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

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include "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 #include "compat.h"
  33
  34 #define PENDING_EXTENT_INSERT 0
  35 #define PENDING_EXTENT_DELETE 1
  36 #define PENDING_BACKREF_UPDATE 2
  37
  38 struct pending_extent_op {
  39         int type;
  40         u64 bytenr;
  41         u64 num_bytes;
  42         u64 parent;
  43         u64 orig_parent;
  44         u64 generation;
  45         u64 orig_generation;
  46         int level;
  47         struct list_head list;
  48         int del;
  49 };
  50
  51 static int finish_current_insert(struct btrfs_trans_handle *trans,
  52                                  struct btrfs_root *extent_root, int all);
  53 static int del_pending_extents(struct btrfs_trans_handle *trans,
  54                                struct btrfs_root *extent_root, int all);
  55 static int pin_down_bytes(struct btrfs_trans_handle *trans,
  56                           struct btrfs_root *root,
  57                           u64 bytenr, u64 num_bytes, int is_data);
  58 static int update_block_group(struct btrfs_trans_handle *trans,
  59                               struct btrfs_root *root,
  60                               u64 bytenr, u64 num_bytes, int alloc,
  61                               int mark_free);
  62
  63 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  64 {
  65         return (cache->flags & bits) == bits;
  66 }
  67
  68 /*
  69  * this adds the block group to the fs_info rb tree for the block group
  70  * cache
  71  */
  72 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  73                                 struct btrfs_block_group_cache *block_group)
  74 {
  75         struct rb_node **p;
  76         struct rb_node *parent = NULL;
  77         struct btrfs_block_group_cache *cache;
  78
  79         spin_lock(&info->block_group_cache_lock);
  80         p = &info->block_group_cache_tree.rb_node;
  81
  82         while (*p) {
  83                 parent = *p;
  84                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  85                                  cache_node);
  86                 if (block_group->key.objectid < cache->key.objectid) {
  87                         p = &(*p)->rb_left;
  88                 } else if (block_group->key.objectid > cache->key.objectid) {
  89                         p = &(*p)->rb_right;
  90                 } else {
  91                         spin_unlock(&info->block_group_cache_lock);
  92                         return -EEXIST;
  93                 }
  94         }
  95
  96         rb_link_node(&block_group->cache_node, parent, p);
  97         rb_insert_color(&block_group->cache_node,
  98                         &info->block_group_cache_tree);
  99         spin_unlock(&info->block_group_cache_lock);
 100
 101         return 0;
 102 }
 103
 104 /*
 105  * This will return the block group at or after bytenr if contains is 0, else
 106  * it will return the block group that contains the bytenr
 107  */
 108 static struct btrfs_block_group_cache *
 109 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 110                               int contains)
 111 {
 112         struct btrfs_block_group_cache *cache, *ret = NULL;
 113         struct rb_node *n;
 114         u64 end, start;
 115
 116         spin_lock(&info->block_group_cache_lock);
 117         n = info->block_group_cache_tree.rb_node;
 118
 119         while (n) {
 120                 cache = rb_entry(n, struct btrfs_block_group_cache,
 121                                  cache_node);
 122                 end = cache->key.objectid + cache->key.offset - 1;
 123                 start = cache->key.objectid;
 124
 125                 if (bytenr < start) {
 126                         if (!contains && (!ret || start < ret->key.objectid))
 127                                 ret = cache;
 128                         n = n->rb_left;
 129                 } else if (bytenr > start) {
 130                         if (contains && bytenr <= end) {
 131                                 ret = cache;
 132                                 break;
 133                         }
 134                         n = n->rb_right;
 135                 } else {
 136                         ret = cache;
 137                         break;
 138                 }
 139         }
 140         if (ret)
 141                 atomic_inc(&ret->count);
 142         spin_unlock(&info->block_group_cache_lock);
 143
 144         return ret;
 145 }
 146
 147 /*
 148  * this is only called by cache_block_group, since we could have freed extents
 149  * we need to check the pinned_extents for any extents that can't be used yet
 150  * since their free space will be released as soon as the transaction commits.
 151  */
 152 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 153                               struct btrfs_fs_info *info, u64 start, u64 end)
 154 {
 155         u64 extent_start, extent_end, size;
 156         int ret;
 157
 158         mutex_lock(&info->pinned_mutex);
 159         while (start < end) {
 160                 ret = find_first_extent_bit(&info->pinned_extents, start,
 161                                             &extent_start, &extent_end,
 162                                             EXTENT_DIRTY);
 163                 if (ret)
 164                         break;
 165
 166                 if (extent_start == start) {
 167                         start = extent_end + 1;
 168                 } else if (extent_start > start && extent_start < end) {
 169                         size = extent_start - start;
 170                         ret = btrfs_add_free_space(block_group, start,
 171                                                    size);
 172                         BUG_ON(ret);
 173                         start = extent_end + 1;
 174                 } else {
 175                         break;
 176                 }
 177         }
 178
 179         if (start < end) {
 180                 size = end - start;
 181                 ret = btrfs_add_free_space(block_group, start, size);
 182                 BUG_ON(ret);
 183         }
 184         mutex_unlock(&info->pinned_mutex);
 185
 186         return 0;
 187 }
 188
 189 static int remove_sb_from_cache(struct btrfs_root *root,
 190                                 struct btrfs_block_group_cache *cache)
 191 {
 192         u64 bytenr;
 193         u64 *logical;
 194         int stripe_len;
 195         int i, nr, ret;
 196
 197         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 198                 bytenr = btrfs_sb_offset(i);
 199                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 200                                        cache->key.objectid, bytenr, 0,
 201                                        &logical, &nr, &stripe_len);
 202                 BUG_ON(ret);
 203                 while (nr--) {
 204                         btrfs_remove_free_space(cache, logical[nr],
 205                                                 stripe_len);
 206                 }
 207                 kfree(logical);
 208         }
 209         return 0;
 210 }
 211
 212 static int cache_block_group(struct btrfs_root *root,
 213                              struct btrfs_block_group_cache *block_group)
 214 {
 215         struct btrfs_path *path;
 216         int ret = 0;
 217         struct btrfs_key key;
 218         struct extent_buffer *leaf;
 219         int slot;
 220         u64 last;
 221
 222         if (!block_group)
 223                 return 0;
 224
 225         root = root->fs_info->extent_root;
 226
 227         if (block_group->cached)
 228                 return 0;
 229
 230         path = btrfs_alloc_path();
 231         if (!path)
 232                 return -ENOMEM;
 233
 234         path->reada = 2;
 235         /*
 236          * we get into deadlocks with paths held by callers of this function.
 237          * since the alloc_mutex is protecting things right now, just
 238          * skip the locking here
 239          */
 240         path->skip_locking = 1;
 241         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 242         key.objectid = last;
 243         key.offset = 0;
 244         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 245         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 246         if (ret < 0)
 247                 goto err;
 248
 249         while (1) {
 250                 leaf = path->nodes[0];
 251                 slot = path->slots[0];
 252                 if (slot >= btrfs_header_nritems(leaf)) {
 253                         ret = btrfs_next_leaf(root, path);
 254                         if (ret < 0)
 255                                 goto err;
 256                         if (ret == 0)
 257                                 continue;
 258                         else
 259                                 break;
 260                 }
 261                 btrfs_item_key_to_cpu(leaf, &key, slot);
 262                 if (key.objectid < block_group->key.objectid)
 263                         goto next;
 264
 265                 if (key.objectid >= block_group->key.objectid +
 266                     block_group->key.offset)
 267                         break;
 268
 269                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 270                         add_new_free_space(block_group, root->fs_info, last,
 271                                            key.objectid);
 272
 273                         last = key.objectid + key.offset;
 274                 }
 275 next:
 276                 path->slots[0]++;
 277         }
 278
 279         add_new_free_space(block_group, root->fs_info, last,
 280                            block_group->key.objectid +
 281                            block_group->key.offset);
 282
 283         remove_sb_from_cache(root, block_group);
 284         block_group->cached = 1;
 285         ret = 0;
 286 err:
 287         btrfs_free_path(path);
 288         return ret;
 289 }
 290
 291 /*
 292  * return the block group that starts at or after bytenr
 293  */
 294 static struct btrfs_block_group_cache *
 295 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 296 {
 297         struct btrfs_block_group_cache *cache;
 298
 299         cache = block_group_cache_tree_search(info, bytenr, 0);
 300
 301         return cache;
 302 }
 303
 304 /*
 305  * return the block group that contains teh given bytenr
 306  */
 307 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 308                                                  struct btrfs_fs_info *info,
 309                                                  u64 bytenr)
 310 {
 311         struct btrfs_block_group_cache *cache;
 312
 313         cache = block_group_cache_tree_search(info, bytenr, 1);
 314
 315         return cache;
 316 }
 317
 318 static inline void put_block_group(struct btrfs_block_group_cache *cache)
 319 {
 320         if (atomic_dec_and_test(&cache->count))
 321                 kfree(cache);
 322 }
 323
 324 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 325                                                   u64 flags)
 326 {
 327         struct list_head *head = &info->space_info;
 328         struct btrfs_space_info *found;
 329         list_for_each_entry(found, head, list) {
 330                 if (found->flags == flags)
 331                         return found;
 332         }
 333         return NULL;
 334 }
 335
 336 static u64 div_factor(u64 num, int factor)
 337 {
 338         if (factor == 10)
 339                 return num;
 340         num *= factor;
 341         do_div(num, 10);
 342         return num;
 343 }
 344
 345 u64 btrfs_find_block_group(struct btrfs_root *root,
 346                            u64 search_start, u64 search_hint, int owner)
 347 {
 348         struct btrfs_block_group_cache *cache;
 349         u64 used;
 350         u64 last = max(search_hint, search_start);
 351         u64 group_start = 0;
 352         int full_search = 0;
 353         int factor = 9;
 354         int wrapped = 0;
 355 again:
 356         while (1) {
 357                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 358                 if (!cache)
 359                         break;
 360
 361                 spin_lock(&cache->lock);
 362                 last = cache->key.objectid + cache->key.offset;
 363                 used = btrfs_block_group_used(&cache->item);
 364
 365                 if ((full_search || !cache->ro) &&
 366                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 367                         if (used + cache->pinned + cache->reserved <
 368                             div_factor(cache->key.offset, factor)) {
 369                                 group_start = cache->key.objectid;
 370                                 spin_unlock(&cache->lock);
 371                                 put_block_group(cache);
 372                                 goto found;
 373                         }
 374                 }
 375                 spin_unlock(&cache->lock);
 376                 put_block_group(cache);
 377                 cond_resched();
 378         }
 379         if (!wrapped) {
 380                 last = search_start;
 381                 wrapped = 1;
 382                 goto again;
 383         }
 384         if (!full_search && factor < 10) {
 385                 last = search_start;
 386                 full_search = 1;
 387                 factor = 10;
 388                 goto again;
 389         }
 390 found:
 391         return group_start;
 392 }
 393
 394 /* simple helper to search for an existing extent at a given offset */
 395 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 396 {
 397         int ret;
 398         struct btrfs_key key;
 399         struct btrfs_path *path;
 400
 401         path = btrfs_alloc_path();
 402         BUG_ON(!path);
 403         key.objectid = start;
 404         key.offset = len;
 405         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 406         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 407                                 0, 0);
 408         btrfs_free_path(path);
 409         return ret;
 410 }
 411
 412 /*
 413  * Back reference rules.  Back refs have three main goals:
 414  *
 415  * 1) differentiate between all holders of references to an extent so that
 416  *    when a reference is dropped we can make sure it was a valid reference
 417  *    before freeing the extent.
 418  *
 419  * 2) Provide enough information to quickly find the holders of an extent
 420  *    if we notice a given block is corrupted or bad.
 421  *
 422  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 423  *    maintenance.  This is actually the same as #2, but with a slightly
 424  *    different use case.
