 0;
1895
1896         mutex_lock(&extent_root->fs_info->chunk_mutex);
1897
1898         flags = btrfs_reduce_alloc_profile(extent_root, flags);
1899
1900         space_info = __find_space_info(extent_root->fs_info, flags);
1901         if (!space_info) {
1902                 ret = update_space_info(extent_root->fs_info, flags,
1903                                         0, 0, &space_info);
1904                 BUG_ON(ret);
1905         }
1906         BUG_ON(!space_info);
1907
1908         spin_lock(&space_info->lock);
1909         if (space_info->force_alloc) {
1910                 force = 1;
1911                 space_info->force_alloc = 0;
1912         }
1913         if (space_info->full) {
1914                 spin_unlock(&space_info->lock);
1915                 goto out;
1916         }
1917
1918         thresh = space_info->total_bytes - space_info->bytes_readonly;
1919         thresh = div_factor(thresh, 6);
1920         if (!force &&
1921            (space_info->bytes_used + space_info->bytes_pinned +
1922             space_info->bytes_reserved + alloc_bytes) < thresh) {
1923                 spin_unlock(&space_info->lock);
1924                 goto out;
1925         }
1926         spin_unlock(&space_info->lock);
1927
1928         ret = btrfs_alloc_chunk(trans, extent_root, flags);
1929         if (ret)
1930                 space_info->full = 1;
1931 out:
1932         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1933         return ret;
1934 }
1935
1936 static int update_block_group(struct btrfs_trans_handle *trans,
1937                               struct btrfs_root *root,
1938                               u64 bytenr, u64 num_bytes, int alloc,
1939                               int mark_free)
1940 {
1941         struct btrfs_block_group_cache *cache;
1942         struct btrfs_fs_info *info = root->fs_info;
1943         u64 total = num_bytes;
1944         u64 old_val;
1945         u64 byte_in_group;
1946
1947         while (total) {
1948                 cache = btrfs_lookup_block_group(info, bytenr);
1949                 if (!cache)
1950                         return -1;
1951                 byte_in_group = bytenr - cache->key.objectid;
1952                 WARN_ON(byte_in_group > cache->key.offset);
1953
1954                 spin_lock(&cache->space_info->lock);
1955                 spin_lock(&cache->lock);
1956                 cache->dirty = 1;
1957                 old_val = btrfs_block_group_used(&cache->item);
1958                 num_bytes = min(total, cache->key.offset - byte_in_group);
1959                 if (alloc) {
1960                         old_val += num_bytes;
1961                         cache->space_info->bytes_used += num_bytes;
1962                         if (cache->ro)
1963                                 cache->space_info->bytes_readonly -= num_bytes;
1964                         btrfs_set_block_group_used(&cache->item, old_val);
1965                         spin_unlock(&cache->lock);
1966                         spin_unlock(&cache->space_info->lock);
1967                 } else {
1968                         old_val -= num_bytes;
1969                         cache->space_info->bytes_used -= num_bytes;
1970                         if (cache->ro)
1971                                 cache->space_info->bytes_readonly += num_bytes;
1972                         btrfs_set_block_group_used(&cache->item, old_val);
1973                         spin_unlock(&cache->lock);
1974                         spin_unlock(&cache->space_info->lock);
1975                         if (mark_free) {
1976                                 int ret;
1977
1978                                 ret = btrfs_discard_extent(root, bytenr,
1979                                                            num_bytes);
1980                                 WARN_ON(ret);
1981
1982                                 ret = btrfs_add_free_space(cache, bytenr,
1983                                                            num_bytes);
1984                                 WARN_ON(ret);
1985                         }
1986                 }
1987                 put_block_group(cache);
1988                 total -= num_bytes;
1989                 bytenr += num_bytes;
1990         }
1991         return 0;
1992 }
1993
1994 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1995 {
1996         struct btrfs_block_group_cache *cache;
1997         u64 bytenr;
1998
1999         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2000         if (!cache)
2001                 return 0;
2002
2003         bytenr = cache->key.objectid;
2004         put_block_group(cache);
2005
2006         return bytenr;
2007 }
2008
2009 int btrfs_update_pinned_extents(struct btrfs_root *root,
2010                                 u64 bytenr, u64 num, int pin)
2011 {
2012         u64 len;
2013         struct btrfs_block_group_cache *cache;
2014         struct btrfs_fs_info *fs_info = root->fs_info;
2015
2016         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2017         if (pin) {
2018                 set_extent_dirty(&fs_info->pinned_extents,
2019                                 bytenr, bytenr + num - 1, GFP_NOFS);
2020         } else {
2021                 clear_extent_dirty(&fs_info->pinned_extents,
2022                                 bytenr, bytenr + num - 1, GFP_NOFS);
2023         }
2024         while (num > 0) {
2025                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2026                 BUG_ON(!cache);
2027                 len = min(num, cache->key.offset -
2028                           (bytenr - cache->key.objectid));
2029                 if (pin) {
2030                         spin_lock(&cache->space_info->lock);
2031                         spin_lock(&cache->lock);
2032                         cache->pinned += len;
2033                         cache->space_info->bytes_pinned += len;
2034                         spin_unlock(&cache->lock);
2035                         spin_unlock(&cache->space_info->lock);
2036                         fs_info->total_pinned += len;
2037                 } else {
2038                         spin_lock(&cache->space_info->lock);
2039                         spin_lock(&cache->lock);
2040                         cache->pinned -= len;
2041                         cache->space_info->bytes_pinned -= len;
2042                         spin_unlock(&cache->lock);
2043                         spin_unlock(&cache->space_info->lock);
2044                         fs_info->total_pinned -= len;
2045                         if (cache->cached)
2046                                 btrfs_add_free_space(cache, bytenr, len);
2047                 }
2048                 put_block_group(cache);
2049                 bytenr += len;
2050                 num -= len;
2051         }
2052         return 0;
2053 }
2054
2055 static int update_reserved_extents(struct btrfs_root *root,
2056                                    u64 bytenr, u64 num, int reserve)
2057 {
2058         u64 len;
2059         struct btrfs_block_group_cache *cache;
2060         struct btrfs_fs_info *fs_info = root->fs_info;
2061
2062         while (num > 0) {
2063                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2064                 BUG_ON(!cache);
2065                 len = min(num, cache->key.offset -
2066                           (bytenr - cache->key.objectid));
2067
2068                 spin_lock(&cache->space_info->lock);
2069                 spin_lock(&cache->lock);
2070                 if (reserve) {
2071                         cache->reserved += len;
2072                         cache->space_info->bytes_reserved += len;
2073                 } else {
2074                         cache->reserved -= len;
2075                         cache->space_info->bytes_reserved -= len;
2076                 }
2077                 spin_unlock(&cache->lock);
2078                 spin_unlock(&cache->space_info->lock);
2079                 put_block_group(cache);
2080                 bytenr += len;
2081                 num -= len;
2082         }
2083         return 0;
2084 }
2085
2086 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2087 {
2088         u64 last = 0;
2089         u64 start;
2090         u64 end;
2091         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2092         int ret;
2093
2094         mutex_lock(&root->fs_info->pinned_mutex);
2095         while (1) {
2096                 ret = find_first_extent_bit(pinned_extents, last,
2097                                             &start, &end, EXTENT_DIRTY);
2098                 if (ret)
2099                         break;
2100                 set_extent_dirty(copy, start, end, GFP_NOFS);
2101                 last = end + 1;
2102         }
2103         mutex_unlock(&root->fs_info->pinned_mutex);
2104         return 0;
2105 }
2106
2107 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2108                                struct btrfs_root *root,
2109                                struct extent_io_tree *unpin)
2110 {
2111         u64 start;
2112         u64 end;
2113         int ret;
2114
2115         mutex_lock(&root->fs_info->pinned_mutex);
2116         while (1) {
2117                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2118                                             EXTENT_DIRTY);
2119                 if (ret)
2120                         break;
2121
2122                 ret = btrfs_discard_extent(root, start, end + 1 - start);
2123
2124                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2125                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2126
2127                 if (need_resched()) {
2128                         mutex_unlock(&root->fs_info->pinned_mutex);
2129                         cond_resched();
2130                         mutex_lock(&root->fs_info->pinned_mutex);
2131                 }
2132         }
2133         mutex_unlock(&root->fs_info->pinned_mutex);
2134         return ret;
2135 }
2136
2137 static int finish_current_insert(struct btrfs_trans_handle *trans,
2138                                  struct btrfs_root *extent_root, int all)
2139 {
2140         u64 start;
2141         u64 end;
2142         u64 priv;
2143         u64 search = 0;
2144         u64 skipped = 0;
2145         struct btrfs_fs_info *info = extent_root->fs_info;
2146         struct btrfs_path *path;
2147         struct pending_extent_op *extent_op, *tmp;
2148         struct list_head insert_list, update_list;
2149         int ret;
2150         int num_inserts = 0, max_inserts;
2151
2152         path = btrfs_alloc_path();
2153         INIT_LIST_HEAD(&insert_list);
2154         INIT_LIST_HEAD(&update_list);
2155
2156         max_inserts = extent_root->leafsize /
2157                 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2158                  sizeof(struct btrfs_extent_ref) +
2159                  sizeof(struct btrfs_extent_item));
2160 again:
2161         mutex_lock(&info->extent_ins_mutex);
2162         while (1) {
2163                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2164                                             &end, EXTENT_WRITEBACK);
2165                 if (ret) {
2166                         if (skipped && all && !num_inserts) {
2167                                 skipped = 0;
2168                                 search = 0;
2169                                 continue;
2170                         }
2171                         mutex_unlock(&info->extent_ins_mutex);
2172                         break;
2173                 }
2174
2175                 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2176                 if (!ret) {
2177                         skipped = 1;
2178                         search = end + 1;
2179                         if (need_resched()) {
2180                                 mutex_unlock(&info->extent_ins_mutex);
2181                                 cond_resched();
2182                                 mutex_lock(&info->extent_ins_mutex);
2183                         }
2184                         continue;
2185                 }
2186
2187                 ret = get_state_private(&info->extent_ins, start, &priv);
2188                 BUG_ON(ret);
2189                 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2190
2191                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2192                         num_inserts++;
2193                         list_add_tail(&extent_op->list, &insert_list);
2194                         search = end + 1;
2195                         if (num_inserts == max_inserts) {
2196                                 mutex_unlock(&info->extent_ins_mutex);
2197                                 break;
2198                         }
2199                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2200                         list_add_tail(&extent_op->list, &update_list);
2201                         search = end + 1;
2202                 } else {
2203                         BUG();
2204                 }
2205         }
2206
2207         /*
2208          * process the update list, clear the writeback bit for it, and if
2209          * somebody marked this thing for deletion then just unlock it and be
2210          * done, the free_extents will handle it
2211          */
2212         mutex_lock(&info->extent_ins_mutex);
2213         list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2214                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2215                                   extent_op->bytenr + extent_op->num_bytes - 1,
2216                                   EXTENT_WRITEBACK, GFP_NOFS);
2217                 if (extent_op->del) {
2218                         list_del_init(&extent_op->list);
2219                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2220                                       extent_op->bytenr + extent_op->num_bytes
2221                                       - 1, GFP_NOFS);
2222                         kfree(extent_op);
2223                 }
2224         }
2225         mutex_unlock(&info->extent_ins_mutex);
2226
2227         /*
2228          * still have things left on the update list, go ahead an update
2229          * everything
2230          */
2231         if (!list_empty(&update_list)) {
2232                 ret = update_backrefs(trans, extent_root, path, &update_list);
2233                 BUG_ON(ret);
2234         }
2235
2236         /*
2237          * if no inserts need to be done, but we skipped some extents and we
2238          * need to make sure everything is cleaned then reset everything and
2239          * go back to the beginning
2240          */
2241         if (!num_inserts && all && skipped) {
2242                 search = 0;
2243                 skipped = 0;
2244                 INIT_LIST_HEAD(&update_list);
2245                 INIT_LIST_HEAD(&insert_list);
2246                 goto again;
2247         } else if (!num_inserts) {
2248                 goto out;
2249         }
2250
2251         /*
2252          * process the insert extents list.  Again if we are deleting this
2253          * extent, then just unlock it, pin down the bytes if need be, and be
2254          * done with it.  Saves us from having to actually insert the extent
2255          * into the tree and then subsequently come along and delete it
2256          */
2257         mutex_lock(&info->extent_ins_mutex);
2258         list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2259                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2260                                   extent_op->bytenr + extent_op->num_bytes - 1,
2261                                   EXTENT_WRITEBACK, GFP_NOFS);
2262                 if (extent_op->del) {
2263                         u64 used;
2264                         list_del_init(&extent_op->list);
2265                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2266                                       extent_op->bytenr + extent_op->num_bytes
2267                                       - 1, GFP_NOFS);
2268
2269                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2270                         ret = pin_down_bytes(trans, extent_root,
2271                                              extent_op->bytenr,
2272                                              extent_op->num_bytes, 0);
2273                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2274
2275                         spin_lock(&info->delalloc_lock);
2276                         used = btrfs_super_bytes_used(&info->super_copy);
2277                         btrfs_set_super_bytes_used(&info->super_copy,
2278                                         used - extent_op->num_bytes);
2279                         used = btrfs_root_used(&extent_root->root_item);
2280                         btrfs_set_root_used(&extent_root->root_item,
2281                                         used - extent_op->num_bytes);
2282                         spin_unlock(&info->delalloc_lock);
2283
2284                         ret = update_block_group(trans, extent_root,
2285                                                  extent_op->bytenr,
2286                                                  extent_op->num_bytes,
2287                                                  0, ret > 0);
2288                         BUG_ON(ret);
2289                         kfree(extent_op);
2290                         num_inserts--;
2291                 }
2292         }
2293         mutex_unlock(&info->extent_ins_mutex);
2294
2295         ret = insert_extents(trans, extent_root, path, &insert_list,
2296                              num_inserts);
2297         BUG_ON(ret);
2298
2299         /*
2300          * if we broke out of the loop in order to insert stuff because we hit
2301          * the maximum number of inserts at a time we can handle, then loop
2302          * back and pick up where we left off
2303          */
2304         if (num_inserts == max_inserts) {
2305                 INIT_LIST_HEAD(&insert_list);
2306                 INIT_LIST_HEAD(&update_list);
2307                 num_inserts = 0;
2308                 goto again;
2309         }
2310
2311         /*
2312          * again, if we need to make absolutely sure there are no more pending
2313          * extent operations left and we know that we skipped some, go back to
2314          * the beginning and do it all again
2315          */
2316         if (all && skipped) {
2317                 INIT_LIST_HEAD(&insert_list);
2318                 INIT_LIST_HEAD(&update_list);
2319                 search = 0;
2320                 skipped = 0;
2321                 num_inserts = 0;
2322                 goto again;
2323         }
2324 out:
2325         btrfs_free_path(path);
2326         return 0;
2327 }
2328
2329 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2330                           struct btrfs_root *root,
2331                           u64 bytenr, u64 num_bytes, int is_data)
2332 {
2333         int err = 0;
2334         struct extent_buffer *buf;
2335
2336         if (is_data)
2337                 goto pinit;
2338
2339         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2340         if (!buf)
2341                 goto pinit;
2342
2343         /* we can reuse a block if it hasn't been written
2344          * and it is from this transaction.  We can't
2345          * reuse anything from the tree log root because
2346          * it has tiny sub-transactions.
