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

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