 425  *
 426  * File extents can be referenced by:
 427  *
 428  * - multiple snapshots, subvolumes, or different generations in one subvol
 429  * - different files inside a single subvolume
 430  * - different offsets inside a file (bookend extents in file.c)
 431  *
 432  * The extent ref structure has fields for:
 433  *
 434  * - Objectid of the subvolume root
 435  * - Generation number of the tree holding the reference
 436  * - objectid of the file holding the reference
 437  * - number of references holding by parent node (alway 1 for tree blocks)
 438  *
 439  * Btree leaf may hold multiple references to a file extent. In most cases,
 440  * these references are from same file and the corresponding offsets inside
 441  * the file are close together.
 442  *
 443  * When a file extent is allocated the fields are filled in:
 444  *     (root_key.objectid, trans->transid, inode objectid, 1)
 445  *
 446  * When a leaf is cow'd new references are added for every file extent found
 447  * in the leaf.  It looks similar to the create case, but trans->transid will
 448  * be different when the block is cow'd.
 449  *
 450  *     (root_key.objectid, trans->transid, inode objectid,
 451  *      number of references in the leaf)
 452  *
 453  * When a file extent is removed either during snapshot deletion or
 454  * file truncation, we find the corresponding back reference and check
 455  * the following fields:
 456  *
 457  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 458  *      inode objectid)
 459  *
 460  * Btree extents can be referenced by:
 461  *
 462  * - Different subvolumes
 463  * - Different generations of the same subvolume
 464  *
 465  * When a tree block is created, back references are inserted:
 466  *
 467  * (root->root_key.objectid, trans->transid, level, 1)
 468  *
 469  * When a tree block is cow'd, new back references are added for all the
 470  * blocks it points to. If the tree block isn't in reference counted root,
 471  * the old back references are removed. These new back references are of
 472  * the form (trans->transid will have increased since creation):
 473  *
 474  * (root->root_key.objectid, trans->transid, level, 1)
 475  *
 476  * When a backref is in deleting, the following fields are checked:
 477  *
 478  * if backref was for a tree root:
 479  *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
 480  * else
 481  *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
 482  *
 483  * Back Reference Key composing:
 484  *
 485  * The key objectid corresponds to the first byte in the extent, the key
 486  * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
 487  * byte of parent extent. If a extent is tree root, the key offset is set
 488  * to the key objectid.
 489  */
 490
 491 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
 492                                           struct btrfs_root *root,
 493                                           struct btrfs_path *path,
 494                                           u64 bytenr, u64 parent,
 495                                           u64 ref_root, u64 ref_generation,
 496                                           u64 owner_objectid, int del)
 497 {
 498         struct btrfs_key key;
 499         struct btrfs_extent_ref *ref;
 500         struct extent_buffer *leaf;
 501         u64 ref_objectid;
 502         int ret;
 503
 504         key.objectid = bytenr;
 505         key.type = BTRFS_EXTENT_REF_KEY;
 506         key.offset = parent;
 507
 508         ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
 509         if (ret < 0)
 510                 goto out;
 511         if (ret > 0) {
 512                 ret = -ENOENT;
 513                 goto out;
 514         }
 515
 516         leaf = path->nodes[0];
 517         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 518         ref_objectid = btrfs_ref_objectid(leaf, ref);
 519         if (btrfs_ref_root(leaf, ref) != ref_root ||
 520             btrfs_ref_generation(leaf, ref) != ref_generation ||
 521             (ref_objectid != owner_objectid &&
 522              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 523                 ret = -EIO;
 524                 WARN_ON(1);
 525                 goto out;
 526         }
 527         ret = 0;
 528 out:
 529         return ret;
 530 }
 531
 532 /*
 533  * updates all the backrefs that are pending on update_list for the
 534  * extent_root
 535  */
 536 static noinline int update_backrefs(struct btrfs_trans_handle *trans,
 537                                     struct btrfs_root *extent_root,
 538                                     struct btrfs_path *path,
 539                                     struct list_head *update_list)
 540 {
 541         struct btrfs_key key;
 542         struct btrfs_extent_ref *ref;
 543         struct btrfs_fs_info *info = extent_root->fs_info;
 544         struct pending_extent_op *op;
 545         struct extent_buffer *leaf;
 546         int ret = 0;
 547         struct list_head *cur = update_list->next;
 548         u64 ref_objectid;
 549         u64 ref_root = extent_root->root_key.objectid;
 550
 551         op = list_entry(cur, struct pending_extent_op, list);
 552
 553 search:
 554         key.objectid = op->bytenr;
 555         key.type = BTRFS_EXTENT_REF_KEY;
 556         key.offset = op->orig_parent;
 557
 558         ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
 559         BUG_ON(ret);
 560
 561         leaf = path->nodes[0];
 562
 563 loop:
 564         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 565
 566         ref_objectid = btrfs_ref_objectid(leaf, ref);
 567
 568         if (btrfs_ref_root(leaf, ref) != ref_root ||
 569             btrfs_ref_generation(leaf, ref) != op->orig_generation ||
 570             (ref_objectid != op->level &&
 571              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 572                 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
 573                        "root %llu, owner %u\n",
 574                        (unsigned long long)op->bytenr,
 575                        (unsigned long long)op->orig_parent,
 576                        (unsigned long long)ref_root, op->level);
 577                 btrfs_print_leaf(extent_root, leaf);
 578                 BUG();
 579         }
 580
 581         key.objectid = op->bytenr;
 582         key.offset = op->parent;
 583         key.type = BTRFS_EXTENT_REF_KEY;
 584         ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
 585         BUG_ON(ret);
 586         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 587         btrfs_set_ref_generation(leaf, ref, op->generation);
 588
 589         cur = cur->next;
 590
 591         list_del_init(&op->list);
 592         unlock_extent(&info->extent_ins, op->bytenr,
 593                       op->bytenr + op->num_bytes - 1, GFP_NOFS);
 594         kfree(op);
 595
 596         if (cur == update_list) {
 597                 btrfs_mark_buffer_dirty(path->nodes[0]);
 598                 btrfs_release_path(extent_root, path);
 599                 goto out;
 600         }
 601
 602         op = list_entry(cur, struct pending_extent_op, list);
 603
 604         path->slots[0]++;
 605         while (path->slots[0] < btrfs_header_nritems(leaf)) {
 606                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 607                 if (key.objectid == op->bytenr &&
 608                     key.type == BTRFS_EXTENT_REF_KEY)
 609                         goto loop;
 610                 path->slots[0]++;
 611         }
 612
 613         btrfs_mark_buffer_dirty(path->nodes[0]);
 614         btrfs_release_path(extent_root, path);
 615         goto search;
 616
 617 out:
 618         return 0;
 619 }
 620
 621 static noinline int insert_extents(struct btrfs_trans_handle *trans,
 622                                    struct btrfs_root *extent_root,
 623                                    struct btrfs_path *path,
 624                                    struct list_head *insert_list, int nr)
 625 {
 626         struct btrfs_key *keys;
 627         u32 *data_size;
 628         struct pending_extent_op *op;
 629         struct extent_buffer *leaf;
 630         struct list_head *cur = insert_list->next;
 631         struct btrfs_fs_info *info = extent_root->fs_info;
 632         u64 ref_root = extent_root->root_key.objectid;
 633         int i = 0, last = 0, ret;
 634         int total = nr * 2;
 635
 636         if (!nr)
 637                 return 0;
 638
 639         keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
 640         if (!keys)
 641                 return -ENOMEM;
 642
 643         data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
 644         if (!data_size) {
 645                 kfree(keys);
 646                 return -ENOMEM;
 647         }
 648
 649         list_for_each_entry(op, insert_list, list) {
 650                 keys[i].objectid = op->bytenr;
 651                 keys[i].offset = op->num_bytes;
 652                 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
 653                 data_size[i] = sizeof(struct btrfs_extent_item);
 654                 i++;
 655
 656                 keys[i].objectid = op->bytenr;
 657                 keys[i].offset = op->parent;
 658                 keys[i].type = BTRFS_EXTENT_REF_KEY;
 659                 data_size[i] = sizeof(struct btrfs_extent_ref);
 660                 i++;
 661         }
 662
 663         op = list_entry(cur, struct pending_extent_op, list);
 664         i = 0;
 665         while (i < total) {
 666                 int c;
 667                 ret = btrfs_insert_some_items(trans, extent_root, path,
 668                                               keys+i, data_size+i, total-i);
 669                 BUG_ON(ret < 0);
 670
 671                 if (last && ret > 1)
 672                         BUG();
 673
 674                 leaf = path->nodes[0];
 675                 for (c = 0; c < ret; c++) {
 676                         int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
 677
 678                         /*
 679                          * if the first item we inserted was a backref, then
 680                          * the EXTENT_ITEM will be the odd c's, else it will
 681                          * be the even c's
 682                          */
 683                         if ((ref_first && (c % 2)) ||
 684                             (!ref_first && !(c % 2))) {
 685                                 struct btrfs_extent_item *itm;
 686
 687                                 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
 688                                                      struct btrfs_extent_item);
 689                                 btrfs_set_extent_refs(path->nodes[0], itm, 1);
 690                                 op->del++;
 691                         } else {
 692                                 struct btrfs_extent_ref *ref;
 693
 694                                 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
 695                                                      struct btrfs_extent_ref);
 696                                 btrfs_set_ref_root(leaf, ref, ref_root);
 697                                 btrfs_set_ref_generation(leaf, ref,
 698                                                          op->generation);
 699                                 btrfs_set_ref_objectid(leaf, ref, op->level);
 700                                 btrfs_set_ref_num_refs(leaf, ref, 1);
 701                                 op->del++;
 702                         }
 703
 704                         /*
 705                          * using del to see when its ok to free up the
 706                          * pending_extent_op.  In the case where we insert the
 707                          * last item on the list in order to help do batching
 708                          * we need to not free the extent op until we actually
 709                          * insert the extent_item
 710                          */
 711                         if (op->del == 2) {
 712                                 unlock_extent(&info->extent_ins, op->bytenr,
 713                                               op->bytenr + op->num_bytes - 1,
 714                                               GFP_NOFS);
 715                                 cur = cur->next;
 716                                 list_del_init(&op->list);
 717                                 kfree(op);
 718                                 if (cur != insert_list)
 719                                         op = list_entry(cur,
 720                                                 struct pending_extent_op,
 721                                                 list);
 722                         }
 723                 }
 724                 btrfs_mark_buffer_dirty(leaf);
 725                 btrfs_release_path(extent_root, path);
 726
 727                 /*
 728                  * Ok backref's and items usually go right next to eachother,
 729                  * but if we could only insert 1 item that means that we
 730                  * inserted on the end of a leaf, and we have no idea what may
 731                  * be on the next leaf so we just play it safe.  In order to
 732                  * try and help this case we insert the last thing on our
 733                  * insert list so hopefully it will end up being the last
 734                  * thing on the leaf and everything else will be before it,
 735                  * which will let us insert a whole bunch of items at the same
 736                  * time.