2347          */
2348         if (btrfs_buffer_uptodate(buf, 0) &&
2349             btrfs_try_tree_lock(buf)) {
2350                 u64 header_owner = btrfs_header_owner(buf);
2351                 u64 header_transid = btrfs_header_generation(buf);
2352                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2353                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2354                     header_transid == trans->transid &&
2355                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2356                         clean_tree_block(NULL, root, buf);
2357                         btrfs_tree_unlock(buf);
2358                         free_extent_buffer(buf);
2359                         return 1;
2360                 }
2361                 btrfs_tree_unlock(buf);
2362         }
2363         free_extent_buffer(buf);
2364 pinit:
2365         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2366
2367         BUG_ON(err < 0);
2368         return 0;
2369 }
2370
2371 /*
2372  * remove an extent from the root, returns 0 on success
2373  */
2374 static int __free_extent(struct btrfs_trans_handle *trans,
2375                          struct btrfs_root *root,
2376                          u64 bytenr, u64 num_bytes, u64 parent,
2377                          u64 root_objectid, u64 ref_generation,
2378                          u64 owner_objectid, int pin, int mark_free)
2379 {
2380         struct btrfs_path *path;
2381         struct btrfs_key key;
2382         struct btrfs_fs_info *info = root->fs_info;
2383         struct btrfs_root *extent_root = info->extent_root;
2384         struct extent_buffer *leaf;
2385         int ret;
2386         int extent_slot = 0;
2387         int found_extent = 0;
2388         int num_to_del = 1;
2389         struct btrfs_extent_item *ei;
2390         u32 refs;
2391
2392         key.objectid = bytenr;
2393         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2394         key.offset = num_bytes;
2395         path = btrfs_alloc_path();
2396         if (!path)
2397                 return -ENOMEM;
2398
2399         path->reada = 1;
2400         ret = lookup_extent_backref(trans, extent_root, path,
2401                                     bytenr, parent, root_objectid,
2402                                     ref_generation, owner_objectid, 1);
2403         if (ret == 0) {
2404                 struct btrfs_key found_key;
2405                 extent_slot = path->slots[0];
2406                 while (extent_slot > 0) {
2407                         extent_slot--;
2408                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2409                                               extent_slot);
2410                         if (found_key.objectid != bytenr)
2411                                 break;
2412                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2413                             found_key.offset == num_bytes) {
2414                                 found_extent = 1;
2415                                 break;
2416                         }
2417                         if (path->slots[0] - extent_slot > 5)
2418                                 break;
2419                 }
2420                 if (!found_extent) {
2421                         ret = remove_extent_backref(trans, extent_root, path);
2422                         BUG_ON(ret);
2423                         btrfs_release_path(extent_root, path);
2424                         ret = btrfs_search_slot(trans, extent_root,
2425                                                 &key, path, -1, 1);
2426                         if (ret) {
2427                                 printk(KERN_ERR "umm, got %d back from search"
2428                                        ", was looking for %llu\n", ret,
2429                                        (unsigned long long)bytenr);
2430                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2431                         }
2432                         BUG_ON(ret);
2433                         extent_slot = path->slots[0];
2434                 }
2435         } else {
2436                 btrfs_print_leaf(extent_root, path->nodes[0]);
2437                 WARN_ON(1);
2438                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2439                        "root %llu gen %llu owner %llu\n",
2440                        (unsigned long long)bytenr,
2441                        (unsigned long long)root_objectid,
2442                        (unsigned long long)ref_generation,
2443                        (unsigned long long)owner_objectid);
2444         }
2445
2446         leaf = path->nodes[0];
2447         ei = btrfs_item_ptr(leaf, extent_slot,
2448                             struct btrfs_extent_item);
2449         refs = btrfs_extent_refs(leaf, ei);
2450         BUG_ON(refs == 0);
2451         refs -= 1;
2452         btrfs_set_extent_refs(leaf, ei, refs);
2453
2454         btrfs_mark_buffer_dirty(leaf);
2455
2456         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2457                 struct btrfs_extent_ref *ref;
2458                 ref = btrfs_item_ptr(leaf, path->slots[0],
2459                                      struct btrfs_extent_ref);
2460                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2461                 /* if the back ref and the extent are next to each other
2462                  * they get deleted below in one shot
2463                  */
2464                 path->slots[0] = extent_slot;
2465                 num_to_del = 2;
2466         } else if (found_extent) {
2467                 /* otherwise delete the extent back ref */
2468                 ret = remove_extent_backref(trans, extent_root, path);
2469                 BUG_ON(ret);
2470                 /* if refs are 0, we need to setup the path for deletion */
2471                 if (refs == 0) {
2472                         btrfs_release_path(extent_root, path);
2473                         ret = btrfs_search_slot(trans, extent_root, &key, path,
2474                                                 -1, 1);
2475                         BUG_ON(ret);
2476                 }
2477         }
2478
2479         if (refs == 0) {
2480                 u64 super_used;
2481                 u64 root_used;
2482
2483                 if (pin) {
2484                         mutex_lock(&root->fs_info->pinned_mutex);
2485                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2486                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2487                         mutex_unlock(&root->fs_info->pinned_mutex);
2488                         if (ret > 0)
2489                                 mark_free = 1;
2490                         BUG_ON(ret < 0);
2491                 }
2492                 /* block accounting for super block */
2493                 spin_lock(&info->delalloc_lock);
2494                 super_used = btrfs_super_bytes_used(&info->super_copy);
2495                 btrfs_set_super_bytes_used(&info->super_copy,
2496                                            super_used - num_bytes);
2497
2498                 /* block accounting for root item */
2499                 root_used = btrfs_root_used(&root->root_item);
2500                 btrfs_set_root_used(&root->root_item,
2501                                            root_used - num_bytes);
2502                 spin_unlock(&info->delalloc_lock);
2503                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2504                                       num_to_del);
2505                 BUG_ON(ret);
2506                 btrfs_release_path(extent_root, path);
2507
2508                 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2509                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2510                         BUG_ON(ret);
2511                 }
2512
2513                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2514                                          mark_free);
2515                 BUG_ON(ret);
2516         }
2517         btrfs_free_path(path);
2518         finish_current_insert(trans, extent_root, 0);
2519         return ret;
2520 }
2521
2522 /*
2523  * find all the blocks marked as pending in the radix tree and remove
2524  * them from the extent map
2525  */
2526 static int del_pending_extents(struct btrfs_trans_handle *trans,
2527                                struct btrfs_root *extent_root, int all)
2528 {
2529         int ret;
2530         int err = 0;
2531         u64 start;
2532         u64 end;
2533         u64 priv;
2534         u64 search = 0;
2535         int nr = 0, skipped = 0;
2536         struct extent_io_tree *pending_del;
2537         struct extent_io_tree *extent_ins;
2538         struct pending_extent_op *extent_op;
2539         struct btrfs_fs_info *info = extent_root->fs_info;
2540         struct list_head delete_list;
2541
2542         INIT_LIST_HEAD(&delete_list);
2543         extent_ins = &extent_root->fs_info->extent_ins;
2544         pending_del = &extent_root->fs_info->pending_del;
2545
2546 again:
2547         mutex_lock(&info->extent_ins_mutex);
2548         while (1) {
2549                 ret = find_first_extent_bit(pending_del, search, &start, &end,
2550                                             EXTENT_WRITEBACK);
2551                 if (ret) {
2552                         if (all && skipped && !nr) {
2553                                 search = 0;
2554                                 continue;
2555                         }
2556                         mutex_unlock(&info->extent_ins_mutex);
2557                         break;
2558                 }
2559
2560                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2561                 if (!ret) {
2562                         search = end+1;
2563                         skipped = 1;
2564
2565                         if (need_resched()) {
2566                                 mutex_unlock(&info->extent_ins_mutex);
2567                                 cond_resched();
2568                                 mutex_lock(&info->extent_ins_mutex);
2569                         }
2570
2571                         continue;
2572                 }
2573                 BUG_ON(ret < 0);
2574
2575                 ret = get_state_private(pending_del, start, &priv);
2576                 BUG_ON(ret);
2577                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2578
2579                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2580                                   GFP_NOFS);
2581                 if (!test_range_bit(extent_ins, start, end,
2582                                     EXTENT_WRITEBACK, 0)) {
2583                         list_add_tail(&extent_op->list, &delete_list);
2584                         nr++;
2585                 } else {
2586                         kfree(extent_op);
2587
2588                         ret = get_state_private(&info->extent_ins, start,
2589                                                 &priv);
2590                         BUG_ON(ret);
2591                         extent_op = (struct pending_extent_op *)
2592                                                 (unsigned long)priv;
2593
2594                         clear_extent_bits(&info->extent_ins, start, end,
2595                                           EXTENT_WRITEBACK, GFP_NOFS);
2596
2597                         if (extent_op->type == PENDING_BACKREF_UPDATE) {
2598                                 list_add_tail(&extent_op->list, &delete_list);
2599                                 search = end + 1;
2600                                 nr++;
2601                                 continue;
2602                         }
2603
2604                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2605                         ret = pin_down_bytes(trans, extent_root, start,
2606                                              end + 1 - start, 0);
2607                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2608
2609                         ret = update_block_group(trans, extent_root, start,
2610                                                 end + 1 - start, 0, ret > 0);
2611
2612                         unlock_extent(extent_ins, start, end, GFP_NOFS);
2613                         BUG_ON(ret);
2614                         kfree(extent_op);
2615                 }
2616                 if (ret)
2617                         err = ret;
2618
2619                 search = end + 1;
2620
2621                 if (need_resched()) {
2622                         mutex_unlock(&info->extent_ins_mutex);
2623                         cond_resched();
2624                         mutex_lock(&info->extent_ins_mutex);
2625                 }
2626         }
2627
2628         if (nr) {
2629                 ret = free_extents(trans, extent_root, &delete_list);
2630                 BUG_ON(ret);
2631         }
2632
2633         if (all && skipped) {
2634                 INIT_LIST_HEAD(&delete_list);
2635                 search = 0;
2636                 nr = 0;
2637                 goto again;
2638         }
2639
2640         return err;
2641 }
2642
2643 /*
2644  * remove an extent from the root, returns 0 on success
2645  */
2646 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2647                                struct btrfs_root *root,
2648                                u64 bytenr, u64 num_bytes, u64 parent,
2649                                u64 root_objectid, u64 ref_generation,
2650                                u64 owner_objectid, int pin)
2651 {
2652         struct btrfs_root *extent_root = root->fs_info->extent_root;
2653         int pending_ret;
2654         int ret;
2655
2656         WARN_ON(num_bytes < root->sectorsize);
2657         if (root == extent_root) {
2658                 struct pending_extent_op *extent_op = NULL;
2659
2660                 mutex_lock(&root->fs_info->extent_ins_mutex);
2661                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2662                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2663                         u64 priv;
2664                         ret = get_state_private(&root->fs_info->extent_ins,
2665                                                 bytenr, &priv);
2666                         BUG_ON(ret);
2667                         extent_op = (struct pending_extent_op *)
2668                                                 (unsigned long)priv;
2669
2670                         extent_op->del = 1;
2671                         if (extent_op->type == PENDING_EXTENT_INSERT) {
2672                                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2673                                 return 0;
2674                         }
2675                 }
2676
2677                 if (extent_op) {
2678                         ref_generation = extent_op->orig_generation;
2679                         parent = extent_op->orig_parent;
2680                 }
2681
2682                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2683                 BUG_ON(!extent_op);
2684
2685                 extent_op->type = PENDING_EXTENT_DELETE;
2686                 extent_op->bytenr = bytenr;
2687                 extent_op->num_bytes = num_bytes;
2688                 extent_op->parent = parent;
2689                 extent_op->orig_parent = parent;
2690                 extent_op->generation = ref_generation;
2691                 extent_op->orig_generation = ref_generation;
2692                 extent_op->level = (int)owner_objectid;
2693                 INIT_LIST_HEAD(&extent_op->list);
2694                 extent_op->del = 0;
2695
2696                 set_extent_bits(&root->fs_info->pending_del,
2697                                 bytenr, bytenr + num_bytes - 1,
2698                                 EXTENT_WRITEBACK, GFP_NOFS);
2699                 set_state_private(&root->fs_info->pending_del,
2700                                   bytenr, (unsigned long)extent_op);
2701                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2702                 return 0;
2703         }
2704         /* if metadata always pin */
2705         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2706                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2707                         struct btrfs_block_group_cache *cache;
2708
2709                         /* btrfs_free_reserved_extent */
2710                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2711                         BUG_ON(!cache);
2712                         btrfs_add_free_space(cache, bytenr, num_bytes);
2713                         put_block_group(cache);
2714                         update_reserved_extents(root, bytenr, num_bytes, 0);
2715                         return 0;
2716                 }
2717                 pin = 1;
2718         }
2719
2720         /* if data pin when any transaction has committed this */
2721         if (ref_generation != trans->transid)
2722                 pin = 1;
2723
2724         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2725                             root_objectid, ref_generation,
2726                             owner_objectid, pin, pin == 0);
2727
2728         finish_current_insert(trans, root->fs_info->extent_root, 0);
2729         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2730         return ret ? ret : pending_ret;
2731 }
2732
2733 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2734                       struct btrfs_root *root,
2735                       u64 bytenr, u64 num_bytes, u64 parent,
2736                       u64 root_objectid, u64 ref_generation,
2737                       u64 owner_objectid, int pin)
2738 {
2739         int ret;
2740
2741         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2742                                   root_objectid, ref_generation,
2743                                   owner_objectid, pin);
2744         return ret;
2745 }
2746
2747 static u64 stripe_align(struct btrfs_root *root, u64 val)
2748 {
2749         u64 mask = ((u64)root->stripesize - 1);
2750         u64 ret = (val + mask) & ~mask;
2751         return ret;
2752 }
2753
2754 /*
2755  * walks the btree of allocated extents and find a hole of a given size.
2756  * The key ins is changed to record the hole:
2757  * ins->objectid == block start
2758  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2759  * ins->offset == number of blocks
2760  * Any available blocks before search_start are skipped.
2761  */
2762 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2763                                      struct btrfs_root *orig_root,
2764                                      u64 num_bytes, u64 empty_size,
2765                                      u64 search_start, u64 search_end,
2766                                      u64 hint_byte, struct btrfs_key *ins,
2767                                      u64 exclude_start, u64 exclude_nr,
2768                                      int data)
2769 {
2770         int ret = 0;
2771         struct btrfs_root *root = orig_root->fs_info->extent_root;
2772         u64 total_needed = num_bytes;
2773         u64 *last_ptr = NULL;
2774         u64 last_wanted = 0;
2775         struct btrfs_block_group_cache *block_group = NULL;
2776         int chunk_alloc_done = 0;
2777         int empty_cluster = 2 * 1024 * 1024;
2778         int allowed_chunk_alloc = 0;
2779         struct list_head *head = NULL, *cur = NULL;
2780         int loop = 0;
2781         int extra_loop = 0;
2782         struct btrfs_space_info *space_info;
2783
2784         WARN_ON(num_bytes < root->sectorsize);
2785         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2786         ins->objectid = 0;
2787         ins->offset = 0;
2788
2789         if (orig_root->ref_cows || empty_size)
2790                 allowed_chunk_alloc = 1;
2791
2792         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2793                 last_ptr = &root->fs_info->last_alloc;
2794                 empty_cluster = 64 * 1024;
2795         }
2796
2797         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2798                 last_ptr = &root->fs_info->last_data_alloc;
2799
2800         if (last_ptr) {
2801                 if (*last_ptr) {
2802                         hint_byte = *last_ptr;
2803                         last_wanted = *last_ptr;
2804                 } else
2805                         empty_size += empty_cluster;
2806         } else {
2807                 empty_cluster = 0;
2808         }
2809         search_start = max(search_start, first_logical_byte(root, 0));
2810         search_start = max(search_start, hint_byte);
2811
2812         if (last_wanted && search_start != last_wanted) {
2813                 last_wanted = 0;
2814                 empty_size += empty_cluster;
2815         }
2816
2817         total_needed += empty_size;
2818         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2819         if (!block_group)
2820                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2821                                                              search_start);
2822         space_info = __find_space_info(root->fs_info, data);
2823
2824         down_read(&space_info->groups_sem);
2825         while (1) {
2826                 struct btrfs_free_space *free_space;
2827                 /*
2828                  * the only way this happens if our hint points to a block
2829                  * group thats not of the proper type, while looping this
2830                  * should never happen
2831                  */
2832                 if (empty_size)
2833                         extra_loop = 1;
2834
2835                 if (!block_group)
2836                         goto new_group_no_lock;
2837
2838                 if (unlikely(!block_group->cached)) {
2839                         mutex_lock(&block_group->cache_mutex);
2840                         ret = cache_block_group(root, block_group);
2841                         mutex_unlock(&block_group->cache_mutex);
2842                         if (ret)
2843                                 break;
2844                 }
2845
2846                 mutex_lock(&block_group->alloc_mutex);
2847                 if (unlikely(!block_group_bits(block_group, data)))
2848                         goto new_group;
2849
2850                 if (unlikely(block_group->ro))
2851                         goto new_group;
2852
2853                 free_space = btrfs_find_free_space(block_group, search_start,
2854                                                    total_needed);
2855                 if (free_space) {
2856                         u64 start = block_group->key.objectid;
2857                         u64 end = block_group->key.objectid +
2858                                 block_group->key.offset;
2859
2860                         search_start = stripe_align(root, free_space->offset);
2861
2862                         /* move on to the next group */
2863                         if (search_start + num_bytes >= search_end)
2864                                 goto new_group;
2865
2866                         /* move on to the next group */
2867                         if (search_start + num_bytes > end)
2868                                 goto new_group;
2869
2870                         if (last_wanted && search_start != last_wanted) {
2871                                 total_needed += empty_cluster;
2872                                 empty_size += empty_cluster;
2873                                 last_wanted = 0;
2874                                 /*
2875                                  * if search_start is still in this block group
2876                                  * then we just re-search this block group
2877                                  */
2878                                 if (search_start >= start &&
2879                                     search_start < end) {
2880                                         mutex_unlock(&block_group->alloc_mutex);
2881                                         continue;
2882                                 }
2883
2884                                 /* else we go to the next block group */
2885                                 goto new_group;
2886                         }
2887
2888                         if (exclude_nr > 0 &&
2889                             (search_start + num_bytes > exclude_start &&
2890                              search_start < exclude_start + exclude_nr)) {
2891                                 search_start = exclude_start + exclude_nr;
2892                                 /*
2893                                  * if search_start is still in this block group
2894                                  * then we just re-search this block group
2895                                  */
2896                                 if (search_start >= start &&
2897                                     search_start < end) {
2898                                         mutex_unlock(&block_group->alloc_mutex);
2899                                         last_wanted = 0;
2900                                         continue;
2901                                 }
2902
2903                                 /* else we go to the next block group */
2904                                 goto new_group;
2905                         }
2906
2907                         ins->objectid = search_start;
2908                         ins->offset = num_bytes;
2909
2910                         btrfs_remove_free_space_lock(block_group, search_start,
2911                                                      num_bytes);
2912                         /* we are all good, lets return */
2913                         mutex_unlock(&block_group->alloc_mutex);
2914                         break;
2915                 }
2916 new_group:
2917                 mutex_unlock(&block_group->alloc_mutex);
2918                 put_block_group(block_group);
2919                 block_group = NULL;
2920 new_group_no_lock:
2921                 /* don't try to compare new allocations against the
2922                  * last allocation any more
2923                  */
2924                 last_wanted = 0;
2925
2926                 /*
2927                  * Here's how this works.