 737                  */
 738                 if (ret == 1 && !last && (i + ret < total)) {
 739                         /*
 740                          * last: where we will pick up the next time around
 741                          * i: our current key to insert, will be total - 1
 742                          * cur: the current op we are screwing with
 743                          * op: duh
 744                          */
 745                         last = i + ret;
 746                         i = total - 1;
 747                         cur = insert_list->prev;
 748                         op = list_entry(cur, struct pending_extent_op, list);
 749                 } else if (last) {
 750                         /*
 751                          * ok we successfully inserted the last item on the
 752                          * list, lets reset everything
 753                          *
 754                          * i: our current key to insert, so where we left off
 755                          *    last time
 756                          * last: done with this
 757                          * cur: the op we are messing with
 758                          * op: duh
 759                          * total: since we inserted the last key, we need to
 760                          *        decrement total so we dont overflow
 761                          */
 762                         i = last;
 763                         last = 0;
 764                         total--;
 765                         if (i < total) {
 766                                 cur = insert_list->next;
 767                                 op = list_entry(cur, struct pending_extent_op,
 768                                                 list);
 769                         }
 770                 } else {
 771                         i += ret;
 772                 }
 773
 774                 cond_resched();
 775         }
 776         ret = 0;
 777         kfree(keys);
 778         kfree(data_size);
 779         return ret;
 780 }
 781
 782 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
 783                                           struct btrfs_root *root,
 784                                           struct btrfs_path *path,
 785                                           u64 bytenr, u64 parent,
 786                                           u64 ref_root, u64 ref_generation,
 787                                           u64 owner_objectid)
 788 {
 789         struct btrfs_key key;
 790         struct extent_buffer *leaf;
 791         struct btrfs_extent_ref *ref;
 792         u32 num_refs;
 793         int ret;
 794
 795         key.objectid = bytenr;
 796         key.type = BTRFS_EXTENT_REF_KEY;
 797         key.offset = parent;
 798
 799         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
 800         if (ret == 0) {
 801                 leaf = path->nodes[0];
 802                 ref = btrfs_item_ptr(leaf, path->slots[0],
 803                                      struct btrfs_extent_ref);
 804                 btrfs_set_ref_root(leaf, ref, ref_root);
 805                 btrfs_set_ref_generation(leaf, ref, ref_generation);
 806                 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
 807                 btrfs_set_ref_num_refs(leaf, ref, 1);
 808         } else if (ret == -EEXIST) {
 809                 u64 existing_owner;
 810                 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
 811                 leaf = path->nodes[0];
 812                 ref = btrfs_item_ptr(leaf, path->slots[0],
 813                                      struct btrfs_extent_ref);
 814                 if (btrfs_ref_root(leaf, ref) != ref_root ||
 815                     btrfs_ref_generation(leaf, ref) != ref_generation) {
 816                         ret = -EIO;
 817                         WARN_ON(1);
 818                         goto out;
 819                 }
 820
 821                 num_refs = btrfs_ref_num_refs(leaf, ref);
 822                 BUG_ON(num_refs == 0);
 823                 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
 824
 825                 existing_owner = btrfs_ref_objectid(leaf, ref);
 826                 if (existing_owner != owner_objectid &&
 827                     existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
 828                         btrfs_set_ref_objectid(leaf, ref,
 829                                         BTRFS_MULTIPLE_OBJECTIDS);
 830                 }
 831                 ret = 0;
 832         } else {
 833                 goto out;
 834         }
 835         btrfs_mark_buffer_dirty(path->nodes[0]);
 836 out:
 837         btrfs_release_path(root, path);
 838         return ret;
 839 }
 840
 841 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
 842                                           struct btrfs_root *root,
 843                                           struct btrfs_path *path)
 844 {
 845         struct extent_buffer *leaf;
 846         struct btrfs_extent_ref *ref;
 847         u32 num_refs;
 848         int ret = 0;
 849
 850         leaf = path->nodes[0];
 851         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 852         num_refs = btrfs_ref_num_refs(leaf, ref);
 853         BUG_ON(num_refs == 0);
 854         num_refs -= 1;
 855         if (num_refs == 0) {
 856                 ret = btrfs_del_item(trans, root, path);
 857         } else {
 858                 btrfs_set_ref_num_refs(leaf, ref, num_refs);
 859                 btrfs_mark_buffer_dirty(leaf);
 860         }
 861         btrfs_release_path(root, path);
 862         return ret;
 863 }
 864
 865 #ifdef BIO_RW_DISCARD
 866 static void btrfs_issue_discard(struct block_device *bdev,
 867                                 u64 start, u64 len)
 868 {
 869         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
 870 }
 871 #endif
 872
 873 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
 874                                 u64 num_bytes)
 875 {
 876 #ifdef BIO_RW_DISCARD
 877         int ret;
 878         u64 map_length = num_bytes;
 879         struct btrfs_multi_bio *multi = NULL;
 880
 881         /* Tell the block device(s) that the sectors can be discarded */
 882         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
 883                               bytenr, &map_length, &multi, 0);
 884         if (!ret) {
 885                 struct btrfs_bio_stripe *stripe = multi->stripes;
 886                 int i;
 887
 888                 if (map_length > num_bytes)
 889                         map_length = num_bytes;
 890
 891                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
 892                         btrfs_issue_discard(stripe->dev->bdev,
 893                                             stripe->physical,
 894                                             map_length);
 895                 }
 896                 kfree(multi);
 897         }
 898
 899         return ret;
 900 #else
 901         return 0;
 902 #endif
 903 }
 904
 905 static noinline int free_extents(struct btrfs_trans_handle *trans,
 906                                  struct btrfs_root *extent_root,
 907                                  struct list_head *del_list)
 908 {
 909         struct btrfs_fs_info *info = extent_root->fs_info;
 910         struct btrfs_path *path;
 911         struct btrfs_key key, found_key;
 912         struct extent_buffer *leaf;
 913         struct list_head *cur;
 914         struct pending_extent_op *op;
 915         struct btrfs_extent_item *ei;
 916         int ret, num_to_del, extent_slot = 0, found_extent = 0;
 917         u32 refs;
 918         u64 bytes_freed = 0;
 919
 920         path = btrfs_alloc_path();
 921         if (!path)
 922                 return -ENOMEM;
 923         path->reada = 1;
 924
 925 search:
 926         /* search for the backref for the current ref we want to delete */
 927         cur = del_list->next;
 928         op = list_entry(cur, struct pending_extent_op, list);
 929         ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
 930                                     op->orig_parent,
 931                                     extent_root->root_key.objectid,
 932                                     op->orig_generation, op->level, 1);
 933         if (ret) {
 934                 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
 935                        "root %llu gen %llu owner %u\n",
 936                        (unsigned long long)op->bytenr,
 937                        (unsigned long long)extent_root->root_key.objectid,
 938                        (unsigned long long)op->orig_generation, op->level);
 939                 btrfs_print_leaf(extent_root, path->nodes[0]);
 940                 WARN_ON(1);
 941                 goto out;
 942         }
 943
 944         extent_slot = path->slots[0];
 945         num_to_del = 1;
 946         found_extent = 0;
 947
 948         /*
 949          * if we aren't the first item on the leaf we can move back one and see
 950          * if our ref is right next to our extent item
 951          */
 952         if (likely(extent_slot)) {
 953                 extent_slot--;
 954                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 955                                       extent_slot);
 956                 if (found_key.objectid == op->bytenr &&
 957                     found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 958                     found_key.offset == op->num_bytes) {
 959                         num_to_del++;
 960                         found_extent = 1;
 961                 }
 962         }
 963
 964         /*
 965          * if we didn't find the extent we need to delete the backref and then
 966          * search for the extent item key so we can update its ref count
 967          */
 968         if (!found_extent) {
 969                 key.objectid = op->bytenr;
 970                 key.type = BTRFS_EXTENT_ITEM_KEY;
 971                 key.offset = op->num_bytes;
 972
 973                 ret = remove_extent_backref(trans, extent_root, path);
 974                 BUG_ON(ret);
 975                 btrfs_release_path(extent_root, path);
 976                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
 977                 BUG_ON(ret);
 978                 extent_slot = path->slots[0];
 979         }
 980
 981         /* this is where we update the ref count for the extent */
 982         leaf = path->nodes[0];
 983         ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
 984         refs = btrfs_extent_refs(leaf, ei);
 985         BUG_ON(refs == 0);
 986         refs--;
 987         btrfs_set_extent_refs(leaf, ei, refs);
 988
 989         btrfs_mark_buffer_dirty(leaf);
 990
 991         /*
 992          * This extent needs deleting.  The reason cur_slot is extent_slot +
 993          * num_to_del is because extent_slot points to the slot where the extent
 994          * is, and if the backref was not right next to the extent we will be
 995          * deleting at least 1 item, and will want to start searching at the
 996          * slot directly next to extent_slot.  However if we did find the
 997          * backref next to the extent item them we will be deleting at least 2
 998          * items and will want to start searching directly after the ref slot
 999          */
1000         if (!refs) {
1001                 struct list_head *pos, *n, *end;
1002                 int cur_slot = extent_slot+num_to_del;
1003                 u64 super_used;
1004                 u64 root_used;
1005
1006                 path->slots[0] = extent_slot;
1007                 bytes_freed = op->num_bytes;
1008
1009                 mutex_lock(&info->pinned_mutex);
1010                 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1011                                      op->num_bytes, op->level >=
1012                                      BTRFS_FIRST_FREE_OBJECTID);
1013                 mutex_unlock(&info->pinned_mutex);
1014                 BUG_ON(ret < 0);
1015                 op->del = ret;
1016
1017                 /*
1018                  * we need to see if we can delete multiple things at once, so
1019                  * start looping through the list of extents we are wanting to
1020                  * delete and see if their extent/backref's are right next to
1021                  * eachother and the extents only have 1 ref
1022                  */
1023                 for (pos = cur->next; pos != del_list; pos = pos->next) {
1024                         struct pending_extent_op *tmp;
1025
1026                         tmp = list_entry(pos, struct pending_extent_op, list);
1027
1028                         /* we only want to delete extent+ref at this stage */
1029                         if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1030                                 break;
1031
1032                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1033                         if (found_key.objectid != tmp->bytenr ||
1034                             found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1035                             found_key.offset != tmp->num_bytes)
1036                                 break;
1037
1038                         /* check to make sure this extent only has one ref */
1039                         ei = btrfs_item_ptr(leaf, cur_slot,
1040                                             struct btrfs_extent_item);
1041                         if (btrfs_extent_refs(leaf, ei) != 1)
1042                                 break;
1043
1044                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1045                         if (found_key.objectid != tmp->bytenr ||
1046                             found_key.type != BTRFS_EXTENT_REF_KEY ||
1047                             found_key.offset != tmp->orig_parent)
1048                                 break;
1049
1050                         /*
1051                          * the ref is right next to the extent, we can set the
1052                          * ref count to 0 since we will delete them both now
1053                          */
1054                         btrfs_set_extent_refs(leaf, ei, 0);
1055
1056                         /* pin down the bytes for this extent */
1057                         mutex_lock(&info->pinned_mutex);
1058                         ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1059                                              tmp->num_bytes, tmp->level >=
1060                                              BTRFS_FIRST_FREE_OBJECTID);
1061                         mutex_unlock(&info->pinned_mutex);
1062                         BUG_ON(ret < 0);
1063
1064                         /*
1065                          * use the del field to tell if we need to go ahead and
1066                          * free up the extent when we delete the item or not.
1067                          */
1068                         tmp->del = ret;
1069                         bytes_freed += tmp->num_bytes;
1070
1071                         num_to_del += 2;
1072                         cur_slot += 2;
1073                 }
1074                 end = pos;
1075
1076                 /* update the free space counters */
1077                 spin_lock(&info->delalloc_lock);
1078                 super_used = btrfs_super_bytes_used(&info->super_copy);
1079                 btrfs_set_super_bytes_used(&info->super_copy,
1080                                            super_used - bytes_freed);
1081
1082                 root_used = btrfs_root_used(&extent_root->root_item);
1083                 btrfs_set_root_used(&extent_root->root_item,
1084                                     root_used - bytes_freed);
1085                 spin_unlock(&info->delalloc_lock);
1086
1087                 /* delete the items */
1088                 ret = btrfs_del_items(trans, extent_root, path,
1089                                       path->slots[0], num_to_del);
1090                 BUG_ON(ret);
1091
1092                 /*
1093                  * loop through the extents we deleted and do the cleanup work
1094                  * on them
1095                  */
1096                 for (pos = cur, n = pos->next; pos != end;
1097                      pos = n, n = pos->next) {
1098                         struct pending_extent_op *tmp;
1099                         tmp = list_entry(pos, struct pending_extent_op, list);
1100
1101                         /*
1102                          * remember tmp->del tells us wether or not we pinned
1103                          * down the extent
1104                          */
1105                         ret = update_block_group(trans, extent_root,
1106                                                  tmp->bytenr, tmp->num_bytes, 0,
1107                                                  tmp->del);
1108                         BUG_ON(ret);
1109
1110                         list_del_init(&tmp->list);
1111                         unlock_extent(&info->extent_ins, tmp->bytenr,
1112                                       tmp->bytenr + tmp->num_bytes - 1,
1113                                       GFP_NOFS);
1114                         kfree(tmp);
1115                 }
1116         } else if (refs && found_extent) {
1117                 /*
1118                  * the ref and extent were right next to eachother, but the
1119                  * extent still has a ref, so just free the backref and keep
1120                  * going
1121                  */
1122                 ret = remove_extent_backref(trans, extent_root, path);
1123                 BUG_ON(ret);
1124
1125                 list_del_init(&op->list);
1126                 unlock_extent(&info->extent_ins, op->bytenr,
1127                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1128                 kfree(op);
1129         } else {
1130                 /*
1131                  * the extent has multiple refs and the backref we were looking
1132                  * for was not right next to it, so just unlock and go next,
1133                  * we're good to go
1134                  */
1135                 list_del_init(&op->list);
1136                 unlock_extent(&info->extent_ins, op->bytenr,
1137                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1138                 kfree(op);
1139         }
1140
1141         btrfs_release_path(extent_root, path);
1142         if (!list_empty(del_list))
1143                 goto search;
1144
1145 out:
1146         btrfs_free_path(path);
1147         return ret;
1148 }
1149
1150 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1151                                      struct btrfs_root *root, u64 bytenr,
1152                                      u64 orig_parent, u64 parent,
1153                                      u64 orig_root, u64 ref_root,
1154                                      u64 orig_generation, u64 ref_generation,
1155                                      u64 owner_objectid)
1156 {
1157         int ret;
1158         struct btrfs_root *extent_root = root->fs_info->extent_root;
1159         struct btrfs_path *path;
1160
1161         if (root == root->fs_info->extent_root) {
1162                 struct pending_extent_op *extent_op;
1163                 u64 num_bytes;
1164
1165                 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1166                 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1167                 mutex_lock(&root->fs_info->extent_ins_mutex);
1168                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1169                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1170                         u64 priv;
1171                         ret = get_state_private(&root->fs_info->extent_ins,
1172                                                 bytenr, &priv);
1173                         BUG_ON(ret);
1174                         extent_op = (struct pending_extent_op *)
1175                                                         (unsigned long)priv;
1176                         BUG_ON(extent_op->parent != orig_parent);
1177                         BUG_ON(extent_op->generation != orig_generation);
1178
1179                         extent_op->parent = parent;
1180                         extent_op->generation = ref_generation;
1181                 } else {
1182                         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1183                         BUG_ON(!extent_op);
1184
1185                         extent_op->type = PENDING_BACKREF_UPDATE;
1186                         extent_op->bytenr = bytenr;
1187                         extent_op->num_bytes = num_bytes;
1188                         extent_op->parent = parent;
1189                         extent_op->orig_parent = orig_parent;
1190                         extent_op->generation = ref_generation;
1191                         extent_op->orig_generation = orig_generation;
1192                         extent_op->level = (int)owner_objectid;
1193                         INIT_LIST_HEAD(&extent_op->list);
1194                         extent_op->del = 0;
1195
1196                         set_extent_bits(&root->fs_info->extent_ins,
1197                                         bytenr, bytenr + num_bytes - 1,
1198                                         EXTENT_WRITEBACK, GFP_NOFS);
1199                         set_state_private(&root->fs_info->extent_ins,
1200                                           bytenr, (unsigned long)extent_op);
1201                 }
1202                 mutex_unlock(&root->fs_info->extent_ins_mutex);
1203                 return 0;
1204         }
1205
1206         path = btrfs_alloc_path();
1207         if (!path)
1208                 return -ENOMEM;
1209         ret = lookup_extent_backref(trans, extent_root, path,
1210                                     bytenr, orig_parent, orig_root,
1211                                     orig_generation, owner_objectid, 1);
1212         if (ret)
1213                 goto out;
1214         ret = remove_extent_backref(trans, extent_root, path);
1215         if (ret)
1216                 goto out;
1217         ret = insert_extent_backref(trans, extent_root, path, bytenr,
1218                                     parent, ref_root, ref_generation,
1219                                     owner_objectid);
1220         BUG_ON(ret);
1221         finish_current_insert(trans, extent_root, 0);
1222         del_pending_extents(trans, extent_root, 0);
1223 out:
1224         btrfs_free_path(path);
1225         return ret;
1226 }
1227
1228 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1229                             struct btrfs_root *root, u64 bytenr,
1230                             u64 orig_parent, u64 parent,
1231                             u64 ref_root, u64 ref_generation,
1232                             u64 owner_objectid)
1233 {
1234         int ret;
1235         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1236             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1237                 return 0;
1238         ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1239                                         parent, ref_root, ref_root,
1240                                         ref_generation, ref_generation,
1241                                         owner_objectid);
1242         return ret;
1243 }
1244
1245 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1246                                   struct btrfs_root *root, u64 bytenr,
1247                                   u64 orig_parent, u64 parent,
1248                                   u64 orig_root, u64 ref_root,
1249                                   u64 orig_generation, u64 ref_generation,
1250                                   u64 owner_objectid)
1251 {
1252         struct btrfs_path *path;
1253         int ret;
1254         struct btrfs_key key;
1255         struct extent_buffer *l;
1256         struct btrfs_extent_item *item;
1257         u32 refs;
1258
1259         path = btrfs_alloc_path();
1260         if (!path)
1261                 return -ENOMEM;
1262
1263         path->reada = 1;
1264         key.objectid = bytenr;
1265         key.type = BTRFS_EXTENT_ITEM_KEY;
1266         key.offset = (u64)-1;
1267
1268         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1269                                 0, 1);
1270         if (ret < 0)
1271                 return ret;
1272         BUG_ON(ret == 0 || path->slots[0] == 0);
1273
1274         path->slots[0]--;
1275         l = path->nodes[0];
1276
1277         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1278         if (key.objectid != bytenr) {
1279                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1280                 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
1281                        (unsigned long long)bytenr,
1282                        (unsigned long long)key.objectid);
1283                 BUG();
1284         }
1285         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1286
1287         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1288         refs = btrfs_extent_refs(l, item);
1289         btrfs_set_extent_refs(l, item, refs + 1);
1290         btrfs_mark_buffer_dirty(path->nodes[0]);
1291
1292         btrfs_release_path(root->fs_info->extent_root, path);
1293
1294         path->reada = 1;
1295         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1296                                     path, bytenr, parent,
1297                                     ref_root, ref_generation,
1298                                     owner_objectid);
1299         BUG_ON(ret);
1300         finish_current_insert(trans, root->fs_info->extent_root, 0);
1301         del_pending_extents(trans, root->fs_info->extent_root, 0);
1302
1303         btrfs_free_path(path);
1304         return 0;
1305 }
1306
1307 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1308                          struct btrfs_root *root,
1309                          u64 bytenr, u64 num_bytes, u64 parent,
1310                          u64 ref_root, u64 ref_generation,
1311                          u64 owner_objectid)
1312 {
1313         int ret;
1314         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1315             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1316                 return 0;
1317         ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1318                                      0, ref_root, 0, ref_generation,
1319                                      owner_objectid);
1320         return ret;
1321 }
1322
1323 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1324                          struct btrfs_root *root)
1325 {
1326         finish_current_insert(trans, root->fs_info->extent_root, 1);
1327         del_pending_extents(trans, root->fs_info->extent_root, 1);
1328         return 0;
1329 }
1330
1331 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1332                             struct btrfs_root *root, u64 bytenr,
1333                             u64 num_bytes, u32 *refs)
1334 {
1335         struct btrfs_path *path;
1336         int ret;
1337         struct btrfs_key key;
1338         struct extent_buffer *l;
1339         struct btrfs_extent_item *item;
1340
1341         WARN_ON(num_bytes < root->sectorsize);
1342         path = btrfs_alloc_path();
1343         path->reada = 1;
1344         key.objectid = bytenr;
1345         key.offset = num_bytes;
1346         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1347         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1348                                 0, 0);
1349         if (ret < 0)
1350                 goto out;
1351         if (ret != 0) {
1352                 btrfs_print_leaf(root, path->nodes[0]);
1353                 printk(KERN_INFO "btrfs failed to find block number %llu\n",
1354                        (unsigned long long)bytenr);
1355                 BUG();
1356         }
1357         l = path->nodes[0];
1358         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1359         *refs = btrfs_extent_refs(l, item);
1360 out:
1361         btrfs_free_path(path);
1362         return 0;
1363 }
1364
1365 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1366                           struct btrfs_root *root, u64 objectid, u64 bytenr)
1367 {
1368         struct btrfs_root *extent_root = root->fs_info->extent_root;
1369         struct btrfs_path *path;
1370         struct extent_buffer *leaf;
1371         struct btrfs_extent_ref *ref_item;
1372         struct btrfs_key key;
1373         struct btrfs_key found_key;
1374         u64 ref_root;
1375         u64 last_snapshot;
1376         u32 nritems;
1377         int ret;
1378
1379         key.objectid = bytenr;
1380         key.offset = (u64)-1;
1381         key.type = BTRFS_EXTENT_ITEM_KEY;
1382
1383         path = btrfs_alloc_path();
1384         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1385         if (ret < 0)
1386                 goto out;
1387         BUG_ON(ret == 0);
1388
1389         ret = -ENOENT;
1390         if (path->slots[0] == 0)
1391                 goto out;
1392
1393         path->slots[0]--;
1394         leaf = path->nodes[0];
1395         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1396
1397         if (found_key.objectid != bytenr ||
1398             found_key.type != BTRFS_EXTENT_ITEM_KEY)
1399                 goto out;
1400
1401         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1402         while (1) {
1403                 leaf = path->nodes[0];
1404                 nritems = btrfs_header_nritems(leaf);
1405                 if (path->slots[0] >= nritems) {
1406                         ret = btrfs_next_leaf(extent_root, path);
1407                         if (ret < 0)
1408                                 goto out;
1409                         if (ret == 0)
1410                                 continue;
1411                         break;
1412                 }
1413                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1414                 if (found_key.objectid != bytenr)
1415                         break;
1416
1417                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1418                         path->slots[0]++;
1419                         continue;
1420                 }
1421
1422                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1423                                           struct btrfs_extent_ref);
1424                 ref_root = btrfs_ref_root(leaf, ref_item);
1425                 if ((ref_root != root->root_key.objectid &&
1426                      ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1427                      objectid != btrfs_ref_objectid(leaf, ref_item)) {
1428                         ret = 1;
1429                         goto out;
1430                 }
1431                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1432                         ret = 1;