2928                  * loop == 0: we were searching a block group via a hint
2929                  *              and didn't find anything, so we start at
2930                  *              the head of the block groups and keep searching
2931                  * loop == 1: we're searching through all of the block groups
2932                  *              if we hit the head again we have searched
2933                  *              all of the block groups for this space and we
2934                  *              need to try and allocate, if we cant error out.
2935                  * loop == 2: we allocated more space and are looping through
2936                  *              all of the block groups again.
2937                  */
2938                 if (loop == 0) {
2939                         head = &space_info->block_groups;
2940                         cur = head->next;
2941                         loop++;
2942                 } else if (loop == 1 && cur == head) {
2943                         int keep_going;
2944
2945                         /* at this point we give up on the empty_size
2946                          * allocations and just try to allocate the min
2947                          * space.
2948                          *
2949                          * The extra_loop field was set if an empty_size
2950                          * allocation was attempted above, and if this
2951                          * is try we need to try the loop again without
2952                          * the additional empty_size.
2953                          */
2954                         total_needed -= empty_size;
2955                         empty_size = 0;
2956                         keep_going = extra_loop;
2957                         loop++;
2958
2959                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2960                                 up_read(&space_info->groups_sem);
2961                                 ret = do_chunk_alloc(trans, root, num_bytes +
2962                                                      2 * 1024 * 1024, data, 1);
2963                                 down_read(&space_info->groups_sem);
2964                                 if (ret < 0)
2965                                         goto loop_check;
2966                                 head = &space_info->block_groups;
2967                                 /*
2968                                  * we've allocated a new chunk, keep
2969                                  * trying
2970                                  */
2971                                 keep_going = 1;
2972                                 chunk_alloc_done = 1;
2973                         } else if (!allowed_chunk_alloc) {
2974                                 space_info->force_alloc = 1;
2975                         }
2976 loop_check:
2977                         if (keep_going) {
2978                                 cur = head->next;
2979                                 extra_loop = 0;
2980                         } else {
2981                                 break;
2982                         }
2983                 } else if (cur == head) {
2984                         break;
2985                 }
2986
2987                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2988                                          list);
2989                 atomic_inc(&block_group->count);
2990
2991                 search_start = block_group->key.objectid;
2992                 cur = cur->next;
2993         }
2994
2995         /* we found what we needed */
2996         if (ins->objectid) {
2997                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2998                         trans->block_group = block_group->key.objectid;
2999
3000                 if (last_ptr)
3001                         *last_ptr = ins->objectid + ins->offset;
3002                 ret = 0;
3003         } else if (!ret) {
3004                 printk(KERN_ERR "btrfs searching for %llu bytes, "
3005                        "num_bytes %llu, loop %d, allowed_alloc %d\n",
3006                        (unsigned long long)total_needed,
3007                        (unsigned long long)num_bytes,
3008                        loop, allowed_chunk_alloc);
3009                 ret = -ENOSPC;
3010         }
3011         if (block_group)
3012                 put_block_group(block_group);
3013
3014         up_read(&space_info->groups_sem);
3015         return ret;
3016 }
3017
3018 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3019 {
3020         struct btrfs_block_group_cache *cache;
3021         struct list_head *l;
3022
3023         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3024                (unsigned long long)(info->total_bytes - info->bytes_used -
3025                                     info->bytes_pinned - info->bytes_reserved),
3026                (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 %llu has %llu bytes, %llu used "
3033                        "%llu pinned %llu reserved\n",
3034                        (unsigned long long)cache->key.objectid,
3035                        (unsigned long long)cache->key.offset,
3036                        (unsigned long long)btrfs_block_group_used(&cache->item),
3037                        (unsigned long long)cache->pinned,
3038                        (unsigned long long)cache->reserved);
3039                 btrfs_dump_free_space(cache, bytes);
3040                 spin_unlock(&cache->lock);
3041         }
3042         up_read(&info->groups_sem);
3043 }
3044
3045 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3046                                   struct btrfs_root *root,
3047                                   u64 num_bytes, u64 min_alloc_size,
3048                                   u64 empty_size, u64 hint_byte,
3049                                   u64 search_end, struct btrfs_key *ins,
3050                                   u64 data)
3051 {
3052         int ret;
3053         u64 search_start = 0;
3054         u64 alloc_profile;
3055         struct btrfs_fs_info *info = root->fs_info;
3056
3057         if (data) {
3058                 alloc_profile = info->avail_data_alloc_bits &
3059                         info->data_alloc_profile;
3060                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3061         } else if (root == root->fs_info->chunk_root) {
3062                 alloc_profile = info->avail_system_alloc_bits &
3063                         info->system_alloc_profile;
3064                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3065         } else {
3066                 alloc_profile = info->avail_metadata_alloc_bits &
3067                         info->metadata_alloc_profile;
3068                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3069         }
3070 again:
3071         data = btrfs_reduce_alloc_profile(root, data);
3072         /*
3073          * the only place that sets empty_size is btrfs_realloc_node, which
3074          * is not called recursively on allocations
3075          */
3076         if (empty_size || root->ref_cows) {
3077                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3078                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3079                                      2 * 1024 * 1024,
3080                                      BTRFS_BLOCK_GROUP_METADATA |
3081                                      (info->metadata_alloc_profile &
3082                                       info->avail_metadata_alloc_bits), 0);
3083                 }
3084                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3085                                      num_bytes + 2 * 1024 * 1024, data, 0);
3086         }
3087
3088         WARN_ON(num_bytes < root->sectorsize);
3089         ret = find_free_extent(trans, root, num_bytes, empty_size,
3090                                search_start, search_end, hint_byte, ins,
3091                                trans->alloc_exclude_start,
3092                                trans->alloc_exclude_nr, data);
3093
3094         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3095                 num_bytes = num_bytes >> 1;
3096                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3097                 num_bytes = max(num_bytes, min_alloc_size);
3098                 do_chunk_alloc(trans, root->fs_info->extent_root,
3099                                num_bytes, data, 1);
3100                 goto again;
3101         }
3102         if (ret) {
3103                 struct btrfs_space_info *sinfo;
3104
3105                 sinfo = __find_space_info(root->fs_info, data);
3106                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3107                        "wanted %llu\n", (unsigned long long)data,
3108                        (unsigned long long)num_bytes);
3109                 dump_space_info(sinfo, num_bytes);
3110                 BUG();
3111         }
3112
3113         return ret;
3114 }
3115
3116 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3117 {
3118         struct btrfs_block_group_cache *cache;
3119         int ret = 0;
3120
3121         cache = btrfs_lookup_block_group(root->fs_info, start);
3122         if (!cache) {
3123                 printk(KERN_ERR "Unable to find block group for %llu\n",
3124                        (unsigned long long)start);
3125                 return -ENOSPC;
3126         }
3127
3128         ret = btrfs_discard_extent(root, start, len);
3129
3130         btrfs_add_free_space(cache, start, len);
3131         put_block_group(cache);
3132         update_reserved_extents(root, start, len, 0);
3133
3134         return ret;
3135 }
3136
3137 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3138                                   struct btrfs_root *root,
3139                                   u64 num_bytes, u64 min_alloc_size,
3140                                   u64 empty_size, u64 hint_byte,
3141                                   u64 search_end, struct btrfs_key *ins,
3142                                   u64 data)
3143 {
3144         int ret;
3145         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3146                                      empty_size, hint_byte, search_end, ins,
3147                                      data);
3148         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3149         return ret;
3150 }
3151
3152 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3153                                          struct btrfs_root *root, u64 parent,
3154                                          u64 root_objectid, u64 ref_generation,
3155                                          u64 owner, struct btrfs_key *ins)
3156 {
3157         int ret;
3158         int pending_ret;
3159         u64 super_used;
3160         u64 root_used;
3161         u64 num_bytes = ins->offset;
3162         u32 sizes[2];
3163         struct btrfs_fs_info *info = root->fs_info;
3164         struct btrfs_root *extent_root = info->extent_root;
3165         struct btrfs_extent_item *extent_item;
3166         struct btrfs_extent_ref *ref;
3167         struct btrfs_path *path;
3168         struct btrfs_key keys[2];
3169
3170         if (parent == 0)
3171                 parent = ins->objectid;
3172
3173         /* block accounting for super block */
3174         spin_lock(&info->delalloc_lock);
3175         super_used = btrfs_super_bytes_used(&info->super_copy);
3176         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3177
3178         /* block accounting for root item */
3179         root_used = btrfs_root_used(&root->root_item);
3180         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3181         spin_unlock(&info->delalloc_lock);
3182
3183         if (root == extent_root) {
3184                 struct pending_extent_op *extent_op;
3185
3186                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3187                 BUG_ON(!extent_op);
3188
3189                 extent_op->type = PENDING_EXTENT_INSERT;
3190                 extent_op->bytenr = ins->objectid;
3191                 extent_op->num_bytes = ins->offset;
3192                 extent_op->parent = parent;
3193                 extent_op->orig_parent = 0;
3194                 extent_op->generation = ref_generation;
3195                 extent_op->orig_generation = 0;
3196                 extent_op->level = (int)owner;
3197                 INIT_LIST_HEAD(&extent_op->list);
3198                 extent_op->del = 0;
3199
3200                 mutex_lock(&root->fs_info->extent_ins_mutex);
3201                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3202                                 ins->objectid + ins->offset - 1,
3203                                 EXTENT_WRITEBACK, GFP_NOFS);
3204                 set_state_private(&root->fs_info->extent_ins,
3205                                   ins->objectid, (unsigned long)extent_op);
3206                 mutex_unlock(&root->fs_info->extent_ins_mutex);
3207                 goto update_block;
3208         }
3209
3210         memcpy(&keys[0], ins, sizeof(*ins));
3211         keys[1].objectid = ins->objectid;
3212         keys[1].type = BTRFS_EXTENT_REF_KEY;
3213         keys[1].offset = parent;
3214         sizes[0] = sizeof(*extent_item);
3215         sizes[1] = sizeof(*ref);
3216
3217         path = btrfs_alloc_path();
3218         BUG_ON(!path);
3219
3220         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3221                                        sizes, 2);
3222         BUG_ON(ret);
3223
3224         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3225                                      struct btrfs_extent_item);
3226         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3227         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3228                              struct btrfs_extent_ref);
3229
3230         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3231         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3232         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3233         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3234
3235         btrfs_mark_buffer_dirty(path->nodes[0]);
3236
3237         trans->alloc_exclude_start = 0;
3238         trans->alloc_exclude_nr = 0;
3239         btrfs_free_path(path);
3240         finish_current_insert(trans, extent_root, 0);
3241         pending_ret = del_pending_extents(trans, extent_root, 0);
3242
3243         if (ret)
3244                 goto out;
3245         if (pending_ret) {
3246                 ret = pending_ret;
3247                 goto out;
3248         }
3249
3250 update_block:
3251         ret = update_block_group(trans, root, ins->objectid,
3252                                  ins->offset, 1, 0);
3253         if (ret) {
3254                 printk(KERN_ERR "btrfs update block group failed for %llu "
3255                        "%llu\n", (unsigned long long)ins->objectid,
3256                        (unsigned long long)ins->offset);
3257                 BUG();
3258         }
3259 out:
3260         return ret;
3261 }
3262
3263 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3264                                 struct btrfs_root *root, u64 parent,
3265                                 u64 root_objectid, u64 ref_generation,
3266                                 u64 owner, struct btrfs_key *ins)
3267 {
3268         int ret;
3269
3270         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3271                 return 0;
3272         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3273                                             ref_generation, owner, ins);
3274         update_reserved_extents(root, ins->objectid, ins->offset, 0);
3275         return ret;
3276 }
3277
3278 /*
3279  * this is used by the tree logging recovery code.  It records that
3280  * an extent has been allocated and makes sure to clear the free
3281  * space cache bits as well
3282  */
3283 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3284                                 struct btrfs_root *root, u64 parent,
3285                                 u64 root_objectid, u64 ref_generation,
3286                                 u64 owner, struct btrfs_key *ins)
3287 {
3288         int ret;
3289         struct btrfs_block_group_cache *block_group;
3290
3291         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3292         mutex_lock(&block_group->cache_mutex);
3293         cache_block_group(root, block_group);
3294         mutex_unlock(&block_group->cache_mutex);
3295
3296         ret = btrfs_remove_free_space(block_group, ins->objectid,
3297                                       ins->offset);
3298         BUG_ON(ret);
3299         put_block_group(block_group);
3300         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3301                                             ref_generation, owner, ins);
3302         return ret;
3303 }
3304
3305 /*
3306  * finds a free extent and does all the dirty work required for allocation
3307  * returns the key for the extent through ins, and a tree buffer for
3308  * the first block of the extent through buf.
3309  *
3310  * returns 0 if everything worked, non-zero otherwise.
3311  */
3312 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3313                        struct btrfs_root *root,
3314                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3315                        u64 root_objectid, u64 ref_generation,
3316                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3317                        u64 search_end, struct btrfs_key *ins, u64 data)
3318 {
3319         int ret;
3320
3321         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3322                                      min_alloc_size, empty_size, hint_byte,
3323                                      search_end, ins, data);
3324         BUG_ON(ret);
3325         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3326                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3327                                         root_objectid, ref_generation,
3328                                         owner_objectid, ins);
3329                 BUG_ON(ret);
3330
3331         } else {
3332                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3333         }
3334         return ret;
3335 }
3336
3337 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3338                                             struct btrfs_root *root,
3339                                             u64 bytenr, u32 blocksize)
3340 {
3341         struct extent_buffer *buf;
3342
3343         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3344         if (!buf)
3345                 return ERR_PTR(-ENOMEM);
3346         btrfs_set_header_generation(buf, trans->transid);
3347         btrfs_tree_lock(buf);
3348         clean_tree_block(trans, root, buf);
3349         btrfs_set_buffer_uptodate(buf);
3350         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3351                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3352                          buf->start + buf->len - 1, GFP_NOFS);
3353         } else {
3354                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3355                          buf->start + buf->len - 1, GFP_NOFS);
3356         }
3357         trans->blocks_used++;
3358         return buf;
3359 }
3360
3361 /*
3362  * helper function to allocate a block for a given tree
3363  * returns the tree buffer or NULL.