1433                         goto out;
1434                 }
1435
1436                 path->slots[0]++;
1437         }
1438         ret = 0;
1439 out:
1440         btrfs_free_path(path);
1441         return ret;
1442 }
1443
1444 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1445                     struct extent_buffer *buf, u32 nr_extents)
1446 {
1447         struct btrfs_key key;
1448         struct btrfs_file_extent_item *fi;
1449         u64 root_gen;
1450         u32 nritems;
1451         int i;
1452         int level;
1453         int ret = 0;
1454         int shared = 0;
1455
1456         if (!root->ref_cows)
1457                 return 0;
1458
1459         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1460                 shared = 0;
1461                 root_gen = root->root_key.offset;
1462         } else {
1463                 shared = 1;
1464                 root_gen = trans->transid - 1;
1465         }
1466
1467         level = btrfs_header_level(buf);
1468         nritems = btrfs_header_nritems(buf);
1469
1470         if (level == 0) {
1471                 struct btrfs_leaf_ref *ref;
1472                 struct btrfs_extent_info *info;
1473
1474                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1475                 if (!ref) {
1476                         ret = -ENOMEM;
1477                         goto out;
1478                 }
1479
1480                 ref->root_gen = root_gen;
1481                 ref->bytenr = buf->start;
1482                 ref->owner = btrfs_header_owner(buf);
1483                 ref->generation = btrfs_header_generation(buf);
1484                 ref->nritems = nr_extents;
1485                 info = ref->extents;
1486
1487                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1488                         u64 disk_bytenr;
1489                         btrfs_item_key_to_cpu(buf, &key, i);
1490                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1491                                 continue;
1492                         fi = btrfs_item_ptr(buf, i,
1493                                             struct btrfs_file_extent_item);
1494                         if (btrfs_file_extent_type(buf, fi) ==
1495                             BTRFS_FILE_EXTENT_INLINE)
1496                                 continue;
1497                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1498                         if (disk_bytenr == 0)
1499                                 continue;
1500
1501                         info->bytenr = disk_bytenr;
1502                         info->num_bytes =
1503                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1504                         info->objectid = key.objectid;
1505                         info->offset = key.offset;
1506                         info++;
1507                 }
1508
1509                 ret = btrfs_add_leaf_ref(root, ref, shared);
1510                 if (ret == -EEXIST && shared) {
1511                         struct btrfs_leaf_ref *old;
1512                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1513                         BUG_ON(!old);
1514                         btrfs_remove_leaf_ref(root, old);
1515                         btrfs_free_leaf_ref(root, old);
1516                         ret = btrfs_add_leaf_ref(root, ref, shared);
1517                 }
1518                 WARN_ON(ret);
1519                 btrfs_free_leaf_ref(root, ref);
1520         }
1521 out:
1522         return ret;
1523 }
1524
1525 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1526                   struct extent_buffer *orig_buf, struct extent_buffer *buf,
1527                   u32 *nr_extents)
1528 {
1529         u64 bytenr;
1530         u64 ref_root;
1531         u64 orig_root;
1532         u64 ref_generation;
1533         u64 orig_generation;
1534         u32 nritems;
1535         u32 nr_file_extents = 0;
1536         struct btrfs_key key;
1537         struct btrfs_file_extent_item *fi;
1538         int i;
1539         int level;
1540         int ret = 0;
1541         int faili = 0;
1542         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1543                             u64, u64, u64, u64, u64, u64, u64, u64);
1544
1545         ref_root = btrfs_header_owner(buf);
1546         ref_generation = btrfs_header_generation(buf);
1547         orig_root = btrfs_header_owner(orig_buf);
1548         orig_generation = btrfs_header_generation(orig_buf);
1549
1550         nritems = btrfs_header_nritems(buf);
1551         level = btrfs_header_level(buf);
1552
1553         if (root->ref_cows) {
1554                 process_func = __btrfs_inc_extent_ref;
1555         } else {
1556                 if (level == 0 &&
1557                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1558                         goto out;
1559                 if (level != 0 &&
1560                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1561                         goto out;
1562                 process_func = __btrfs_update_extent_ref;
1563         }
1564
1565         for (i = 0; i < nritems; i++) {
1566                 cond_resched();
1567                 if (level == 0) {
1568                         btrfs_item_key_to_cpu(buf, &key, i);
1569                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1570                                 continue;
1571                         fi = btrfs_item_ptr(buf, i,
1572                                             struct btrfs_file_extent_item);
1573                         if (btrfs_file_extent_type(buf, fi) ==
1574                             BTRFS_FILE_EXTENT_INLINE)
1575                                 continue;
1576                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1577                         if (bytenr == 0)
1578                                 continue;
1579
1580                         nr_file_extents++;
1581
1582                         ret = process_func(trans, root, bytenr,
1583                                            orig_buf->start, buf->start,
1584                                            orig_root, ref_root,
1585                                            orig_generation, ref_generation,
1586                                            key.objectid);
1587
1588                         if (ret) {
1589                                 faili = i;
1590                                 WARN_ON(1);
1591                                 goto fail;
1592                         }
1593                 } else {
1594                         bytenr = btrfs_node_blockptr(buf, i);
1595                         ret = process_func(trans, root, bytenr,
1596                                            orig_buf->start, buf->start,
1597                                            orig_root, ref_root,
1598                                            orig_generation, ref_generation,
1599                                            level - 1);
1600                         if (ret) {
1601                                 faili = i;
1602                                 WARN_ON(1);
1603                                 goto fail;
1604                         }
1605                 }
1606         }
1607 out:
1608         if (nr_extents) {
1609                 if (level == 0)
1610                         *nr_extents = nr_file_extents;
1611                 else
1612                         *nr_extents = nritems;
1613         }
1614         return 0;
1615 fail:
1616         WARN_ON(1);
1617         return ret;
1618 }
1619
1620 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1621                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1622                      struct extent_buffer *buf, int start_slot, int nr)
1623
1624 {
1625         u64 bytenr;
1626         u64 ref_root;
1627         u64 orig_root;
1628         u64 ref_generation;
1629         u64 orig_generation;
1630         struct btrfs_key key;
1631         struct btrfs_file_extent_item *fi;
1632         int i;
1633         int ret;
1634         int slot;
1635         int level;
1636
1637         BUG_ON(start_slot < 0);
1638         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1639
1640         ref_root = btrfs_header_owner(buf);
1641         ref_generation = btrfs_header_generation(buf);
1642         orig_root = btrfs_header_owner(orig_buf);
1643         orig_generation = btrfs_header_generation(orig_buf);
1644         level = btrfs_header_level(buf);
1645
1646         if (!root->ref_cows) {
1647                 if (level == 0 &&
1648                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1649                         return 0;
1650                 if (level != 0 &&
1651                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1652                         return 0;
1653         }
1654
1655         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1656                 cond_resched();
1657                 if (level == 0) {
1658                         btrfs_item_key_to_cpu(buf, &key, slot);
1659                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1660                                 continue;
1661                         fi = btrfs_item_ptr(buf, slot,
1662                                             struct btrfs_file_extent_item);
1663                         if (btrfs_file_extent_type(buf, fi) ==
1664                             BTRFS_FILE_EXTENT_INLINE)
1665                                 continue;
1666                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1667                         if (bytenr == 0)
1668                                 continue;
1669                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1670                                             orig_buf->start, buf->start,
1671                                             orig_root, ref_root,
1672                                             orig_generation, ref_generation,
1673                                             key.objectid);
1674                         if (ret)
1675                                 goto fail;
1676                 } else {
1677                         bytenr = btrfs_node_blockptr(buf, slot);
1678                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1679                                             orig_buf->start, buf->start,
1680                                             orig_root, ref_root,
1681                                             orig_generation, ref_generation,
1682                                             level - 1);
1683                         if (ret)
1684                                 goto fail;
1685                 }
1686         }
1687         return 0;
1688 fail:
1689         WARN_ON(1);
1690         return -1;
1691 }
1692
1693 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1694                                  struct btrfs_root *root,
1695                                  struct btrfs_path *path,
1696                                  struct btrfs_block_group_cache *cache)
1697 {
1698         int ret;
1699         int pending_ret;
1700         struct btrfs_root *extent_root = root->fs_info->extent_root;
1701         unsigned long bi;
1702         struct extent_buffer *leaf;
1703
1704         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1705         if (ret < 0)
1706                 goto fail;
1707         BUG_ON(ret);
1708
1709         leaf = path->nodes[0];
1710         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1711         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1712         btrfs_mark_buffer_dirty(leaf);
1713         btrfs_release_path(extent_root, path);
1714 fail:
1715         finish_current_insert(trans, extent_root, 0);
1716         pending_ret = del_pending_extents(trans, extent_root, 0);
1717         if (ret)
1718                 return ret;
1719         if (pending_ret)
1720                 return pending_ret;
1721         return 0;
1722
1723 }
1724
1725 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1726                                    struct btrfs_root *root)
1727 {
1728         struct btrfs_block_group_cache *cache, *entry;
1729         struct rb_node *n;
1730         int err = 0;
1731         int werr = 0;
1732         struct btrfs_path *path;
1733         u64 last = 0;
1734
1735         path = btrfs_alloc_path();
1736         if (!path)
1737                 return -ENOMEM;
1738
1739         while (1) {
1740                 cache = NULL;
1741                 spin_lock(&root->fs_info->block_group_cache_lock);
1742                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1743                      n; n = rb_next(n)) {
1744                         entry = rb_entry(n, struct btrfs_block_group_cache,
1745                                          cache_node);
1746                         if (entry->dirty) {
1747                                 cache = entry;
1748                                 break;
1749                         }
1750                 }
1751                 spin_unlock(&root->fs_info->block_group_cache_lock);
1752
1753                 if (!cache)
1754                         break;
1755
1756                 cache->dirty = 0;
1757                 last += cache->key.offset;
1758
1759                 err = write_one_cache_group(trans, root,
1760                                             path, cache);
1761                 /*
1762                  * if we fail to write the cache group, we want
1763                  * to keep it marked dirty in hopes that a later
1764                  * write will work
1765                  */
1766                 if (err) {
1767                         werr = err;
1768                         continue;
1769                 }
1770         }
1771         btrfs_free_path(path);
1772         return werr;
1773 }
1774
1775 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1776 {
1777         struct btrfs_block_group_cache *block_group;
1778         int readonly = 0;
1779
1780         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1781         if (!block_group || block_group->ro)
1782                 readonly = 1;
1783         if (block_group)
1784                 put_block_group(block_group);
1785         return readonly;
1786 }
1787
1788 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1789                              u64 total_bytes, u64 bytes_used,
1790                              struct btrfs_space_info **space_info)
1791 {
1792         struct btrfs_space_info *found;
1793
1794         found = __find_space_info(info, flags);
1795         if (found) {
1796                 spin_lock(&found->lock);
1797                 found->total_bytes += total_bytes;
1798                 found->bytes_used += bytes_used;
1799                 found->full = 0;
1800                 spin_unlock(&found->lock);
1801                 *space_info = found;
1802                 return 0;
1803         }
1804         found = kzalloc(sizeof(*found), GFP_NOFS);
1805         if (!found)
1806                 return -ENOMEM;
1807
1808         list_add(&found->list, &info->space_info);
1809         INIT_LIST_HEAD(&found->block_groups);
1810         init_rwsem(&found->groups_sem);
1811         spin_lock_init(&found->lock);
1812         found->flags = flags;
1813         found->total_bytes = total_bytes;
1814         found->bytes_used = bytes_used;
1815         found->bytes_pinned = 0;
1816         found->bytes_reserved = 0;
1817         found->bytes_readonly = 0;
1818         found->full = 0;
1819         found->force_alloc = 0;
1820         *space_info = found;
1821         return 0;
1822 }
1823
1824 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1825 {
1826         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1827                                    BTRFS_BLOCK_GROUP_RAID1 |
1828                                    BTRFS_BLOCK_GROUP_RAID10 |
1829                                    BTRFS_BLOCK_GROUP_DUP);
1830         if (extra_flags) {