3364  */
3365 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3366                                              struct btrfs_root *root,
3367                                              u32 blocksize, u64 parent,
3368                                              u64 root_objectid,
3369                                              u64 ref_generation,
3370                                              int level,
3371                                              u64 hint,
3372                                              u64 empty_size)
3373 {
3374         struct btrfs_key ins;
3375         int ret;
3376         struct extent_buffer *buf;
3377
3378         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3379                                  root_objectid, ref_generation, level,
3380                                  empty_size, hint, (u64)-1, &ins, 0);
3381         if (ret) {
3382                 BUG_ON(ret > 0);
3383                 return ERR_PTR(ret);
3384         }
3385
3386         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3387         return buf;
3388 }
3389
3390 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3391                         struct btrfs_root *root, struct extent_buffer *leaf)
3392 {
3393         u64 leaf_owner;
3394         u64 leaf_generation;
3395         struct btrfs_key key;
3396         struct btrfs_file_extent_item *fi;
3397         int i;
3398         int nritems;
3399         int ret;
3400
3401         BUG_ON(!btrfs_is_leaf(leaf));
3402         nritems = btrfs_header_nritems(leaf);
3403         leaf_owner = btrfs_header_owner(leaf);
3404         leaf_generation = btrfs_header_generation(leaf);
3405
3406         for (i = 0; i < nritems; i++) {
3407                 u64 disk_bytenr;
3408                 cond_resched();
3409
3410                 btrfs_item_key_to_cpu(leaf, &key, i);
3411                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3412                         continue;
3413                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3414                 if (btrfs_file_extent_type(leaf, fi) ==
3415                     BTRFS_FILE_EXTENT_INLINE)
3416                         continue;
3417                 /*
3418                  * FIXME make sure to insert a trans record that
3419                  * repeats the snapshot del on crash
3420                  */
3421                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3422                 if (disk_bytenr == 0)
3423                         continue;
3424
3425                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3426                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3427                                 leaf->start, leaf_owner, leaf_generation,
3428                                 key.objectid, 0);
3429                 BUG_ON(ret);
3430
3431                 atomic_inc(&root->fs_info->throttle_gen);
3432                 wake_up(&root->fs_info->transaction_throttle);
3433                 cond_resched();
3434         }
3435         return 0;
3436 }
3437
3438 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3439                                         struct btrfs_root *root,
3440                                         struct btrfs_leaf_ref *ref)
3441 {
3442         int i;
3443         int ret;
3444         struct btrfs_extent_info *info = ref->extents;
3445
3446         for (i = 0; i < ref->nritems; i++) {
3447                 ret = __btrfs_free_extent(trans, root, info->bytenr,
3448                                           info->num_bytes, ref->bytenr,
3449                                           ref->owner, ref->generation,
3450                                           info->objectid, 0);
3451
3452                 atomic_inc(&root->fs_info->throttle_gen);
3453                 wake_up(&root->fs_info->transaction_throttle);
3454                 cond_resched();
3455
3456                 BUG_ON(ret);
3457                 info++;
3458         }
3459
3460         return 0;
3461 }
3462
3463 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3464                                      u64 len, u32 *refs)
3465 {
3466         int ret;
3467
3468         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3469         BUG_ON(ret);
3470
3471 #if 0 /* some debugging code in case we see problems here */
3472         /* if the refs count is one, it won't get increased again.  But
3473          * if the ref count is > 1, someone may be decreasing it at
3474          * the same time we are.
3475          */
3476         if (*refs != 1) {
3477                 struct extent_buffer *eb = NULL;
3478                 eb = btrfs_find_create_tree_block(root, start, len);
3479                 if (eb)
3480                         btrfs_tree_lock(eb);
3481
3482                 mutex_lock(&root->fs_info->alloc_mutex);
3483                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3484                 BUG_ON(ret);
3485                 mutex_unlock(&root->fs_info->alloc_mutex);
3486
3487                 if (eb) {
3488                         btrfs_tree_unlock(eb);
3489                         free_extent_buffer(eb);
3490                 }
3491                 if (*refs == 1) {
3492                         printk(KERN_ERR "btrfs block %llu went down to one "
3493                                "during drop_snap\n", (unsigned long long)start);
3494                 }
3495
3496         }
3497 #endif
3498
3499         cond_resched();
3500         return ret;
3501 }
3502
3503 /*
3504  * helper function for drop_snapshot, this walks down the tree dropping ref
3505  * counts as it goes.
3506  */
3507 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3508                                    struct btrfs_root *root,
3509                                    struct btrfs_path *path, int *level)
3510 {
3511         u64 root_owner;
3512         u64 root_gen;
3513         u64 bytenr;
3514         u64 ptr_gen;
3515         struct extent_buffer *next;
3516         struct extent_buffer *cur;
3517         struct extent_buffer *parent;
3518         struct btrfs_leaf_ref *ref;
3519         u32 blocksize;
3520         int ret;
3521         u32 refs;
3522
3523         WARN_ON(*level < 0);
3524         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3525         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3526                                 path->nodes[*level]->len, &refs);
3527         BUG_ON(ret);
3528         if (refs > 1)
3529                 goto out;
3530
3531         /*
3532          * walk down to the last node level and free all the leaves
3533          */
3534         while (*level >= 0) {
3535                 WARN_ON(*level < 0);
3536                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3537                 cur = path->nodes[*level];
3538
3539                 if (btrfs_header_level(cur) != *level)
3540                         WARN_ON(1);
3541
3542                 if (path->slots[*level] >=
3543                     btrfs_header_nritems(cur))
3544                         break;
3545                 if (*level == 0) {
3546                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3547                         BUG_ON(ret);
3548                         break;
3549                 }
3550                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3551                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3552                 blocksize = btrfs_level_size(root, *level - 1);
3553
3554                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3555                 BUG_ON(ret);
3556                 if (refs != 1) {
3557                         parent = path->nodes[*level];
3558                         root_owner = btrfs_header_owner(parent);
3559                         root_gen = btrfs_header_generation(parent);
3560                         path->slots[*level]++;
3561
3562                         ret = __btrfs_free_extent(trans, root, bytenr,
3563                                                 blocksize, parent->start,
3564                                                 root_owner, root_gen,
3565                                                 *level - 1, 1);
3566                         BUG_ON(ret);
3567
3568                         atomic_inc(&root->fs_info->throttle_gen);
3569                         wake_up(&root->fs_info->transaction_throttle);
3570                         cond_resched();
3571
3572                         continue;
3573                 }
3574                 /*
3575                  * at this point, we have a single ref, and since the
3576                  * only place referencing this extent is a dead root
3577                  * the reference count should never go higher.
3578                  * So, we don't need to check it again
3579                  */
3580                 if (*level == 1) {
3581                         ref = btrfs_lookup_leaf_ref(root, bytenr);
3582                         if (ref && ref->generation != ptr_gen) {
3583                                 btrfs_free_leaf_ref(root, ref);
3584                                 ref = NULL;
3585                         }
3586                         if (ref) {
3587                                 ret = cache_drop_leaf_ref(trans, root, ref);
3588                                 BUG_ON(ret);
3589                                 btrfs_remove_leaf_ref(root, ref);
3590                                 btrfs_free_leaf_ref(root, ref);
3591                                 *level = 0;
3592                                 break;
3593                         }
3594                 }
3595                 next = btrfs_find_tree_block(root, bytenr, blocksize);
3596                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3597                         free_extent_buffer(next);
3598
3599                         next = read_tree_block(root, bytenr, blocksize,
3600                                                ptr_gen);
3601                         cond_resched();
3602 #if 0
3603                         /*
3604                          * this is a debugging check and can go away
3605                          * the ref should never go all the way down to 1
3606                          * at this point
3607                          */
3608                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3609                                                 &refs);
3610                         BUG_ON(ret);
3611                         WARN_ON(refs != 1);
3612 #endif
3613                 }
3614                 WARN_ON(*level <= 0);
3615                 if (path->nodes[*level-1])
3616                         free_extent_buffer(path->nodes[*level-1]);
3617                 path->nodes[*level-1] = next;
3618                 *level = btrfs_header_level(next);
3619                 path->slots[*level] = 0;
3620                 cond_resched();
3621         }
3622 out:
3623         WARN_ON(*level < 0);
3624         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3625
3626         if (path->nodes[*level] == root->node) {
3627                 parent = path->nodes[*level];
3628                 bytenr = path->nodes[*level]->start;
3629         } else {
3630                 parent = path->nodes[*level + 1];
3631                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3632         }
3633
3634         blocksize = btrfs_level_size(root, *level);
3635         root_owner = btrfs_header_owner(parent);
3636         root_gen = btrfs_header_generation(parent);
3637
3638         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3639                                   parent->start, root_owner, root_gen,
3640                                   *level, 1);
3641         free_extent_buffer(path->nodes[*level]);
3642         path->nodes[*level] = NULL;
3643         *level += 1;
3644         BUG_ON(ret);
3645
3646         cond_resched();
3647         return 0;
3648 }
3649
3650 /*
3651  * helper function for drop_subtree, this function is similar to
3652  * walk_down_tree. The main difference is that it checks reference
3653  * counts while tree blocks are locked.
3654  */
3655 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3656                                       struct btrfs_root *root,
3657                                       struct btrfs_path *path, int *level)
3658 {
3659         struct extent_buffer *next;
3660         struct extent_buffer *cur;
3661         struct extent_buffer *parent;
3662         u64 bytenr;
3663         u64 ptr_gen;
3664         u32 blocksize;
3665         u32 refs;
3666         int ret;
3667
3668         cur = path->nodes[*level];
3669         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3670                                       &refs);
3671         BUG_ON(ret);
3672         if (refs > 1)
3673                 goto out;
3674
3675         while (*level >= 0) {
3676                 cur = path->nodes[*level];
3677                 if (*level == 0) {
3678                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3679                         BUG_ON(ret);
3680                         clean_tree_block(trans, root, cur);
3681                         break;
3682                 }
3683                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3684                         clean_tree_block(trans, root, cur);
3685                         break;
3686                 }
3687
3688                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3689                 blocksize = btrfs_level_size(root, *level - 1);
3690                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3691
3692                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3693                 btrfs_tree_lock(next);
3694
3695                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3696                                               &refs);
3697                 BUG_ON(ret);
3698                 if (refs > 1) {
3699                         parent = path->nodes[*level];
3700                         ret = btrfs_free_extent(trans, root, bytenr,
3701                                         blocksize, parent->start,
3702                                         btrfs_header_owner(parent),
3703                                         btrfs_header_generation(parent),
3704                                         *level - 1, 1);
3705                         BUG_ON(ret);
3706                         path->slots[*level]++;
3707                         btrfs_tree_unlock(next);
3708                         free_extent_buffer(next);
3709                         continue;
3710                 }
3711
3712                 *level = btrfs_header_level(next);
3713                 path->nodes[*level] = next;
3714                 path->slots[*level] = 0;
3715                 path->locks[*level] = 1;
3716                 cond_resched();
3717         }
3718 out:
3719         parent = path->nodes[*level + 1];
3720         bytenr = path->nodes[*level]->start;
3721         blocksize = path->nodes[*level]->len;
3722
3723         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3724                         parent->start, btrfs_header_owner(parent),
3725                         btrfs_header_generation(parent), *level, 1);
3726         BUG_ON(ret);
3727
3728         if (path->locks[*level]) {
3729                 btrfs_tree_unlock(path->nodes[*level]);
3730                 path->locks[*level] = 0;
3731         }
3732         free_extent_buffer(path->nodes[*level]);
3733         path->nodes[*level] = NULL;
3734         *level += 1;
3735         cond_resched();
3736         return 0;
3737 }
3738
3739 /*
3740  * helper for dropping snapshots.  This walks back up the tree in the path
3741  * to find the first node higher up where we haven't yet gone through
3742  * all the slots
3743  */
3744 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3745                                  struct btrfs_root *root,
3746                                  struct btrfs_path *path,
3747                                  int *level, int max_level)
3748 {
3749         u64 root_owner;
3750         u64 root_gen;
3751         struct btrfs_root_item *root_item = &root->root_item;
3752         int i;
3753         int slot;
3754         int ret;
3755
3756         for (i = *level; i < max_level && path->nodes[i]; i++) {
3757                 slot = path->slots[i];
3758                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3759                         struct extent_buffer *node;
3760                         struct btrfs_disk_key disk_key;
3761                         node = path->nodes[i];
3762                         path->slots[i]++;
3763                         *level = i;
3764                         WARN_ON(*level == 0);
3765                         btrfs_node_key(node, &disk_key, path->slots[i]);
3766                         memcpy(&root_item->drop_progress,
3767                                &disk_key, sizeof(disk_key));
3768                         root_item->drop_level = i;
3769                         return 0;
3770                 } else {
3771                         struct extent_buffer *parent;
3772                         if (path->nodes[*level] == root->node)
3773                                 parent = path->nodes[*level];
3774                         else
3775                                 parent = path->nodes[*level + 1];
3776
3777                         root_owner = btrfs_header_owner(parent);
3778                         root_gen = btrfs_header_generation(parent);
3779
3780                         clean_tree_block(trans, root, path->nodes[*level]);
3781                         ret = btrfs_free_extent(trans, root,
3782                                                 path->nodes[*level]->start,
3783                                                 path->nodes[*level]->len,
3784                                                 parent->start, root_owner,
3785                                                 root_gen, *level, 1);
3786                         BUG_ON(ret);
3787                         if (path->locks[*level]) {
3788                                 btrfs_tree_unlock(path->nodes[*level]);
3789                                 path->locks[*level] = 0;
3790                         }
3791                         free_extent_buffer(path->nodes[*level]);
3792                         path->nodes[*level] = NULL;
3793                         *level = i + 1;
3794                 }
3795         }
3796         return 1;
3797 }
3798
3799 /*
3800  * drop the reference count on the tree rooted at 'snap'.  This traverses
3801  * the tree freeing any blocks that have a ref count of zero after being
3802  * decremented.
3803  */
3804 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3805                         *root)
3806 {
3807         int ret = 0;
3808         int wret;
3809         int level;
3810         struct btrfs_path *path;
3811         int i;
3812         int orig_level;
3813         struct btrfs_root_item *root_item = &root->root_item;
3814
3815         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3816         path = btrfs_alloc_path();
3817         BUG_ON(!path);
3818
3819         level = btrfs_header_level(root->node);
3820         orig_level = level;
3821         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3822                 path->nodes[level] = root->node;
3823                 extent_buffer_get(root->node);
3824                 path->slots[level] = 0;
3825         } else {
3826                 struct btrfs_key key;
3827                 struct btrfs_disk_key found_key;
3828                 struct extent_buffer *node;
3829
3830                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3831                 level = root_item->drop_level;
3832                 path->lowest_level = level;
3833                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3834                 if (wret < 0) {
3835                         ret = wret;
3836                         goto out;
3837                 }
3838                 node = path->nodes[level];
3839                 btrfs_node_key(node, &found_key, path->slots[level]);
3840                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3841                                sizeof(found_key)));
3842                 /*
3843                  * unlock our path, this is safe because only this
3844                  * function is allowed to delete this snapshot
3845                  */
3846                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3847                         if (path->nodes[i] && path->locks[i]) {
3848                                 path->locks[i] = 0;
3849                                 btrfs_tree_unlock(path->nodes[i]);
3850                         }
3851                 }
3852         }
3853         while (1) {
3854                 wret = walk_down_tree(trans, root, path, &level);
3855                 if (wret > 0)
3856                         break;
3857                 if (wret < 0)
3858                         ret = wret;
3859
3860                 wret = walk_up_tree(trans, root, path, &level,
3861                                     BTRFS_MAX_LEVEL);
3862                 if (wret > 0)
3863                         break;
3864                 if (wret < 0)
3865                         ret = wret;
3866                 if (trans->transaction->in_commit) {
3867                         ret = -EAGAIN;
3868                         break;
3869                 }
3870                 atomic_inc(&root->fs_info->throttle_gen);
3871                 wake_up(&root->fs_info->transaction_throttle);
3872         }
3873         for (i = 0; i <= orig_level; i++) {
3874                 if (path->nodes[i]) {
3875                         free_extent_buffer(path->nodes[i]);
3876                         path->nodes[i] = NULL;
3877                 }
3878         }
3879 out:
3880         btrfs_free_path(path);
3881         return ret;
3882 }
3883
3884 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3885                         struct btrfs_root *root,
3886                         struct extent_buffer *node,
3887                         struct extent_buffer *parent)
3888 {
3889         struct btrfs_path *path;
3890         int level;
3891         int parent_level;
3892         int ret = 0;
3893         int wret;
3894
3895         path = btrfs_alloc_path();
3896         BUG_ON(!path);
3897
3898         BUG_ON(!btrfs_tree_locked(parent));
3899         parent_level = btrfs_header_level(parent);
3900         extent_buffer_get(parent);
3901         path->nodes[parent_level] = parent;
3902         path->slots[parent_level] = btrfs_header_nritems(parent);
3903
3904         BUG_ON(!btrfs_tree_locked(node));
3905         level = btrfs_header_level(node);
3906         extent_buffer_get(node);
3907         path->nodes[level] = node;
3908         path->slots[level] = 0;
3909
3910         while (1) {
3911                 wret = walk_down_subtree(trans, root, path, &level);
3912                 if (wret < 0)
3913                         ret = wret;
3914                 if (wret != 0)
3915                         break;
3916
3917                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3918                 if (wret < 0)
3919                         ret = wret;
3920                 if (wret != 0)
3921                         break;
3922         }
3923
3924         btrfs_free_path(path);
3925         return ret;
3926 }
3927
3928 static unsigned long calc_ra(unsigned long start, unsigned long last,
3929                              unsigned long nr)
3930 {
3931         return min(last, start + nr - 1);
3932 }
3933
3934 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3935                                          u64 len)
3936 {
3937         u64 page_start;
3938         u64 page_end;
3939         unsigned long first_index;
3940         unsigned long last_index;
3941         unsigned long i;
3942         struct page *page;
3943         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3944         struct file_ra_state *ra;
3945         struct btrfs_ordered_extent *ordered;
3946         unsigned int total_read = 0;
3947         unsigned int total_dirty = 0;
3948         int ret = 0;
3949
3950         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3951
3952         mutex_lock(&inode->i_mutex);
3953         first_index = start >> PAGE_CACHE_SHIFT;
3954         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3955
3956         /* make sure the dirty trick played by the caller work */
3957         ret = invalidate_inode_pages2_range(inode->i_mapping,
3958                                             first_index, last_index);
3959         if (ret)
3960                 goto out_unlock;
3961
3962         file_ra_state_init(ra, inode->i_mapping);
3963
3964         for (i = first_index ; i <= last_index; i++) {
3965                 if (total_read % ra->ra_pages == 0) {
3966                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3967                                        calc_ra(i, last_index, ra->ra_pages));
3968                 }
3969                 total_read++;
3970 again:
3971                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3972                         BUG_ON(1);
3973                 page = grab_cache_page(inode->i_mapping, i);
3974                 if (!page) {
3975                         ret = -ENOMEM;
3976                         goto out_unlock;
3977                 }
3978                 if (!PageUptodate(page)) {
3979                         btrfs_readpage(NULL, page);
3980                         lock_page(page);
3981                         if (!PageUptodate(page)) {
3982                                 unlock_page(page);
3983                                 page_cache_release(page);
3984                                 ret = -EIO;
3985                                 goto out_unlock;
3986                         }
3987                 }
3988                 wait_on_page_writeback(page);
3989
3990                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3991                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3992                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3993
3994                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3995                 if (ordered) {
3996                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3997                         unlock_page(page);
3998                         page_cache_release(page);
3999                         btrfs_start_ordered_extent(inode, ordered, 1);
4000                         btrfs_put_ordered_extent(ordered);
4001                         goto again;
4002                 }
4003                 set_page_extent_mapped(page);
4004
4005                 if (i == first_index)
4006                         set_extent_bits(io_tree, page_start, page_end,
4007                                         EXTENT_BOUNDARY, GFP_NOFS);
4008                 btrfs_set_extent_delalloc(inode, page_start, page_end);
4009
4010                 set_page_dirty(page);
4011                 total_dirty++;
4012
4013                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4014                 unlock_page(page);
4015                 page_cache_release(page);
4016         }
4017
4018 out_unlock:
4019         kfree(ra);
4020         mutex_unlock(&inode->i_mutex);
4021         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4022         return ret;
4023 }
4024
4025 static noinline int relocate_data_extent(struct inode *reloc_inode,
4026                                          struct btrfs_key *extent_key,
4027                                          u64 offset)
4028 {
4029         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4030         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4031         struct extent_map *em;
4032         u64 start = extent_key->objectid - offset;
4033         u64 end = start + extent_key->offset - 1;
4034
4035         em = alloc_extent_map(GFP_NOFS);
4036         BUG_ON(!em || IS_ERR(em));
4037
4038         em->start = start;
4039         em->len = extent_key->offset;
4040         em->block_len = extent_key->offset;
4041         em->block_start = extent_key->objectid;
4042         em->bdev = root->fs_info->fs_devices->latest_bdev;
4043         set_bit(EXTENT_FLAG_PINNED, &em->flags);
4044
4045         /* setup extent map to cheat btrfs_readpage */
4046         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4047         while (1) {
4048                 int ret;
4049                 spin_lock(&em_tree->lock);
4050                 ret = add_extent_mapping(em_tree, em);
4051                 spin_unlock(&em_tree->lock);
4052                 if (ret != -EEXIST) {
4053                         free_extent_map(em);
4054                         break;
4055                 }
4056                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4057         }
4058         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4059
4060         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4061 }
4062
4063 struct btrfs_ref_path {
4064         u64 extent_start;
4065         u64 nodes[BTRFS_MAX_LEVEL];
4066         u64 root_objectid;
4067         u64 root_generation;
4068         u64 owner_objectid;
4069         u32 num_refs;
4070         int lowest_level;
4071         int current_level;
4072         int shared_level;
4073
4074         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4075         u64 new_nodes[BTRFS_MAX_LEVEL];
4076 };
4077
4078 struct disk_extent {
4079         u64 ram_bytes;
4080         u64 disk_bytenr;
4081         u64 disk_num_bytes;
4082         u64 offset;
4083         u64 num_bytes;
4084         u8 compression;
4085         u8 encryption;
4086         u16 other_encoding;
4087 };
4088
4089 static int is_cowonly_root(u64 root_objectid)
4090 {
4091         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4092             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4093             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4094             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4095             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4096             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4097                 return 1;
4098         return 0;
4099 }
4100
4101 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4102                                     struct btrfs_root *extent_root,
4103                                     struct btrfs_ref_path *ref_path,
4104                                     int first_time)
4105 {
4106         struct extent_buffer *leaf;
4107         struct btrfs_path *path;
4108         struct btrfs_extent_ref *ref;
4109         struct btrfs_key key;
4110         struct btrfs_key found_key;
4111         u64 bytenr;
4112         u32 nritems;
4113         int level;
4114         int ret = 1;
4115
4116         path = btrfs_alloc_path();
4117         if (!path)
4118                 return -ENOMEM;
4119
4120         if (first_time) {
4121                 ref_path->lowest_level = -1;
4122                 ref_path->current_level = -1;
4123                 ref_path->shared_level = -1;
4124                 goto walk_up;
4125         }
4126 walk_down:
4127         level = ref_path->current_level - 1;
4128         while (level >= -1) {
4129                 u64 parent;
4130                 if (level < ref_path->lowest_level)
4131                         break;
4132
4133                 if (level >= 0)
4134                         bytenr = ref_path->nodes[level];
4135                 else
4136                         bytenr = ref_path->extent_start;
4137                 BUG_ON(bytenr == 0);
4138
4139                 parent = ref_path->nodes[level + 1];
4140                 ref_path->nodes[level + 1] = 0;
4141                 ref_path->current_level = level;
4142                 BUG_ON(parent == 0);
4143
4144                 key.objectid = bytenr;
4145                 key.offset = parent + 1;
4146                 key.type = BTRFS_EXTENT_REF_KEY;
4147
4148                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4149                 if (ret < 0)
4150                         goto out;
4151                 BUG_ON(ret == 0);
4152
4153                 leaf = path->nodes[0];
4154                 nritems = btrfs_header_nritems(leaf);
4155                 if (path->slots[0] >= nritems) {
4156                         ret = btrfs_next_leaf(extent_root, path);
4157                         if (ret < 0)
4158                                 goto out;
4159                         if (ret > 0)
4160                                 goto next;
4161                         leaf = path->nodes[0];
4162                 }
4163
4164                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4165                 if (found_key.objectid == bytenr &&
4166                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4167                         if (level < ref_path->shared_level)
4168                                 ref_path->shared_level = level;
4169                         goto found;
4170                 }
4171 next:
4172                 level--;
4173                 btrfs_release_path(extent_root, path);
4174                 cond_resched();
4175         }
4176         /* reached lowest level */
4177         ret = 1;
4178         goto out;
4179 walk_up:
4180         level = ref_path->current_level;
4181         while (level < BTRFS_MAX_LEVEL - 1) {
4182                 u64 ref_objectid;
4183
4184                 if (level >= 0)
4185                         bytenr = ref_path->nodes[level];
4186                 else
4187                         bytenr = ref_path->extent_start;
4188
4189                 BUG_ON(bytenr == 0);
4190
4191                 key.objectid = bytenr;
4192                 key.offset = 0;
4193                 key.type = BTRFS_EXTENT_REF_KEY;
4194
4195                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4196                 if (ret < 0)
4197                         goto out;
4198
4199                 leaf = path->nodes[0];
4200                 nritems = btrfs_header_nritems(leaf);
4201                 if (path->slots[0] >= nritems) {
4202                         ret = btrfs_next_leaf(extent_root, path);
4203                         if (ret < 0)
4204                                 goto out;
4205                         if (ret > 0) {
4206                                 /* the extent was freed by someone */
4207                                 if (ref_path->lowest_level == level)
4208                                         goto out;
4209                                 btrfs_release_path(extent_root, path);
4210                                 goto walk_down;
4211                         }
4212                         leaf = path->nodes[0];
4213                 }
4214
4215                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4216                 if (found_key.objectid != bytenr ||
4217                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4218                         /* the extent was freed by someone */
4219                         if (ref_path->lowest_level == level) {
4220                                 ret = 1;
4221                                 goto out;
4222                         }
4223                         btrfs_release_path(extent_root, path);
4224                         goto walk_down;
4225                 }
4226 found:
4227                 ref = btrfs_item_ptr(leaf, path->slots[0],
4228                                 struct btrfs_extent_ref);
4229                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4230                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4231                         if (first_time) {
4232                                 level = (int)ref_objectid;
4233                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4234                                 ref_path->lowest_level = level;
4235                                 ref_path->current_level = level;
4236                                 ref_path->nodes[level] = bytenr;
4237                         } else {
4238                                 WARN_ON(ref_objectid != level);
4239                         }
4240                 } else {
4241                         WARN_ON(level != -1);
4242                 }
4243                 first_time = 0;
4244
4245                 if (ref_path->lowest_level == level) {
4246                         ref_path->owner_objectid = ref_objectid;
4247                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4248                 }
4249
4250                 /*
4251                  * the block is tree root or the block isn't in reference
4252                  * counted tree.
4253                  */
4254                 if (found_key.objectid == found_key.offset ||
4255                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4256                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4257                         ref_path->root_generation =
4258                                 btrfs_ref_generation(leaf, ref);
4259                         if (level < 0) {
4260                                 /* special reference from the tree log */
4261                                 ref_path->nodes[0] = found_key.offset;
4262                                 ref_path->current_level = 0;
4263                         }
4264                         ret = 0;
4265                         goto out;
4266                 }
4267
4268                 level++;
4269                 BUG_ON(ref_path->nodes[level] != 0);
4270                 ref_path->nodes[level] = found_key.offset;
4271                 ref_path->current_level = level;
4272
4273                 /*
4274                  * the reference was created in the running transaction,
4275                  * no need to continue walking up.
4276                  */
4277                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4278                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4279                         ref_path->root_generation =
4280                                 btrfs_ref_generation(leaf, ref);
4281                         ret = 0;
4282                         goto out;
4283                 }
4284
4285                 btrfs_release_path(extent_root, path);
4286                 cond_resched();
4287         }
4288         /* reached max tree level, but no tree root found. */
4289         BUG();
4290 out:
4291         btrfs_free_path(path);
4292         return ret;
4293 }
4294
4295 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4296                                 struct btrfs_root *extent_root,
4297                                 struct btrfs_ref_path *ref_path,
4298                                 u64 extent_start)
4299 {
4300         memset(ref_path, 0, sizeof(*ref_path));
4301         ref_path->extent_start = extent_start;
4302
4303         return __next_ref_path(trans, extent_root, ref_path, 1);
4304 }
4305
4306 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4307                                struct btrfs_root *extent_root,
4308                                struct btrfs_ref_path *ref_path)
4309 {
4310         return __next_ref_path(trans, extent_root, ref_path, 0);
4311 }
4312
4313 static noinline int get_new_locations(struct inode *reloc_inode,
4314                                       struct btrfs_key *extent_key,
4315                                       u64 offset, int no_fragment,
4316                                       struct disk_extent **extents,
4317                                       int *nr_extents)
4318 {
4319         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4320         struct btrfs_path *path;
4321         struct btrfs_file_extent_item *fi;
4322         struct extent_buffer *leaf;
4323         struct disk_extent *exts = *extents;
4324         struct btrfs_key found_key;
4325         u64 cur_pos;
4326         u64 last_byte;
4327         u32 nritems;
4328         int nr = 0;
4329         int max = *nr_extents;
4330         int ret;
4331
4332         WARN_ON(!no_fragment && *extents);
4333         if (!exts) {
4334                 max = 1;
4335                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4336                 if (!exts)
4337                         return -ENOMEM;
4338         }
4339
4340         path = btrfs_alloc_path();
4341         BUG_ON(!path);
4342
4343         cur_pos = extent_key->objectid - offset;
4344         last_byte = extent_key->objectid + extent_key->offset;
4345         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4346                                        cur_pos, 0);
4347         if (ret < 0)
4348                 goto out;
4349         if (ret > 0) {
4350                 ret = -ENOENT;
4351                 goto out;
4352         }
4353
4354         while (1) {
4355                 leaf = path->nodes[0];
4356                 nritems = btrfs_header_nritems(leaf);
4357                 if (path->slots[0] >= nritems) {
4358                         ret = btrfs_next_leaf(root, path);
4359                         if (ret < 0)
4360                                 goto out;
4361                         if (ret > 0)
4362                                 break;
4363                         leaf = path->nodes[0];
4364                 }
4365
4366                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4367                 if (found_key.offset != cur_pos ||
4368                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4369                     found_key.objectid != reloc_inode->i_ino)
4370                         break;
4371
4372                 fi = btrfs_item_ptr(leaf, path->slots[0],
4373                                     struct btrfs_file_extent_item);
4374                 if (btrfs_file_extent_type(leaf, fi) !=
4375                     BTRFS_FILE_EXTENT_REG ||
4376                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4377                         break;
4378
4379                 if (nr == max) {
4380                         struct disk_extent *old = exts;
4381                         max *= 2;
4382                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4383                         memcpy(exts, old, sizeof(*exts) * nr);
4384                         if (old != *extents)
4385                                 kfree(old);
4386                 }
4387
4388                 exts[nr].disk_bytenr =
4389                         btrfs_file_extent_disk_bytenr(leaf, fi);
4390                 exts[nr].disk_num_bytes =
4391                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4392                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4393                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4394                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4395                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4396                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4397                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4398                                                                            fi);
4399                 BUG_ON(exts[nr].offset > 0);
4400                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4401                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4402
4403                 cur_pos += exts[nr].num_bytes;
4404                 nr++;
4405
4406                 if (cur_pos + offset >= last_byte)
4407                         break;
4408
4409                 if (no_fragment) {
4410                         ret = 1;
4411                         goto out;
4412                 }
4413                 path->slots[0]++;
4414         }
4415
4416         BUG_ON(cur_pos + offset > last_byte);
4417         if (cur_pos + offset < last_byte) {
4418                 ret = -ENOENT;
4419                 goto out;
4420         }
4421         ret = 0;
4422 out:
4423         btrfs_free_path(path);
4424         if (ret) {
4425                 if (exts != *extents)
4426                         kfree(exts);
4427         } else {
4428                 *extents = exts;
4429                 *nr_extents = nr;
4430         }
4431         return ret;
4432 }
4433
4434 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4435                                         struct btrfs_root *root,
4436                                         struct btrfs_path *path,
4437                                         struct btrfs_key *extent_key,
4438                                         struct btrfs_key *leaf_key,
4439                                         struct btrfs_ref_path *ref_path,
4440                                         struct disk_extent *new_extents,
4441                                         int nr_extents)
4442 {
4443         struct extent_buffer *leaf;
4444         struct btrfs_file_extent_item *fi;
4445         struct inode *inode = NULL;
4446         struct btrfs_key key;
4447         u64 lock_start = 0;
4448         u64 lock_end = 0;
4449         u64 num_bytes;
4450         u64 ext_offset;
4451         u64 first_pos;
4452         u32 nritems;
4453         int nr_scaned = 0;
4454         int extent_locked = 0;
4455         int extent_type;
4456         int ret;
4457
4458         memcpy(&key, leaf_key, sizeof(key));
4459         first_pos = INT_LIMIT(loff_t) - extent_key->offset;
4460         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4461                 if (key.objectid < ref_path->owner_objectid ||
4462                     (key.objectid == ref_path->owner_objectid &&
4463                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4464                         key.objectid = ref_path->owner_objectid;
4465                         key.type = BTRFS_EXTENT_DATA_KEY;
4466                         key.offset = 0;
4467                 }
4468         }
4469
4470         while (1) {
4471                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4472                 if (ret < 0)
4473                         goto out;
4474
4475                 leaf = path->nodes[0];
4476                 nritems = btrfs_header_nritems(leaf);
4477 next:
4478                 if (extent_locked && ret > 0) {
4479                         /*
4480                          * the file extent item was modified by someone
4481                          * before the extent got locked.