1831                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1832                         fs_info->avail_data_alloc_bits |= extra_flags;
1833                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1834                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1835                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1836                         fs_info->avail_system_alloc_bits |= extra_flags;
1837         }
1838 }
1839
1840 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1841 {
1842         spin_lock(&cache->space_info->lock);
1843         spin_lock(&cache->lock);
1844         if (!cache->ro) {
1845                 cache->space_info->bytes_readonly += cache->key.offset -
1846                                         btrfs_block_group_used(&cache->item);
1847                 cache->ro = 1;
1848         }
1849         spin_unlock(&cache->lock);
1850         spin_unlock(&cache->space_info->lock);
1851 }
1852
1853 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1854 {
1855         u64 num_devices = root->fs_info->fs_devices->rw_devices;
1856
1857         if (num_devices == 1)
1858                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1859         if (num_devices < 4)
1860                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1861
1862         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1863             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1864                       BTRFS_BLOCK_GROUP_RAID10))) {
1865                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1866         }
1867
1868         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1869             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1870                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1871         }
1872
1873         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1874             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1875              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1876              (flags & BTRFS_BLOCK_GROUP_DUP)))
1877                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1878         return flags;
1879 }
1880
1881 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1882                           struct btrfs_root *extent_root, u64 alloc_bytes,
1883                           u64 flags, int force)
1884 {
1885         struct btrfs_space_info *space_info;
1886         u64 thresh;
1887         int ret = 0;
1888
1889         mutex_lock(&extent_root->fs_info->chunk_mutex);
1890
1891         flags = btrfs_reduce_alloc_profile(extent_root, flags);
1892
1893         space_info = __find_space_info(extent_root->fs_info, flags);
1894         if (!space_info) {
1895                 ret = update_space_info(extent_root->fs_info, flags,
1896                                         0, 0, &space_info);
1897                 BUG_ON(ret);
1898         }
1899         BUG_ON(!space_info);
1900
1901         spin_lock(&space_info->lock);
1902         if (space_info->force_alloc) {
1903                 force = 1;
1904                 space_info->force_alloc = 0;
1905         }
1906         if (space_info->full) {
1907                 spin_unlock(&space_info->lock);
1908                 goto out;
1909         }
1910
1911         thresh = space_info->total_bytes - space_info->bytes_readonly;
1912         thresh = div_factor(thresh, 6);
1913         if (!force &&
1914            (space_info->bytes_used + space_info->bytes_pinned +
1915             space_info->bytes_reserved + alloc_bytes) < thresh) {
1916                 spin_unlock(&space_info->lock);
1917                 goto out;
1918         }
1919         spin_unlock(&space_info->lock);
1920
1921         ret = btrfs_alloc_chunk(trans, extent_root, flags);
1922         if (ret)
1923                 space_info->full = 1;
1924 out:
1925         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1926         return ret;
1927 }
1928
1929 static int update_block_group(struct btrfs_trans_handle *trans,
1930                               struct btrfs_root *root,
1931                               u64 bytenr, u64 num_bytes, int alloc,
1932                               int mark_free)
1933 {
1934         struct btrfs_block_group_cache *cache;
1935         struct btrfs_fs_info *info = root->fs_info;
1936         u64 total = num_bytes;
1937         u64 old_val;
1938         u64 byte_in_group;
1939
1940         while (total) {
1941                 cache = btrfs_lookup_block_group(info, bytenr);
1942                 if (!cache)
1943                         return -1;
1944                 byte_in_group = bytenr - cache->key.objectid;
1945                 WARN_ON(byte_in_group > cache->key.offset);
1946
1947                 spin_lock(&cache->space_info->lock);
1948                 spin_lock(&cache->lock);
1949                 cache->dirty = 1;
1950                 old_val = btrfs_block_group_used(&cache->item);
1951                 num_bytes = min(total, cache->key.offset - byte_in_group);
1952                 if (alloc) {
1953                         old_val += num_bytes;
1954                         cache->space_info->bytes_used += num_bytes;
1955                         if (cache->ro)
1956                                 cache->space_info->bytes_readonly -= num_bytes;
1957                         btrfs_set_block_group_used(&cache->item, old_val);
1958                         spin_unlock(&cache->lock);
1959                         spin_unlock(&cache->space_info->lock);
1960                 } else {
1961                         old_val -= num_bytes;
1962                         cache->space_info->bytes_used -= num_bytes;
1963                         if (cache->ro)
1964                                 cache->space_info->bytes_readonly += num_bytes;
1965                         btrfs_set_block_group_used(&cache->item, old_val);
1966                         spin_unlock(&cache->lock);
1967                         spin_unlock(&cache->space_info->lock);
1968                         if (mark_free) {
1969                                 int ret;
1970
1971                                 ret = btrfs_discard_extent(root, bytenr,
1972                                                            num_bytes);
1973                                 WARN_ON(ret);
1974
1975                                 ret = btrfs_add_free_space(cache, bytenr,
1976                                                            num_bytes);
1977                                 WARN_ON(ret);
1978                         }
1979                 }
1980                 put_block_group(cache);
1981                 total -= num_bytes;
1982                 bytenr += num_bytes;
1983         }
1984         return 0;
1985 }
1986
1987 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1988 {
1989         struct btrfs_block_group_cache *cache;
1990         u64 bytenr;
1991
1992         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1993         if (!cache)
1994                 return 0;
1995
1996         bytenr = cache->key.objectid;
1997         put_block_group(cache);
1998
1999         return bytenr;
2000 }
2001
2002 int btrfs_update_pinned_extents(struct btrfs_root *root,
2003                                 u64 bytenr, u64 num, int pin)
2004 {
2005         u64 len;
2006         struct btrfs_block_group_cache *cache;
2007         struct btrfs_fs_info *fs_info = root->fs_info;
2008
2009         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2010         if (pin) {
2011                 set_extent_dirty(&fs_info->pinned_extents,
2012                                 bytenr, bytenr + num - 1, GFP_NOFS);
2013         } else {
2014                 clear_extent_dirty(&fs_info->pinned_extents,
2015                                 bytenr, bytenr + num - 1, GFP_NOFS);
2016         }
2017         while (num > 0) {
2018                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2019                 BUG_ON(!cache);
2020                 len = min(num, cache->key.offset -
2021                           (bytenr - cache->key.objectid));
2022                 if (pin) {
2023                         spin_lock(&cache->space_info->lock);
2024                         spin_lock(&cache->lock);
2025                         cache->pinned += len;
2026                         cache->space_info->bytes_pinned += len;
2027                         spin_unlock(&cache->lock);
2028                         spin_unlock(&cache->space_info->lock);
2029                         fs_info->total_pinned += len;
2030                 } else {
2031                         spin_lock(&cache->space_info->lock);
2032                         spin_lock(&cache->lock);
2033                         cache->pinned -= len;
2034                         cache->space_info->bytes_pinned -= len;
2035                         spin_unlock(&cache->lock);
2036                         spin_unlock(&cache->space_info->lock);
2037                         fs_info->total_pinned -= len;
2038                         if (cache->cached)
2039                                 btrfs_add_free_space(cache, bytenr, len);
2040                 }
2041                 put_block_group(cache);
2042                 bytenr += len;
2043                 num -= len;
2044         }
2045         return 0;
2046 }
2047
2048 static int update_reserved_extents(struct btrfs_root *root,
2049                                    u64 bytenr, u64 num, int reserve)
2050 {
2051         u64 len;
2052         struct btrfs_block_group_cache *cache;
2053         struct btrfs_fs_info *fs_info = root->fs_info;
2054
2055         while (num > 0) {
2056                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2057                 BUG_ON(!cache);
2058                 len = min(num, cache->key.offset -
2059                           (bytenr - cache->key.objectid));
2060
2061                 spin_lock(&cache->space_info->lock);
2062                 spin_lock(&cache->lock);
2063                 if (reserve) {
2064                         cache->reserved += len;
2065                         cache->space_info->bytes_reserved += len;
2066                 } else {
2067                         cache->reserved -= len;
2068                         cache->space_info->bytes_reserved -= len;
2069                 }
2070                 spin_unlock(&cache->lock);
2071                 spin_unlock(&cache->space_info->lock);
2072                 put_block_group(cache);
2073                 bytenr += len;
2074                 num -= len;
2075         }
2076         return 0;
2077 }
2078
2079 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2080 {
2081         u64 last = 0;
2082         u64 start;
2083         u64 end;
2084         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2085         int ret;
2086
2087         mutex_lock(&root->fs_info->pinned_mutex);
2088         while (1) {
2089                 ret = find_first_extent_bit(pinned_extents, last,
2090                                             &start, &end, EXTENT_DIRTY);
2091                 if (ret)
2092                         break;
2093                 set_extent_dirty(copy, start, end, GFP_NOFS);
2094                 last = end + 1;
2095         }
2096         mutex_unlock(&root->fs_info->pinned_mutex);
2097         return 0;
2098 }
2099
2100 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2101                                struct btrfs_root *root,
2102                                struct extent_io_tree *unpin)
2103 {
2104         u64 start;
2105         u64 end;
2106         int ret;
2107
2108         mutex_lock(&root->fs_info->pinned_mutex);
2109         while (1) {
2110                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2111                                             EXTENT_DIRTY);
2112                 if (ret)
2113                         break;
2114
2115                 ret = btrfs_discard_extent(root, start, end + 1 - start);
2116
2117                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2118                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2119
2120                 if (need_resched()) {
2121                         mutex_unlock(&root->fs_info->pinned_mutex);
2122                         cond_resched();
2123                         mutex_lock(&root->fs_info->pinned_mutex);
2124                 }
2125         }
2126         mutex_unlock(&root->fs_info->pinned_mutex);
2127         return ret;
2128 }
2129
2130 static int finish_current_insert(struct btrfs_trans_handle *trans,
2131                                  struct btrfs_root *extent_root, int all)
2132 {
2133         u64 start;
2134         u64 end;
2135         u64 priv;
2136         u64 search = 0;
2137         u64 skipped = 0;
2138         struct btrfs_fs_info *info = extent_root->fs_info;
2139         struct btrfs_path *path;
2140         struct pending_extent_op *extent_op, *tmp;
2141         struct list_head insert_list, update_list;
2142         int ret;
2143         int num_inserts = 0, max_inserts;
2144
2145         path = btrfs_alloc_path();
2146         INIT_LIST_HEAD(&insert_list);
2147         INIT_LIST_HEAD(&update_list);
2148
2149         max_inserts = extent_root->leafsize /
2150                 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2151                  sizeof(struct btrfs_extent_ref) +
2152                  sizeof(struct btrfs_extent_item));
2153 again:
2154         mutex_lock(&info->extent_ins_mutex);
2155         while (1) {
2156                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2157                                             &end, EXTENT_WRITEBACK);
2158                 if (ret) {
2159                         if (skipped && all && !num_inserts) {
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                                 continue;
2548                         }
2549                         mutex_unlock(&info->extent_ins_mutex);
2550                         break;
2551                 }
2552
2553                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2554                 if (!ret) {
2555                         search = end+1;
2556                         skipped = 1;
2557
2558                         if (need_resched()) {
2559                                 mutex_unlock(&info->extent_ins_mutex);
2560                                 cond_resched();
2561                                 mutex_lock(&info->extent_ins_mutex);
2562                         }
2563
2564                         continue;
2565                 }
2566                 BUG_ON(ret < 0);
2567
2568                 ret = get_state_private(pending_del, start, &priv);
2569                 BUG_ON(ret);
2570                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2571
2572                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2573                                   GFP_NOFS);
2574                 if (!test_range_bit(extent_ins, start, end,
2575                                     EXTENT_WRITEBACK, 0)) {
2576                         list_add_tail(&extent_op->list, &delete_list);
2577                         nr++;
2578                 } else {
2579                         kfree(extent_op);
2580
2581                         ret = get_state_private(&info->extent_ins, start,
2582                                                 &priv);
2583                         BUG_ON(ret);
2584                         extent_op = (struct pending_extent_op *)
2585                                                 (unsigned long)priv;
2586
2587                         clear_extent_bits(&info->extent_ins, start, end,
2588                                           EXTENT_WRITEBACK, GFP_NOFS);
2589
2590                         if (extent_op->type == PENDING_BACKREF_UPDATE) {
2591                                 list_add_tail(&extent_op->list, &delete_list);
2592                                 search = end + 1;
2593                                 nr++;
2594                                 continue;
2595                         }
2596