4482                          */
4483                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4484                                       lock_end, GFP_NOFS);
4485                         extent_locked = 0;
4486                 }
4487
4488                 if (path->slots[0] >= nritems) {
4489                         if (++nr_scaned > 2)
4490                                 break;
4491
4492                         BUG_ON(extent_locked);
4493                         ret = btrfs_next_leaf(root, path);
4494                         if (ret < 0)
4495                                 goto out;
4496                         if (ret > 0)
4497                                 break;
4498                         leaf = path->nodes[0];
4499                         nritems = btrfs_header_nritems(leaf);
4500                 }
4501
4502                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4503
4504                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4505                         if ((key.objectid > ref_path->owner_objectid) ||
4506                             (key.objectid == ref_path->owner_objectid &&
4507                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4508                             (key.offset >= first_pos + extent_key->offset))
4509                                 break;
4510                 }
4511
4512                 if (inode && key.objectid != inode->i_ino) {
4513                         BUG_ON(extent_locked);
4514                         btrfs_release_path(root, path);
4515                         mutex_unlock(&inode->i_mutex);
4516                         iput(inode);
4517                         inode = NULL;
4518                         continue;
4519                 }
4520
4521                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4522                         path->slots[0]++;
4523                         ret = 1;
4524                         goto next;
4525                 }
4526                 fi = btrfs_item_ptr(leaf, path->slots[0],
4527                                     struct btrfs_file_extent_item);
4528                 extent_type = btrfs_file_extent_type(leaf, fi);
4529                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4530                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4531                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4532                      extent_key->objectid)) {
4533                         path->slots[0]++;
4534                         ret = 1;
4535                         goto next;
4536                 }
4537
4538                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4539                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4540
4541                 if (first_pos > key.offset - ext_offset)
4542                         first_pos = key.offset - ext_offset;
4543
4544                 if (!extent_locked) {
4545                         lock_start = key.offset;
4546                         lock_end = lock_start + num_bytes - 1;
4547                 } else {
4548                         if (lock_start > key.offset ||
4549                             lock_end + 1 < key.offset + num_bytes) {
4550                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4551                                               lock_start, lock_end, GFP_NOFS);
4552                                 extent_locked = 0;
4553                         }
4554                 }
4555
4556                 if (!inode) {
4557                         btrfs_release_path(root, path);
4558
4559                         inode = btrfs_iget_locked(root->fs_info->sb,
4560                                                   key.objectid, root);
4561                         if (inode->i_state & I_NEW) {
4562                                 BTRFS_I(inode)->root = root;
4563                                 BTRFS_I(inode)->location.objectid =
4564                                         key.objectid;
4565                                 BTRFS_I(inode)->location.type =
4566                                         BTRFS_INODE_ITEM_KEY;
4567                                 BTRFS_I(inode)->location.offset = 0;
4568                                 btrfs_read_locked_inode(inode);
4569                                 unlock_new_inode(inode);
4570                         }
4571                         /*
4572                          * some code call btrfs_commit_transaction while
4573                          * holding the i_mutex, so we can't use mutex_lock
4574                          * here.
4575                          */
4576                         if (is_bad_inode(inode) ||
4577                             !mutex_trylock(&inode->i_mutex)) {
4578                                 iput(inode);
4579                                 inode = NULL;
4580                                 key.offset = (u64)-1;
4581                                 goto skip;
4582                         }
4583                 }
4584
4585                 if (!extent_locked) {
4586                         struct btrfs_ordered_extent *ordered;
4587
4588                         btrfs_release_path(root, path);
4589
4590                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4591                                     lock_end, GFP_NOFS);
4592                         ordered = btrfs_lookup_first_ordered_extent(inode,
4593                                                                     lock_end);
4594                         if (ordered &&
4595                             ordered->file_offset <= lock_end &&
4596                             ordered->file_offset + ordered->len > lock_start) {
4597                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4598                                               lock_start, lock_end, GFP_NOFS);
4599                                 btrfs_start_ordered_extent(inode, ordered, 1);
4600                                 btrfs_put_ordered_extent(ordered);
4601                                 key.offset += num_bytes;
4602                                 goto skip;
4603                         }
4604                         if (ordered)
4605                                 btrfs_put_ordered_extent(ordered);
4606
4607                         extent_locked = 1;
4608                         continue;
4609                 }
4610
4611                 if (nr_extents == 1) {
4612                         /* update extent pointer in place */
4613                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4614                                                 new_extents[0].disk_bytenr);
4615                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4616                                                 new_extents[0].disk_num_bytes);
4617                         btrfs_mark_buffer_dirty(leaf);
4618
4619                         btrfs_drop_extent_cache(inode, key.offset,
4620                                                 key.offset + num_bytes - 1, 0);
4621
4622                         ret = btrfs_inc_extent_ref(trans, root,
4623                                                 new_extents[0].disk_bytenr,
4624                                                 new_extents[0].disk_num_bytes,
4625                                                 leaf->start,
4626                                                 root->root_key.objectid,
4627                                                 trans->transid,
4628                                                 key.objectid);
4629                         BUG_ON(ret);
4630
4631                         ret = btrfs_free_extent(trans, root,
4632                                                 extent_key->objectid,
4633                                                 extent_key->offset,
4634                                                 leaf->start,
4635                                                 btrfs_header_owner(leaf),
4636                                                 btrfs_header_generation(leaf),
4637                                                 key.objectid, 0);
4638                         BUG_ON(ret);
4639
4640                         btrfs_release_path(root, path);
4641                         key.offset += num_bytes;
4642                 } else {
4643                         BUG_ON(1);
4644 #if 0
4645                         u64 alloc_hint;
4646                         u64 extent_len;
4647                         int i;
4648                         /*
4649                          * drop old extent pointer at first, then insert the
4650                          * new pointers one bye one
4651                          */
4652                         btrfs_release_path(root, path);
4653                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4654                                                  key.offset + num_bytes,
4655                                                  key.offset, &alloc_hint);
4656                         BUG_ON(ret);
4657
4658                         for (i = 0; i < nr_extents; i++) {
4659                                 if (ext_offset >= new_extents[i].num_bytes) {
4660                                         ext_offset -= new_extents[i].num_bytes;
4661                                         continue;
4662                                 }
4663                                 extent_len = min(new_extents[i].num_bytes -
4664                                                  ext_offset, num_bytes);
4665
4666                                 ret = btrfs_insert_empty_item(trans, root,
4667                                                               path, &key,
4668                                                               sizeof(*fi));
4669                                 BUG_ON(ret);
4670
4671                                 leaf = path->nodes[0];
4672                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4673                                                 struct btrfs_file_extent_item);
4674                                 btrfs_set_file_extent_generation(leaf, fi,
4675                                                         trans->transid);
4676                                 btrfs_set_file_extent_type(leaf, fi,
4677                                                         BTRFS_FILE_EXTENT_REG);
4678                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4679                                                 new_extents[i].disk_bytenr);
4680                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4681                                                 new_extents[i].disk_num_bytes);
4682                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4683                                                 new_extents[i].ram_bytes);
4684
4685                                 btrfs_set_file_extent_compression(leaf, fi,
4686                                                 new_extents[i].compression);
4687                                 btrfs_set_file_extent_encryption(leaf, fi,
4688                                                 new_extents[i].encryption);
4689                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4690                                                 new_extents[i].other_encoding);
4691
4692                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4693                                                         extent_len);
4694                                 ext_offset += new_extents[i].offset;
4695                                 btrfs_set_file_extent_offset(leaf, fi,
4696                                                         ext_offset);
4697                                 btrfs_mark_buffer_dirty(leaf);
4698
4699                                 btrfs_drop_extent_cache(inode, key.offset,
4700                                                 key.offset + extent_len - 1, 0);
4701
4702                                 ret = btrfs_inc_extent_ref(trans, root,
4703                                                 new_extents[i].disk_bytenr,
4704                                                 new_extents[i].disk_num_bytes,
4705                                                 leaf->start,
4706                                                 root->root_key.objectid,
4707                                                 trans->transid, key.objectid);
4708                                 BUG_ON(ret);
4709                                 btrfs_release_path(root, path);
4710
4711                                 inode_add_bytes(inode, extent_len);
4712
4713                                 ext_offset = 0;
4714                                 num_bytes -= extent_len;
4715                                 key.offset += extent_len;
4716
4717                                 if (num_bytes == 0)
4718                                         break;
4719                         }
4720                         BUG_ON(i >= nr_extents);
4721 #endif
4722                 }
4723
4724                 if (extent_locked) {
4725                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4726                                       lock_end, GFP_NOFS);
4727                         extent_locked = 0;
4728                 }
4729 skip:
4730                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4731                     key.offset >= first_pos + extent_key->offset)
4732                         break;
4733
4734                 cond_resched();
4735         }
4736         ret = 0;
4737 out:
4738         btrfs_release_path(root, path);
4739         if (inode) {
4740                 mutex_unlock(&inode->i_mutex);
4741                 if (extent_locked) {
4742                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4743                                       lock_end, GFP_NOFS);
4744                 }
4745                 iput(inode);
4746         }
4747         return ret;
4748 }
4749
4750 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4751                                struct btrfs_root *root,
4752                                struct extent_buffer *buf, u64 orig_start)
4753 {
4754         int level;
4755         int ret;
4756
4757         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4758         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4759
4760         level = btrfs_header_level(buf);
4761         if (level == 0) {
4762                 struct btrfs_leaf_ref *ref;
4763                 struct btrfs_leaf_ref *orig_ref;
4764
4765                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4766                 if (!orig_ref)
4767                         return -ENOENT;
4768
4769                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4770                 if (!ref) {
4771                         btrfs_free_leaf_ref(root, orig_ref);
4772                         return -ENOMEM;
4773                 }
4774
4775                 ref->nritems = orig_ref->nritems;
4776                 memcpy(ref->extents, orig_ref->extents,
4777                         sizeof(ref->extents[0]) * ref->nritems);
4778
4779                 btrfs_free_leaf_ref(root, orig_ref);
4780
4781                 ref->root_gen = trans->transid;
4782                 ref->bytenr = buf->start;
4783                 ref->owner = btrfs_header_owner(buf);
4784                 ref->generation = btrfs_header_generation(buf);
4785                 ret = btrfs_add_leaf_ref(root, ref, 0);
4786                 WARN_ON(ret);
4787                 btrfs_free_leaf_ref(root, ref);
4788         }
4789         return 0;
4790 }
4791
4792 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4793                                         struct extent_buffer *leaf,
4794                                         struct btrfs_block_group_cache *group,
4795                                         struct btrfs_root *target_root)
4796 {
4797         struct btrfs_key key;
4798         struct inode *inode = NULL;
4799         struct btrfs_file_extent_item *fi;
4800         u64 num_bytes;
4801         u64 skip_objectid = 0;
4802         u32 nritems;
4803         u32 i;
4804
4805         nritems = btrfs_header_nritems(leaf);
4806         for (i = 0; i < nritems; i++) {
4807                 btrfs_item_key_to_cpu(leaf, &key, i);
4808                 if (key.objectid == skip_objectid ||
4809                     key.type != BTRFS_EXTENT_DATA_KEY)
4810                         continue;
4811                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4812                 if (btrfs_file_extent_type(leaf, fi) ==
4813                     BTRFS_FILE_EXTENT_INLINE)
4814                         continue;
4815                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4816                         continue;
4817                 if (!inode || inode->i_ino != key.objectid) {
4818                         iput(inode);
4819                         inode = btrfs_ilookup(target_root->fs_info->sb,
4820                                               key.objectid, target_root, 1);
4821                 }
4822                 if (!inode) {
4823                         skip_objectid = key.objectid;
4824                         continue;
4825                 }
4826                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4827
4828                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4829                             key.offset + num_bytes - 1, GFP_NOFS);
4830                 btrfs_drop_extent_cache(inode, key.offset,
4831                                         key.offset + num_bytes - 1, 1);
4832                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4833                               key.offset + num_bytes - 1, GFP_NOFS);
4834                 cond_resched();
4835         }
4836         iput(inode);
4837         return 0;
4838 }
4839
4840 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4841                                         struct btrfs_root *root,
4842                                         struct extent_buffer *leaf,
4843                                         struct btrfs_block_group_cache *group,
4844                                         struct inode *reloc_inode)
4845 {
4846         struct btrfs_key key;
4847         struct btrfs_key extent_key;
4848         struct btrfs_file_extent_item *fi;
4849         struct btrfs_leaf_ref *ref;
4850         struct disk_extent *new_extent;
4851         u64 bytenr;
4852         u64 num_bytes;
4853         u32 nritems;
4854         u32 i;
4855         int ext_index;
4856         int nr_extent;
4857         int ret;
4858
4859         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4860         BUG_ON(!new_extent);
4861
4862         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4863         BUG_ON(!ref);
4864
4865         ext_index = -1;
4866         nritems = btrfs_header_nritems(leaf);
4867         for (i = 0; i < nritems; i++) {
4868                 btrfs_item_key_to_cpu(leaf, &key, i);
4869                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4870                         continue;
4871                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4872                 if (btrfs_file_extent_type(leaf, fi) ==
4873                     BTRFS_FILE_EXTENT_INLINE)
4874                         continue;
4875                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4876                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4877                 if (bytenr == 0)
4878                         continue;
4879
4880                 ext_index++;
4881                 if (bytenr >= group->key.objectid + group->key.offset ||
4882                     bytenr + num_bytes <= group->key.objectid)
4883                         continue;
4884
4885                 extent_key.objectid = bytenr;
4886                 extent_key.offset = num_bytes;
4887                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4888                 nr_extent = 1;
4889                 ret = get_new_locations(reloc_inode, &extent_key,
4890                                         group->key.objectid, 1,
4891                                         &new_extent, &nr_extent);
4892                 if (ret > 0)
4893                         continue;
4894                 BUG_ON(ret < 0);
4895
4896                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4897                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4898                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4899                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4900