2597                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2598                         ret = pin_down_bytes(trans, extent_root, start,
2599                                              end + 1 - start, 0);
2600                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2601
2602                         ret = update_block_group(trans, extent_root, start,
2603                                                 end + 1 - start, 0, ret > 0);
2604
2605                         unlock_extent(extent_ins, start, end, GFP_NOFS);
2606                         BUG_ON(ret);
2607                         kfree(extent_op);
2608                 }
2609                 if (ret)
2610                         err = ret;
2611
2612                 search = end + 1;
2613
2614                 if (need_resched()) {
2615                         mutex_unlock(&info->extent_ins_mutex);
2616                         cond_resched();
2617                         mutex_lock(&info->extent_ins_mutex);
2618                 }
2619         }
2620
2621         if (nr) {
2622                 ret = free_extents(trans, extent_root, &delete_list);
2623                 BUG_ON(ret);
2624         }
2625
2626         if (all && skipped) {
2627                 INIT_LIST_HEAD(&delete_list);
2628                 search = 0;
2629                 nr = 0;
2630                 goto again;
2631         }
2632
2633         return err;
2634 }
2635
2636 /*
2637  * remove an extent from the root, returns 0 on success
2638  */
2639 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2640                                struct btrfs_root *root,
2641                                u64 bytenr, u64 num_bytes, u64 parent,
2642                                u64 root_objectid, u64 ref_generation,
2643                                u64 owner_objectid, int pin)
2644 {
2645         struct btrfs_root *extent_root = root->fs_info->extent_root;
2646         int pending_ret;
2647         int ret;
2648
2649         WARN_ON(num_bytes < root->sectorsize);
2650         if (root == extent_root) {
2651                 struct pending_extent_op *extent_op = NULL;
2652
2653                 mutex_lock(&root->fs_info->extent_ins_mutex);
2654                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2655                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2656                         u64 priv;
2657                         ret = get_state_private(&root->fs_info->extent_ins,
2658                                                 bytenr, &priv);
2659                         BUG_ON(ret);
2660                         extent_op = (struct pending_extent_op *)
2661                                                 (unsigned long)priv;
2662
2663                         extent_op->del = 1;
2664                         if (extent_op->type == PENDING_EXTENT_INSERT) {
2665                                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2666                                 return 0;
2667                         }
2668                 }
2669
2670                 if (extent_op) {
2671                         ref_generation = extent_op->orig_generation;
2672                         parent = extent_op->orig_parent;
2673                 }
2674
2675                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2676                 BUG_ON(!extent_op);
2677
2678                 extent_op->type = PENDING_EXTENT_DELETE;
2679                 extent_op->bytenr = bytenr;
2680                 extent_op->num_bytes = num_bytes;
2681                 extent_op->parent = parent;
2682                 extent_op->orig_parent = parent;
2683                 extent_op->generation = ref_generation;
2684                 extent_op->orig_generation = ref_generation;
2685                 extent_op->level = (int)owner_objectid;
2686                 INIT_LIST_HEAD(&extent_op->list);
2687                 extent_op->del = 0;
2688
2689                 set_extent_bits(&root->fs_info->pending_del,
2690                                 bytenr, bytenr + num_bytes - 1,
2691                                 EXTENT_WRITEBACK, GFP_NOFS);
2692                 set_state_private(&root->fs_info->pending_del,
2693                                   bytenr, (unsigned long)extent_op);
2694                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2695                 return 0;
2696         }
2697         /* if metadata always pin */
2698         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2699                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2700                         struct btrfs_block_group_cache *cache;
2701
2702                         /* btrfs_free_reserved_extent */
2703                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2704                         BUG_ON(!cache);
2705                         btrfs_add_free_space(cache, bytenr, num_bytes);
2706                         put_block_group(cache);
2707                         update_reserved_extents(root, bytenr, num_bytes, 0);
2708                         return 0;
2709                 }
2710                 pin = 1;
2711         }
2712
2713         /* if data pin when any transaction has committed this */
2714         if (ref_generation != trans->transid)
2715                 pin = 1;
2716
2717         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2718                             root_objectid, ref_generation,
2719                             owner_objectid, pin, pin == 0);
2720
2721         finish_current_insert(trans, root->fs_info->extent_root, 0);
2722         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2723         return ret ? ret : pending_ret;
2724 }
2725
2726 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2727                       struct btrfs_root *root,
2728                       u64 bytenr, u64 num_bytes, u64 parent,
2729                       u64 root_objectid, u64 ref_generation,
2730                       u64 owner_objectid, int pin)
2731 {
2732         int ret;
2733
2734         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2735                                   root_objectid, ref_generation,
2736                                   owner_objectid, pin);
2737         return ret;
2738 }
2739
2740 static u64 stripe_align(struct btrfs_root *root, u64 val)
2741 {
2742         u64 mask = ((u64)root->stripesize - 1);
2743         u64 ret = (val + mask) & ~mask;
2744         return ret;
2745 }
2746
2747 /*
2748  * walks the btree of allocated extents and find a hole of a given size.
2749  * The key ins is changed to record the hole:
2750  * ins->objectid == block start
2751  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2752  * ins->offset == number of blocks
2753  * Any available blocks before search_start are skipped.
2754  */
2755 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2756                                      struct btrfs_root *orig_root,
2757                                      u64 num_bytes, u64 empty_size,
2758                                      u64 search_start, u64 search_end,
2759                                      u64 hint_byte, struct btrfs_key *ins,
2760                                      u64 exclude_start, u64 exclude_nr,
2761                                      int data)
2762 {
2763         int ret = 0;
2764         struct btrfs_root *root = orig_root->fs_info->extent_root;
2765         u64 total_needed = num_bytes;
2766         u64 *last_ptr = NULL;
2767         u64 last_wanted = 0;
2768         struct btrfs_block_group_cache *block_group = NULL;
2769         int chunk_alloc_done = 0;
2770         int empty_cluster = 2 * 1024 * 1024;
2771         int allowed_chunk_alloc = 0;
2772         struct list_head *head = NULL, *cur = NULL;
2773         int loop = 0;
2774         int extra_loop = 0;
2775         struct btrfs_space_info *space_info;
2776
2777         WARN_ON(num_bytes < root->sectorsize);
2778         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2779         ins->objectid = 0;
2780         ins->offset = 0;
2781
2782         if (orig_root->ref_cows || empty_size)
2783                 allowed_chunk_alloc = 1;
2784
2785         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2786                 last_ptr = &root->fs_info->last_alloc;
2787                 empty_cluster = 64 * 1024;
2788         }
2789
2790         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2791                 last_ptr = &root->fs_info->last_data_alloc;
2792
2793         if (last_ptr) {
2794                 if (*last_ptr) {
2795                         hint_byte = *last_ptr;
2796                         last_wanted = *last_ptr;
2797                 } else
2798                         empty_size += empty_cluster;
2799         } else {
2800                 empty_cluster = 0;
2801         }
2802         search_start = max(search_start, first_logical_byte(root, 0));
2803         search_start = max(search_start, hint_byte);
2804
2805         if (last_wanted && search_start != last_wanted) {
2806                 last_wanted = 0;
2807                 empty_size += empty_cluster;
2808         }
2809
2810         total_needed += empty_size;
2811         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2812         if (!block_group)
2813                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2814                                                              search_start);
2815         space_info = __find_space_info(root->fs_info, data);
2816
2817         down_read(&space_info->groups_sem);
2818         while (1) {
2819                 struct btrfs_free_space *free_space;
2820                 /*
2821                  * the only way this happens if our hint points to a block
2822                  * group thats not of the proper type, while looping this
2823                  * should never happen
2824                  */
2825                 if (empty_size)
2826                         extra_loop = 1;
2827
2828                 if (!block_group)
2829                         goto new_group_no_lock;
2830
2831                 if (unlikely(!block_group->cached)) {
2832                         mutex_lock(&block_group->cache_mutex);
2833                         ret = cache_block_group(root, block_group);
2834                         mutex_unlock(&block_group->cache_mutex);
2835                         if (ret)
2836                                 break;
2837                 }
2838
2839                 mutex_lock(&block_group->alloc_mutex);
2840                 if (unlikely(!block_group_bits(block_group, data)))
2841                         goto new_group;
2842
2843                 if (unlikely(block_group->ro))
2844                         goto new_group;
2845
2846                 free_space = btrfs_find_free_space(block_group, search_start,
2847                                                    total_needed);
2848                 if (free_space) {
2849                         u64 start = block_group->key.objectid;
2850                         u64 end = block_group->key.objectid +
2851                                 block_group->key.offset;
2852
2853                         search_start = stripe_align(root, free_space->offset);
2854
2855                         /* move on to the next group */
2856                         if (search_start + num_bytes >= search_end)
2857                                 goto new_group;
2858
2859                         /* move on to the next group */
2860                         if (search_start + num_bytes > end)
2861                                 goto new_group;
2862
2863                         if (last_wanted && search_start != last_wanted) {
2864                                 total_needed += empty_cluster;
2865                                 empty_size += empty_cluster;
2866                                 last_wanted = 0;
2867                                 /*
2868                                  * if search_start is still in this block group
2869                                  * then we just re-search this block group
2870                                  */
2871                                 if (search_start >= start &&
2872                                     search_start < end) {
2873                                         mutex_unlock(&block_group->alloc_mutex);
2874                                         continue;
2875                                 }
2876
2877                                 /* else we go to the next block group */
2878                                 goto new_group;
2879                         }
2880
2881                         if (exclude_nr > 0 &&
2882                             (search_start + num_bytes > exclude_start &&
2883                              search_start < exclude_start + exclude_nr)) {
2884                                 search_start = exclude_start + exclude_nr;
2885                                 /*
2886                                  * if search_start is still in this block group
2887                                  * then we just re-search this block group
2888                                  */
2889                                 if (search_start >= start &&
2890                                     search_start < end) {
2891                                         mutex_unlock(&block_group->alloc_mutex);
2892                                         last_wanted = 0;
2893                                         continue;
2894                                 }
2895
2896                                 /* else we go to the next block group */
2897                                 goto new_group;
2898                         }
2899
2900                         ins->objectid = search_start;
2901                         ins->offset = num_bytes;
2902
2903                         btrfs_remove_free_space_lock(block_group, search_start,
2904                                                      num_bytes);
2905                         /* we are all good, lets return */
2906                         mutex_unlock(&block_group->alloc_mutex);
2907                         break;
2908                 }
2909 new_group:
2910                 mutex_unlock(&block_group->alloc_mutex);
2911                 put_block_group(block_group);
2912                 block_group = NULL;
2913 new_group_no_lock:
2914                 /* don't try to compare new allocations against the
2915                  * last allocation any more
2916                  */
2917                 last_wanted = 0;
2918
2919                 /*
2920                  * Here's how this works.
2921                  * loop == 0: we were searching a block group via a hint
2922                  *              and didn't find anything, so we start at
2923                  *              the head of the block groups and keep searching
2924                  * loop == 1: we're searching through all of the block groups
2925                  *              if we hit the head again we have searched
2926                  *              all of the block groups for this space and we
2927                  *              need to try and allocate, if we cant error out.
2928                  * loop == 2: we allocated more space and are looping through
2929                  *              all of the block groups again.
2930                  */
2931                 if (loop == 0) {
2932                         head = &space_info->block_groups;
2933                         cur = head->next;
2934                         loop++;
2935                 } else if (loop == 1 && cur == head) {
2936                         int keep_going;
2937
2938                         /* at this point we give up on the empty_size
2939                          * allocations and just try to allocate the min
2940                          * space.
2941                          *
2942                          * The extra_loop field was set if an empty_size
2943                          * allocation was attempted above, and if this
2944                          * is try we need to try the loop again without
2945                          * the additional empty_size.