4901                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4902                                                 new_extent->disk_bytenr);
4903                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4904                                                 new_extent->disk_num_bytes);
4905                 btrfs_mark_buffer_dirty(leaf);
4906
4907                 ret = btrfs_inc_extent_ref(trans, root,
4908                                         new_extent->disk_bytenr,
4909                                         new_extent->disk_num_bytes,
4910                                         leaf->start,
4911                                         root->root_key.objectid,
4912                                         trans->transid, key.objectid);
4913                 BUG_ON(ret);
4914                 ret = btrfs_free_extent(trans, root,
4915                                         bytenr, num_bytes, leaf->start,
4916                                         btrfs_header_owner(leaf),
4917                                         btrfs_header_generation(leaf),
4918                                         key.objectid, 0);
4919                 BUG_ON(ret);
4920                 cond_resched();
4921         }
4922         kfree(new_extent);
4923         BUG_ON(ext_index + 1 != ref->nritems);
4924         btrfs_free_leaf_ref(root, ref);
4925         return 0;
4926 }
4927
4928 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4929                           struct btrfs_root *root)
4930 {
4931         struct btrfs_root *reloc_root;
4932         int ret;
4933
4934         if (root->reloc_root) {
4935                 reloc_root = root->reloc_root;
4936                 root->reloc_root = NULL;
4937                 list_add(&reloc_root->dead_list,
4938                          &root->fs_info->dead_reloc_roots);
4939
4940                 btrfs_set_root_bytenr(&reloc_root->root_item,
4941                                       reloc_root->node->start);
4942                 btrfs_set_root_level(&root->root_item,
4943                                      btrfs_header_level(reloc_root->node));
4944                 memset(&reloc_root->root_item.drop_progress, 0,
4945                         sizeof(struct btrfs_disk_key));
4946                 reloc_root->root_item.drop_level = 0;
4947
4948                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4949                                         &reloc_root->root_key,
4950                                         &reloc_root->root_item);
4951                 BUG_ON(ret);
4952         }
4953         return 0;
4954 }
4955
4956 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4957 {
4958         struct btrfs_trans_handle *trans;
4959         struct btrfs_root *reloc_root;
4960         struct btrfs_root *prev_root = NULL;
4961         struct list_head dead_roots;
4962         int ret;
4963         unsigned long nr;
4964
4965         INIT_LIST_HEAD(&dead_roots);
4966         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4967
4968         while (!list_empty(&dead_roots)) {
4969                 reloc_root = list_entry(dead_roots.prev,
4970                                         struct btrfs_root, dead_list);
4971                 list_del_init(&reloc_root->dead_list);
4972
4973                 BUG_ON(reloc_root->commit_root != NULL);
4974                 while (1) {
4975                         trans = btrfs_join_transaction(root, 1);
4976                         BUG_ON(!trans);
4977
4978                         mutex_lock(&root->fs_info->drop_mutex);
4979                         ret = btrfs_drop_snapshot(trans, reloc_root);
4980                         if (ret != -EAGAIN)
4981                                 break;
4982                         mutex_unlock(&root->fs_info->drop_mutex);
4983
4984                         nr = trans->blocks_used;
4985                         ret = btrfs_end_transaction(trans, root);
4986                         BUG_ON(ret);
4987                         btrfs_btree_balance_dirty(root, nr);
4988                 }
4989
4990                 free_extent_buffer(reloc_root->node);
4991
4992                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4993                                      &reloc_root->root_key);
4994                 BUG_ON(ret);
4995                 mutex_unlock(&root->fs_info->drop_mutex);
4996
4997                 nr = trans->blocks_used;
4998                 ret = btrfs_end_transaction(trans, root);
4999                 BUG_ON(ret);
5000                 btrfs_btree_balance_dirty(root, nr);
5001
5002                 kfree(prev_root);
5003                 prev_root = reloc_root;
5004         }
5005         if (prev_root) {
5006                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5007                 kfree(prev_root);
5008         }
5009         return 0;
5010 }
5011
5012 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5013 {
5014         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5015         return 0;
5016 }
5017
5018 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5019 {
5020         struct btrfs_root *reloc_root;
5021         struct btrfs_trans_handle *trans;
5022         struct btrfs_key location;
5023         int found;
5024         int ret;
5025
5026         mutex_lock(&root->fs_info->tree_reloc_mutex);
5027         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5028         BUG_ON(ret);
5029         found = !list_empty(&root->fs_info->dead_reloc_roots);
5030         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5031
5032         if (found) {
5033                 trans = btrfs_start_transaction(root, 1);
5034                 BUG_ON(!trans);
5035                 ret = btrfs_commit_transaction(trans, root);
5036                 BUG_ON(ret);
5037         }
5038
5039         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5040         location.offset = (u64)-1;
5041         location.type = BTRFS_ROOT_ITEM_KEY;
5042
5043         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5044         BUG_ON(!reloc_root);
5045         btrfs_orphan_cleanup(reloc_root);
5046         return 0;
5047 }
5048
5049 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5050                                     struct btrfs_root *root)
5051 {
5052         struct btrfs_root *reloc_root;
5053         struct extent_buffer *eb;
5054         struct btrfs_root_item *root_item;
5055         struct btrfs_key root_key;
5056         int ret;
5057
5058         BUG_ON(!root->ref_cows);
5059         if (root->reloc_root)
5060                 return 0;
5061
5062         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5063         BUG_ON(!root_item);
5064
5065         ret = btrfs_copy_root(trans, root, root->commit_root,
5066                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5067         BUG_ON(ret);
5068
5069         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5070         root_key.offset = root->root_key.objectid;
5071         root_key.type = BTRFS_ROOT_ITEM_KEY;
5072
5073         memcpy(root_item, &root->root_item, sizeof(root_item));
5074         btrfs_set_root_refs(root_item, 0);
5075         btrfs_set_root_bytenr(root_item, eb->start);
5076         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5077         btrfs_set_root_generation(root_item, trans->transid);
5078
5079         btrfs_tree_unlock(eb);
5080         free_extent_buffer(eb);
5081
5082         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5083                                 &root_key, root_item);
5084         BUG_ON(ret);
5085         kfree(root_item);
5086
5087         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5088                                                  &root_key);
5089         BUG_ON(!reloc_root);
5090         reloc_root->last_trans = trans->transid;
5091         reloc_root->commit_root = NULL;
5092         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5093
5094         root->reloc_root = reloc_root;
5095         return 0;
5096 }
5097
5098 /*
5099  * Core function of space balance.
5100  *
5101  * The idea is using reloc trees to relocate tree blocks in reference
5102  * counted roots. There is one reloc tree for each subvol, and all
5103  * reloc trees share same root key objectid. Reloc trees are snapshots
5104  * of the latest committed roots of subvols (root->commit_root).
5105  *
5106  * To relocate a tree block referenced by a subvol, there are two steps.
5107  * COW the block through subvol's reloc tree, then update block pointer
5108  * in the subvol to point to the new block. Since all reloc trees share
5109  * same root key objectid, doing special handing for tree blocks owned
5110  * by them is easy. Once a tree block has been COWed in one reloc tree,
5111  * we can use the resulting new block directly when the same block is
5112  * required to COW again through other reloc trees. By this way, relocated
5113  * tree blocks are shared between reloc trees, so they are also shared
5114  * between subvols.
5115  */
5116 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5117                                       struct btrfs_root *root,
5118                                       struct btrfs_path *path,
5119                                       struct btrfs_key *first_key,
5120                                       struct btrfs_ref_path *ref_path,
5121                                       struct btrfs_block_group_cache *group,
5122                                       struct inode *reloc_inode)
5123 {
5124         struct btrfs_root *reloc_root;
5125         struct extent_buffer *eb = NULL;
5126         struct btrfs_key *keys;
5127         u64 *nodes;
5128         int level;
5129         int shared_level;
5130         int lowest_level = 0;
5131         int ret;
5132
5133         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5134                 lowest_level = ref_path->owner_objectid;
5135
5136         if (!root->ref_cows) {
5137                 path->lowest_level = lowest_level;
5138                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5139                 BUG_ON(ret < 0);
5140                 path->lowest_level = 0;
5141                 btrfs_release_path(root, path);
5142                 return 0;
5143         }
5144
5145         mutex_lock(&root->fs_info->tree_reloc_mutex);
5146         ret = init_reloc_tree(trans, root);
5147         BUG_ON(ret);
5148         reloc_root = root->reloc_root;
5149
5150         shared_level = ref_path->shared_level;
5151         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5152
5153         keys = ref_path->node_keys;
5154         nodes = ref_path->new_nodes;
5155         memset(&keys[shared_level + 1], 0,
5156                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5157         memset(&nodes[shared_level + 1], 0,
5158                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5159
5160         if (nodes[lowest_level] == 0) {
5161                 path->lowest_level = lowest_level;
5162                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5163                                         0, 1);
5164                 BUG_ON(ret);
5165                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5166                         eb = path->nodes[level];
5167                         if (!eb || eb == reloc_root->node)
5168                                 break;
5169                         nodes[level] = eb->start;
5170                         if (level == 0)
5171                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5172                         else
5173                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5174                 }
5175                 if (nodes[0] &&
5176                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5177                         eb = path->nodes[0];
5178                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5179                                                       group, reloc_inode);
5180                         BUG_ON(ret);
5181                 }
5182                 btrfs_release_path(reloc_root, path);
5183         } else {
5184                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5185                                        lowest_level);
5186                 BUG_ON(ret);
5187         }
5188
5189         /*
5190          * replace tree blocks in the fs tree with tree blocks in
5191          * the reloc tree.
5192          */
5193         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5194         BUG_ON(ret < 0);
5195
5196         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5197                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5198                                         0, 0);
5199                 BUG_ON(ret);
5200                 extent_buffer_get(path->nodes[0]);
5201                 eb = path->nodes[0];
5202                 btrfs_release_path(reloc_root, path);
5203                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5204                 BUG_ON(ret);
5205                 free_extent_buffer(eb);
5206         }
5207
5208         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5209         path->lowest_level = 0;
5210         return 0;
5211 }
5212
5213 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5214                                         struct btrfs_root *root,
5215                                         struct btrfs_path *path,
5216                                         struct btrfs_key *first_key,
5217                                         struct btrfs_ref_path *ref_path)
5218 {
5219         int ret;
5220
5221         ret = relocate_one_path(trans, root, path, first_key,
5222                                 ref_path, NULL, NULL);
5223         BUG_ON(ret);
5224
5225         if (root == root->fs_info->extent_root)
5226                 btrfs_extent_post_op(trans, root);
5227
5228         return 0;
5229 }
5230
5231 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5232                                     struct btrfs_root *extent_root,
5233                                     struct btrfs_path *path,
5234                                     struct btrfs_key *extent_key)
5235 {
5236         int ret;
5237
5238         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5239         if (ret)
5240                 goto out;
5241         ret = btrfs_del_item(trans, extent_root, path);
5242 out:
5243         btrfs_release_path(extent_root, path);
5244         return ret;
5245 }
5246
5247 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5248                                                 struct btrfs_ref_path *ref_path)
5249 {
5250         struct btrfs_key root_key;
5251
5252         root_key.objectid = ref_path->root_objectid;
5253         root_key.type = BTRFS_ROOT_ITEM_KEY;
5254         if (is_cowonly_root(ref_path->root_objectid))
5255                 root_key.offset = 0;
5256         else
5257                 root_key.offset = (u64)-1;
5258
5259         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5260 }
5261
5262 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5263                                         struct btrfs_path *path,
5264                                         struct btrfs_key *extent_key,
5265                                         struct btrfs_block_group_cache *group,
5266                                         struct inode *reloc_inode, int pass)
5267 {
5268         struct btrfs_trans_handle *trans;
5269         struct btrfs_root *found_root;
5270         struct btrfs_ref_path *ref_path = NULL;
5271         struct disk_extent *new_extents = NULL;
5272         int nr_extents = 0;
5273         int loops;
5274         int ret;
5275         int level;
5276         struct btrfs_key first_key;
5277         u64 prev_block = 0;
5278
5279
5280         trans = btrfs_start_transaction(extent_root, 1);
5281         BUG_ON(!trans);
5282
5283         if (extent_key->objectid == 0) {
5284                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5285                 goto out;
5286         }
5287
5288         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5289         if (!ref_path) {
5290                 ret = -ENOMEM;
5291                 goto out;
5292         }
5293
5294         for (loops = 0; ; loops++) {
5295                 if (loops == 0) {
5296                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5297                                                    extent_key->objectid);
5298                 } else {
5299                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5300                 }
5301                 if (ret < 0)
5302                         goto out;
5303                 if (ret > 0)
5304                         break;
5305
5306                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5307                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5308                         continue;
5309
5310                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5311                 BUG_ON(!found_root);
5312                 /*
5313                  * for reference counted tree, only process reference paths
5314                  * rooted at the latest committed root.
5315                  */
5316                 if (found_root->ref_cows &&
5317                     ref_path->root_generation != found_root->root_key.offset)
5318                         continue;
5319
5320                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5321                         if (pass == 0) {
5322                                 /*
5323                                  * copy data extents to new locations
5324                                  */
5325                                 u64 group_start = group->key.objectid;
5326                                 ret = relocate_data_extent(reloc_inode,
5327                                                            extent_key,
5328                                                            group_start);
5329                                 if (ret < 0)
5330                                         goto out;
5331                                 break;
5332                         }
5333                         level = 0;
5334                 } else {
5335                         level = ref_path->owner_objectid;
5336                 }
5337
5338                 if (prev_block != ref_path->nodes[level]) {
5339                         struct extent_buffer *eb;
5340                         u64 block_start = ref_path->nodes[level];
5341                         u64 block_size = btrfs_level_size(found_root, level);
5342
5343                         eb = read_tree_block(found_root, block_start,
5344                                              block_size, 0);
5345                         btrfs_tree_lock(eb);
5346                         BUG_ON(level != btrfs_header_level(eb));
5347
5348                         if (level == 0)
5349                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5350                         else
5351                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5352
5353                         btrfs_tree_unlock(eb);
5354                         free_extent_buffer(eb);
5355                         prev_block = block_start;
5356                 }
5357
5358                 btrfs_record_root_in_trans(found_root);
5359                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5360                         /*
5361                          * try to update data extent references while
5362                          * keeping metadata shared between snapshots.
5363                          */
5364                         if (pass == 1) {
5365                                 ret = relocate_one_path(trans, found_root,
5366                                                 path, &first_key, ref_path,
5367                                                 group, reloc_inode);
5368                                 if (ret < 0)
5369                                         goto out;
5370                                 continue;
5371                         }
5372                         /*
5373                          * use fallback method to process the remaining
5374                          * references.
5375                          */
5376                         if (!new_extents) {
5377                                 u64 group_start = group->key.objectid;
5378                                 new_extents = kmalloc(sizeof(*new_extents),
5379                                                       GFP_NOFS);
5380                                 nr_extents = 1;
5381                                 ret = get_new_locations(reloc_inode,
5382                                                         extent_key,
5383                                                         group_start, 1,
5384                                                         &new_extents,
5385                                                         &nr_extents);
5386                                 if (ret)
5387                                         goto out;
5388                         }
5389                         ret = replace_one_extent(trans, found_root,
5390                                                 path, extent_key,
5391                                                 &first_key, ref_path,
5392                                                 new_extents, nr_extents);
5393                 } else {
5394                         ret = relocate_tree_block(trans, found_root, path,
5395                                                   &first_key, ref_path);
5396                 }
5397                 if (ret < 0)
5398                         goto out;
5399         }
5400         ret = 0;
5401 out:
5402         btrfs_end_transaction(trans, extent_root);
5403         kfree(new_extents);
5404         kfree(ref_path);
5405         return ret;
5406 }
5407
5408 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5409 {
5410         u64 num_devices;
5411         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5412                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5413
5414         num_devices = root->fs_info->fs_devices->rw_devices;
5415         if (num_devices == 1) {
5416                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5417                 stripped = flags & ~stripped;
5418
5419                 /* turn raid0 into single device chunks */
5420                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5421                         return stripped;
5422
5423                 /* turn mirroring into duplication */
5424                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5425                              BTRFS_BLOCK_GROUP_RAID10))
5426                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5427                 return flags;
5428         } else {
5429                 /* they already had raid on here, just return */
5430                 if (flags & stripped)
5431                         return flags;
5432
5433                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5434                 stripped = flags & ~stripped;
5435
5436                 /* switch duplicated blocks with raid1 */
5437                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5438                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5439
5440                 /* turn single device chunks into raid0 */
5441                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5442         }
5443         return flags;
5444 }
5445
5446 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5447                      struct btrfs_block_group_cache *shrink_block_group,
5448                      int force)
5449 {
5450         struct btrfs_trans_handle *trans;
5451         u64 new_alloc_flags;
5452         u64 calc;
5453
5454         spin_lock(&shrink_block_group->lock);
5455         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5456                 spin_unlock(&shrink_block_group->lock);
5457
5458                 trans = btrfs_start_transaction(root, 1);
5459                 spin_lock(&shrink_block_group->lock);
5460
5461                 new_alloc_flags = update_block_group_flags(root,
5462                                                    shrink_block_group->flags);
5463                 if (new_alloc_flags != shrink_block_group->flags) {
5464                         calc =
5465                              btrfs_block_group_used(&shrink_block_group->item);
5466                 } else {
5467                         calc = shrink_block_group->key.offset;
5468                 }
5469                 spin_unlock(&shrink_block_group->lock);
5470
5471                 do_chunk_alloc(trans, root->fs_info->extent_root,
5472                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5473
5474                 btrfs_end_transaction(trans, root);
5475         } else
5476                 spin_unlock(&shrink_block_group->lock);
5477         return 0;
5478 }
5479
5480 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5481                                  struct btrfs_root *root,
5482                                  u64 objectid, u64 size)
5483 {
5484         struct btrfs_path *path;
5485         struct btrfs_inode_item *item;
5486         struct extent_buffer *leaf;
5487         int ret;
5488
5489         path = btrfs_alloc_path();
5490         if (!path)
5491                 return -ENOMEM;
5492
5493         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5494         if (ret)
5495                 goto out;
5496
5497         leaf = path->nodes[0];
5498         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5499         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5500         btrfs_set_inode_generation(leaf, item, 1);
5501         btrfs_set_inode_size(leaf, item, size);
5502         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5503         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5504         btrfs_mark_buffer_dirty(leaf);
5505         btrfs_release_path(root, path);
5506 out:
5507         btrfs_free_path(path);
5508         return ret;
5509 }
5510
5511 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5512                                         struct btrfs_block_group_cache *group)
5513 {
5514         struct inode *inode = NULL;
5515         struct btrfs_trans_handle *trans;
5516         struct btrfs_root *root;
5517         struct btrfs_key root_key;
5518         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5519         int err = 0;
5520
5521         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5522         root_key.type = BTRFS_ROOT_ITEM_KEY;
5523         root_key.offset = (u64)-1;
5524         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5525         if (IS_ERR(root))
5526                 return ERR_CAST(root);
5527
5528         trans = btrfs_start_transaction(root, 1);
5529         BUG_ON(!trans);
5530
5531         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5532         if (err)
5533                 goto out;
5534
5535         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5536         BUG_ON(err);
5537
5538         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5539                                        group->key.offset, 0, group->key.offset,
5540                                        0, 0, 0);
5541         BUG_ON(err);
5542
5543         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5544         if (inode->i_state & I_NEW) {
5545                 BTRFS_I(inode)->root = root;
5546                 BTRFS_I(inode)->location.objectid = objectid;
5547                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5548                 BTRFS_I(inode)->location.offset = 0;
5549                 btrfs_read_locked_inode(inode);
5550                 unlock_new_inode(inode);
5551                 BUG_ON(is_bad_inode(inode));
5552         } else {
5553                 BUG_ON(1);
5554         }
5555         BTRFS_I(inode)->index_cnt = group->key.objectid;
5556
5557         err = btrfs_orphan_add(trans, inode);
5558 out:
5559         btrfs_end_transaction(trans, root);
5560         if (err) {
5561                 if (inode)
5562                         iput(inode);
5563                 inode = ERR_PTR(err);
5564         }
5565         return inode;
5566 }
5567
5568 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5569 {
5570
5571         struct btrfs_ordered_sum *sums;
5572         struct btrfs_sector_sum *sector_sum;
5573         struct btrfs_ordered_extent *ordered;
5574         struct btrfs_root *root = BTRFS_I(inode)->root;
5575         struct list_head list;
5576         size_t offset;
5577         int ret;
5578         u64 disk_bytenr;
5579
5580         INIT_LIST_HEAD(&list);
5581
5582         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5583         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5584
5585         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5586         ret = btrfs_lookup_csums_range(root, disk_bytenr,
5587                                        disk_bytenr + len - 1, &list);
5588
5589         while (!list_empty(&list)) {
5590                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5591                 list_del_init(&sums->list);
5592
5593                 sector_sum = sums->sums;
5594                 sums->bytenr = ordered->start;
5595
5596                 offset = 0;
5597                 while (offset < sums->len) {
5598                         sector_sum->bytenr += ordered->start - disk_bytenr;
5599                         sector_sum++;
5600                         offset += root->sectorsize;
5601                 }
5602
5603                 btrfs_add_ordered_sum(inode, ordered, sums);
5604         }
5605         btrfs_put_ordered_extent(ordered);
5606         return 0;
5607 }
5608
5609 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5610 {
5611         struct btrfs_trans_handle *trans;
5612         struct btrfs_path *path;
5613         struct btrfs_fs_info *info = root->fs_info;
5614         struct extent_buffer *leaf;
5615         struct inode *reloc_inode;
5616         struct btrfs_block_group_cache *block_group;
5617         struct btrfs_key key;
5618         u64 skipped;
5619         u64 cur_byte;
5620         u64 total_found;
5621         u32 nritems;
5622         int ret;
5623         int progress;
5624         int pass = 0;
5625
5626         root = root->fs_info->extent_root;
5627
5628         block_group = btrfs_lookup_block_group(info, group_start);
5629         BUG_ON(!block_group);
5630
5631         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5632                (unsigned long long)block_group->key.objectid,
5633                (unsigned long long)block_group->flags);
5634
5635         path = btrfs_alloc_path();
5636         BUG_ON(!path);
5637
5638         reloc_inode = create_reloc_inode(info, block_group);
5639         BUG_ON(IS_ERR(reloc_inode));
5640
5641         __alloc_chunk_for_shrink(root, block_group, 1);
5642         set_block_group_readonly(block_group);
5643
5644         btrfs_start_delalloc_inodes(info->tree_root);
5645         btrfs_wait_ordered_extents(info->tree_root, 0);
5646 again:
5647         skipped = 0;
5648         total_found = 0;
5649         progress = 0;
5650         key.objectid = block_group->key.objectid;
5651         key.offset = 0;
5652         key.type = 0;
5653         cur_byte = key.objectid;
5654
5655         trans = btrfs_start_transaction(info->tree_root, 1);
5656         btrfs_commit_transaction(trans, info->tree_root);
5657
5658         mutex_lock(&root->fs_info->cleaner_mutex);
5659         btrfs_clean_old_snapshots(info->tree_root);
5660         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5661         mutex_unlock(&root->fs_info->cleaner_mutex);
5662
5663         while (1) {
5664                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5665                 if (ret < 0)
5666                         goto out;
5667 next:
5668                 leaf = path->nodes[0];
5669                 nritems = btrfs_header_nritems(leaf);
5670                 if (path->slots[0] >= nritems) {
5671                         ret = btrfs_next_leaf(root, path);
5672                         if (ret < 0)
5673                                 goto out;
5674                         if (ret == 1) {
5675                                 ret = 0;
5676                                 break;
5677                         }
5678                         leaf = path->nodes[0];
5679                         nritems = btrfs_header_nritems(leaf);
5680                 }
5681
5682                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5683
5684                 if (key.objectid >= block_group->key.objectid +
5685                     block_group->key.offset)
5686                         break;
5687
5688                 if (progress && need_resched()) {
5689                         btrfs_release_path(root, path);
5690                         cond_resched();
5691                         progress = 0;
5692                         continue;
5693                 }
5694                 progress = 1;
5695
5696                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5697                     key.objectid + key.offset <= cur_byte) {
5698                         path->slots[0]++;
5699                         goto next;
5700                 }
5701
5702                 total_found++;
5703                 cur_byte = key.objectid + key.offset;
5704                 btrfs_release_path(root, path);
5705
5706                 __alloc_chunk_for_shrink(root, block_group, 0);
5707                 ret = relocate_one_extent(root, path, &key, block_group,
5708                                           reloc_inode, pass);
5709                 BUG_ON(ret < 0);
5710                 if (ret > 0)
5711                         skipped++;
5712
5713                 key.objectid = cur_byte;
5714                 key.type = 0;
5715                 key.offset = 0;
5716         }
5717
5718         btrfs_release_path(root, path);
5719
5720         if (pass == 0) {
5721                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5722                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5723         }
5724
5725         if (total_found > 0) {
5726                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5727                        (unsigned long long)total_found, pass);
5728                 pass++;
5729                 if (total_found == skipped && pass > 2) {
5730                         iput(reloc_inode);
5731                         reloc_inode = create_reloc_inode(info, block_group);
5732                         pass = 0;
5733                 }
5734                 goto again;
5735         }
5736
5737         /* delete reloc_inode */
5738         iput(reloc_inode);
5739
5740         /* unpin extents in this range */
5741         trans = btrfs_start_transaction(info->tree_root, 1);
5742         btrfs_commit_transaction(trans, info->tree_root);
5743
5744         spin_lock(&block_group->lock);
5745         WARN_ON(block_group->pinned > 0);
5746         WARN_ON(block_group->reserved > 0);
5747         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5748         spin_unlock(&block_group->lock);
5749         put_block_group(block_group);
5750         ret = 0;
5751 out:
5752         btrfs_free_path(path);
5753         return ret;
5754 }
5755
5756 static int find_first_block_group(struct btrfs_root *root,
5757                 struct btrfs_path *path, struct btrfs_key *key)
5758 {
5759         int ret = 0;
5760         struct btrfs_key found_key;
5761         struct extent_buffer *leaf;
5762         int slot;
5763
5764         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5765         if (ret < 0)
5766                 goto out;
5767
5768         while (1) {
5769                 slot = path->slots[0];
5770                 leaf = path->nodes[0];
5771                 if (slot >= btrfs_header_nritems(leaf)) {
5772                         ret = btrfs_next_leaf(root, path);
5773                         if (ret == 0)
5774                                 continue;
5775                         if (ret < 0)
5776                                 goto out;
5777                         break;
5778                 }
5779                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5780
5781                 if (found_key.objectid >= key->objectid &&
5782                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5783                         ret = 0;
5784                         goto out;
5785                 }
5786                 path->slots[0]++;
5787         }
5788         ret = -ENOENT;
5789 out:
5790         return ret;
5791 }
5792
5793 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5794 {
5795         struct btrfs_block_group_cache *block_group;
5796         struct rb_node *n;
5797
5798         spin_lock(&info->block_group_cache_lock);
5799         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5800                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5801                                        cache_node);
5802                 rb_erase(&block_group->cache_node,
5803                          &info->block_group_cache_tree);
5804                 spin_unlock(&info->block_group_cache_lock);
5805
5806                 btrfs_remove_free_space_cache(block_group);
5807                 down_write(&block_group->space_info->groups_sem);
5808                 list_del(&block_group->list);
5809                 up_write(&block_group->space_info->groups_sem);
5810
5811                 WARN_ON(atomic_read(&block_group->count) != 1);
5812                 kfree(block_group);
5813
5814                 spin_lock(&info->block_group_cache_lock);
5815         }
5816         spin_unlock(&info->block_group_cache_lock);
5817         return 0;
5818 }
5819
5820 int btrfs_read_block_groups(struct btrfs_root *root)
5821 {
5822         struct btrfs_path *path;
5823         int ret;
5824         struct btrfs_block_group_cache *cache;
5825         struct btrfs_fs_info *info = root->fs_info;
5826         struct btrfs_space_info *space_info;
5827         struct btrfs_key key;
5828         struct btrfs_key found_key;
5829         struct extent_buffer *leaf;
5830
5831         root = info->extent_root;
5832         key.objectid = 0;
5833         key.offset = 0;
5834         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5835         path = btrfs_alloc_path();
5836         if (!path)
5837                 return -ENOMEM;
5838
5839         while (1) {
5840                 ret = find_first_block_group(root, path, &key);
5841                 if (ret > 0) {
5842                         ret = 0;
5843                         goto error;
5844                 }
5845                 if (ret != 0)
5846                         goto error;
5847
5848                 leaf = path->nodes[0];
5849                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5850                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5851                 if (!cache) {
5852                         ret = -ENOMEM;
5853                         break;
5854                 }
5855
5856                 atomic_set(&cache->count, 1);
5857                 spin_lock_init(&cache->lock);
5858                 mutex_init(&cache->alloc_mutex);
5859                 mutex_init(&cache->cache_mutex);
5860                 INIT_LIST_HEAD(&cache->list);
5861                 read_extent_buffer(leaf, &cache->item,
5862                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5863                                    sizeof(cache->item));
5864                 memcpy(&cache->key, &found_key, sizeof(found_key));
5865
5866                 key.objectid = found_key.objectid + found_key.offset;
5867                 btrfs_release_path(root, path);
5868                 cache->flags = btrfs_block_group_flags(&cache->item);
5869
5870                 ret = update_space_info(info, cache->flags, found_key.offset,
5871                                         btrfs_block_group_used(&cache->item),
5872                                         &space_info);
5873                 BUG_ON(ret);
5874                 cache->space_info = space_info;
5875                 down_write(&space_info->groups_sem);
5876                 list_add_tail(&cache->list, &space_info->block_groups);
5877                 up_write(&space_info->groups_sem);
5878
5879                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5880                 BUG_ON(ret);
5881
5882                 set_avail_alloc_bits(root->fs_info, cache->flags);
5883                 if (btrfs_chunk_readonly(root, cache->key.objectid))
5884                         set_block_group_readonly(cache);
5885         }
5886         ret = 0;
5887 error:
5888         btrfs_free_path(path);
5889         return ret;
5890 }
5891
5892 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5893                            struct btrfs_root *root, u64 bytes_used,
5894                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5895                            u64 size)
5896 {
5897         int ret;
5898         struct btrfs_root *extent_root;
5899         struct btrfs_block_group_cache *cache;
5900
5901         extent_root = root->fs_info->extent_root;
5902
5903         root->fs_info->last_trans_new_blockgroup = trans->transid;
5904
5905         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5906         if (!cache)
5907                 return -ENOMEM;
5908
5909         cache->key.objectid = chunk_offset;
5910         cache->key.offset = size;
5911         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5912         atomic_set(&cache->count, 1);
5913         spin_lock_init(&cache->lock);
5914         mutex_init(&cache->alloc_mutex);
5915         mutex_init(&cache->cache_mutex);
5916         INIT_LIST_HEAD(&cache->list);
5917
5918         btrfs_set_block_group_used(&cache->item, bytes_used);
5919         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5920         cache->flags = type;
5921         btrfs_set_block_group_flags(&cache->item, type);
5922
5923         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5924                                 &cache->space_info);
5925         BUG_ON(ret);
5926         down_write(&cache->space_info->groups_sem);
5927         list_add_tail(&cache->list, &cache->space_info->block_groups);
5928         up_write(&cache->space_info->groups_sem);
5929
5930         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5931         BUG_ON(ret);
5932
5933         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5934                                 sizeof(cache->item));
5935         BUG_ON(ret);
5936
5937         finish_current_insert(trans, extent_root, 0);
5938         ret = del_pending_extents(trans, extent_root, 0);
5939         BUG_ON(ret);
5940         set_avail_alloc_bits(extent_root->fs_info, type);
5941
5942         return 0;
5943 }
5944
5945 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5946                              struct btrfs_root *root, u64 group_start)
5947 {
5948         struct btrfs_path *path;
5949         struct btrfs_block_group_cache *block_group;
5950         struct btrfs_key key;
5951         int ret;
5952
5953         root = root->fs_info->extent_root;
5954
5955         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5956         BUG_ON(!block_group);
5957         BUG_ON(!block_group->ro);
5958
5959         memcpy(&key, &block_group->key, sizeof(key));
5960
5961         path = btrfs_alloc_path();
5962         BUG_ON(!path);
5963
5964         btrfs_remove_free_space_cache(block_group);
5965         rb_erase(&block_group->cache_node,
5966                  &root->fs_info->block_group_cache_tree);
5967         down_write(&block_group->space_info->groups_sem);
5968         list_del(&block_group->list);
5969         up_write(&block_group->space_info->groups_sem);
5970
5971         spin_lock(&block_group->space_info->lock);
5972         block_group->space_info->total_bytes -= block_group->key.offset;
5973         block_group->space_info->bytes_readonly -= block_group->key.offset;
5974         spin_unlock(&block_group->space_info->lock);
5975         block_group->space_info->full = 0;
5976
5977         put_block_group(block_group);
5978         put_block_group(block_group);
5979
5980         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5981         if (ret > 0)
5982                 ret = -EIO;
5983         if (ret < 0)
5984                 goto out;
5985
5986         ret = btrfs_del_item(trans, root, path);
5987 out:
5988         btrfs_free_path(path);
5989         return ret;
5990 }