2946                          */
2947                         total_needed -= empty_size;
2948                         empty_size = 0;
2949                         keep_going = extra_loop;
2950                         loop++;
2951
2952                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2953                                 up_read(&space_info->groups_sem);
2954                                 ret = do_chunk_alloc(trans, root, num_bytes +
2955                                                      2 * 1024 * 1024, data, 1);
2956                                 down_read(&space_info->groups_sem);
2957                                 if (ret < 0)
2958                                         goto loop_check;
2959                                 head = &space_info->block_groups;
2960                                 /*
2961                                  * we've allocated a new chunk, keep
2962                                  * trying
2963                                  */
2964                                 keep_going = 1;
2965                                 chunk_alloc_done = 1;
2966                         } else if (!allowed_chunk_alloc) {
2967                                 space_info->force_alloc = 1;
2968                         }
2969 loop_check:
2970                         if (keep_going) {
2971                                 cur = head->next;
2972                                 extra_loop = 0;
2973                         } else {
2974                                 break;
2975                         }
2976                 } else if (cur == head) {
2977                         break;
2978                 }
2979
2980                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2981                                          list);
2982                 atomic_inc(&block_group->count);
2983
2984                 search_start = block_group->key.objectid;
2985                 cur = cur->next;
2986         }
2987
2988         /* we found what we needed */
2989         if (ins->objectid) {
2990                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2991                         trans->block_group = block_group->key.objectid;
2992
2993                 if (last_ptr)
2994                         *last_ptr = ins->objectid + ins->offset;
2995                 ret = 0;
2996         } else if (!ret) {
2997                 printk(KERN_ERR "btrfs searching for %llu bytes, "
2998                        "num_bytes %llu, loop %d, allowed_alloc %d\n",
2999                        (unsigned long long)total_needed,
3000                        (unsigned long long)num_bytes,
3001                        loop, allowed_chunk_alloc);
3002                 ret = -ENOSPC;
3003         }
3004         if (block_group)
3005                 put_block_group(block_group);
3006
3007         up_read(&space_info->groups_sem);
3008         return ret;
3009 }
3010
3011 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3012 {
3013         struct btrfs_block_group_cache *cache;
3014
3015         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3016                (unsigned long long)(info->total_bytes - info->bytes_used -
3017                                     info->bytes_pinned - info->bytes_reserved),
3018                (info->full) ? "" : "not ");
3019
3020         down_read(&info->groups_sem);
3021         list_for_each_entry(cache, &info->block_groups, list) {
3022                 spin_lock(&cache->lock);
3023                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3024                        "%llu pinned %llu reserved\n",
3025                        (unsigned long long)cache->key.objectid,
3026                        (unsigned long long)cache->key.offset,
3027                        (unsigned long long)btrfs_block_group_used(&cache->item),
3028                        (unsigned long long)cache->pinned,
3029                        (unsigned long long)cache->reserved);
3030                 btrfs_dump_free_space(cache, bytes);
3031                 spin_unlock(&cache->lock);
3032         }
3033         up_read(&info->groups_sem);
3034 }
3035
3036 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3037                                   struct btrfs_root *root,
3038                                   u64 num_bytes, u64 min_alloc_size,
3039                                   u64 empty_size, u64 hint_byte,
3040                                   u64 search_end, struct btrfs_key *ins,
3041                                   u64 data)
3042 {
3043         int ret;
3044         u64 search_start = 0;
3045         u64 alloc_profile;
3046         struct btrfs_fs_info *info = root->fs_info;
3047
3048         if (data) {
3049                 alloc_profile = info->avail_data_alloc_bits &
3050                         info->data_alloc_profile;
3051                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3052         } else if (root == root->fs_info->chunk_root) {
3053                 alloc_profile = info->avail_system_alloc_bits &
3054                         info->system_alloc_profile;
3055                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3056         } else {
3057                 alloc_profile = info->avail_metadata_alloc_bits &
3058                         info->metadata_alloc_profile;
3059                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3060         }
3061 again:
3062         data = btrfs_reduce_alloc_profile(root, data);
3063         /*
3064          * the only place that sets empty_size is btrfs_realloc_node, which
3065          * is not called recursively on allocations
3066          */
3067         if (empty_size || root->ref_cows) {
3068                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3069                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3070                                      2 * 1024 * 1024,
3071                                      BTRFS_BLOCK_GROUP_METADATA |
3072                                      (info->metadata_alloc_profile &
3073                                       info->avail_metadata_alloc_bits), 0);
3074                 }
3075                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3076                                      num_bytes + 2 * 1024 * 1024, data, 0);
3077         }
3078
3079         WARN_ON(num_bytes < root->sectorsize);
3080         ret = find_free_extent(trans, root, num_bytes, empty_size,
3081                                search_start, search_end, hint_byte, ins,
3082                                trans->alloc_exclude_start,
3083                                trans->alloc_exclude_nr, data);
3084
3085         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3086                 num_bytes = num_bytes >> 1;
3087                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3088                 num_bytes = max(num_bytes, min_alloc_size);
3089                 do_chunk_alloc(trans, root->fs_info->extent_root,
3090                                num_bytes, data, 1);
3091                 goto again;
3092         }
3093         if (ret) {
3094                 struct btrfs_space_info *sinfo;
3095
3096                 sinfo = __find_space_info(root->fs_info, data);
3097                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3098                        "wanted %llu\n", (unsigned long long)data,
3099                        (unsigned long long)num_bytes);
3100                 dump_space_info(sinfo, num_bytes);
3101                 BUG();
3102         }
3103
3104         return ret;
3105 }
3106
3107 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3108 {
3109         struct btrfs_block_group_cache *cache;
3110         int ret = 0;
3111
3112         cache = btrfs_lookup_block_group(root->fs_info, start);
3113         if (!cache) {
3114                 printk(KERN_ERR "Unable to find block group for %llu\n",
3115                        (unsigned long long)start);
3116                 return -ENOSPC;
3117         }
3118
3119         ret = btrfs_discard_extent(root, start, len);
3120
3121         btrfs_add_free_space(cache, start, len);
3122         put_block_group(cache);
3123         update_reserved_extents(root, start, len, 0);
3124
3125         return ret;
3126 }
3127
3128 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3129                                   struct btrfs_root *root,
3130                                   u64 num_bytes, u64 min_alloc_size,
3131                                   u64 empty_size, u64 hint_byte,
3132                                   u64 search_end, struct btrfs_key *ins,
3133                                   u64 data)
3134 {
3135         int ret;
3136         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3137                                      empty_size, hint_byte, search_end, ins,
3138                                      data);
3139         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3140         return ret;
3141 }
3142
3143 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3144                                          struct btrfs_root *root, u64 parent,
3145                                          u64 root_objectid, u64 ref_generation,
3146                                          u64 owner, struct btrfs_key *ins)
3147 {
3148         int ret;
3149         int pending_ret;
3150         u64 super_used;
3151         u64 root_used;
3152         u64 num_bytes = ins->offset;
3153         u32 sizes[2];
3154         struct btrfs_fs_info *info = root->fs_info;
3155         struct btrfs_root *extent_root = info->extent_root;
3156         struct btrfs_extent_item *extent_item;
3157         struct btrfs_extent_ref *ref;
3158         struct btrfs_path *path;
3159         struct btrfs_key keys[2];
3160
3161         if (parent == 0)
3162                 parent = ins->objectid;
3163
3164         /* block accounting for super block */
3165         spin_lock(&info->delalloc_lock);
3166         super_used = btrfs_super_bytes_used(&info->super_copy);
3167         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3168
3169         /* block accounting for root item */
3170         root_used = btrfs_root_used(&root->root_item);
3171         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3172         spin_unlock(&info->delalloc_lock);
3173
3174         if (root == extent_root) {
3175                 struct pending_extent_op *extent_op;
3176
3177                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3178                 BUG_ON(!extent_op);
3179
3180                 extent_op->type = PENDING_EXTENT_INSERT;
3181                 extent_op->bytenr = ins->objectid;
3182                 extent_op->num_bytes = ins->offset;
3183                 extent_op->parent = parent;
3184                 extent_op->orig_parent = 0;
3185                 extent_op->generation = ref_generation;
3186                 extent_op->orig_generation = 0;
3187                 extent_op->level = (int)owner;
3188                 INIT_LIST_HEAD(&extent_op->list);
3189                 extent_op->del = 0;
3190
3191                 mutex_lock(&root->fs_info->extent_ins_mutex);
3192                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3193                                 ins->objectid + ins->offset - 1,
3194                                 EXTENT_WRITEBACK, GFP_NOFS);
3195                 set_state_private(&root->fs_info->extent_ins,
3196                                   ins->objectid, (unsigned long)extent_op);
3197                 mutex_unlock(&root->fs_info->extent_ins_mutex);
3198                 goto update_block;
3199         }
3200
3201         memcpy(&keys[0], ins, sizeof(*ins));
3202         keys[1].objectid = ins->objectid;
3203         keys[1].type = BTRFS_EXTENT_REF_KEY;
3204         keys[1].offset = parent;
3205         sizes[0] = sizeof(*extent_item);
3206         sizes[1] = sizeof(*ref);
3207
3208         path = btrfs_alloc_path();
3209         BUG_ON(!path);
3210
3211         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3212                                        sizes, 2);
3213         BUG_ON(ret);
3214
3215         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3216                                      struct btrfs_extent_item);
3217         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3218         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3219                              struct btrfs_extent_ref);
3220
3221         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3222         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3223         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3224         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3225
3226         btrfs_mark_buffer_dirty(path->nodes[0]);
3227
3228         trans->alloc_exclude_start = 0;
3229         trans->alloc_exclude_nr = 0;
3230         btrfs_free_path(path);
3231         finish_current_insert(trans, extent_root, 0);
3232         pending_ret = del_pending_extents(trans, extent_root, 0);
3233
3234         if (ret)
3235                 goto out;
3236         if (pending_ret) {
3237                 ret = pending_ret;
3238                 goto out;
3239         }
3240
3241 update_block:
3242         ret = update_block_group(trans, root, ins->objectid,
3243                                  ins->offset, 1, 0);
3244         if (ret) {
3245                 printk(KERN_ERR "btrfs update block group failed for %llu "
3246                        "%llu\n", (unsigned long long)ins->objectid,
3247                        (unsigned long long)ins->offset);
3248                 BUG();
3249         }
3250 out:
3251         return ret;
3252 }
3253
3254 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3255                                 struct btrfs_root *root, u64 parent,
3256                                 u64 root_objectid, u64 ref_generation,
3257                                 u64 owner, struct btrfs_key *ins)
3258 {
3259         int ret;
3260
3261         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3262                 return 0;
3263         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3264                                             ref_generation, owner, ins);
3265         update_reserved_extents(root, ins->objectid, ins->offset, 0);
3266         return ret;
3267 }
3268
3269 /*
3270  * this is used by the tree logging recovery code.  It records that
3271  * an extent has been allocated and makes sure to clear the free
3272  * space cache bits as well
3273  */
3274 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3275                                 struct btrfs_root *root, u64 parent,
3276                                 u64 root_objectid, u64 ref_generation,
3277                                 u64 owner, struct btrfs_key *ins)
3278 {
3279         int ret;
3280         struct btrfs_block_group_cache *block_group;
3281
3282         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3283         mutex_lock(&block_group->cache_mutex);
3284         cache_block_group(root, block_group);
3285         mutex_unlock(&block_group->cache_mutex);
3286
3287         ret = btrfs_remove_free_space(block_group, ins->objectid,
3288                                       ins->offset);
3289         BUG_ON(ret);
3290         put_block_group(block_group);
3291         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3292                                             ref_generation, owner, ins);
3293         return ret;
3294 }
3295
3296 /*
3297  * finds a free extent and does all the dirty work required for allocation
3298  * returns the key for the extent through ins, and a tree buffer for
3299  * the first block of the extent through buf.
3300  *
3301  * returns 0 if everything worked, non-zero otherwise.
3302  */
3303 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3304                        struct btrfs_root *root,
3305                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3306                        u64 root_objectid, u64 ref_generation,
3307                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3308                        u64 search_end, struct btrfs_key *ins, u64 data)
3309 {
3310         int ret;
3311
3312         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3313                                      min_alloc_size, empty_size, hint_byte,
3314                                      search_end, ins, data);
3315         BUG_ON(ret);
3316         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3317                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3318                                         root_objectid, ref_generation,
3319                                         owner_objectid, ins);
3320                 BUG_ON(ret);
3321
3322         } else {
3323                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3324         }
3325         return ret;
3326 }
3327
3328 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3329                                             struct btrfs_root *root,
3330                                             u64 bytenr, u32 blocksize)
3331 {
3332         struct extent_buffer *buf;
3333
3334         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3335         if (!buf)
3336                 return ERR_PTR(-ENOMEM);
3337         btrfs_set_header_generation(buf, trans->transid);
3338         btrfs_tree_lock(buf);
3339         clean_tree_block(trans, root, buf);
3340         btrfs_set_buffer_uptodate(buf);
3341         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3342                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3343                          buf->start + buf->len - 1, GFP_NOFS);
3344         } else {
3345                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3346                          buf->start + buf->len - 1, GFP_NOFS);
3347         }
3348         trans->blocks_used++;
3349         return buf;
3350 }
3351
3352 /*
3353  * helper function to allocate a block for a given tree
3354  * returns the tree buffer or NULL.
3355  */
3356 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3357                                              struct btrfs_root *root,
3358                                              u32 blocksize, u64 parent,
3359                                              u64 root_objectid,
3360                                              u64 ref_generation,
3361                                              int level,
3362                                              u64 hint,
3363                                              u64 empty_size)
3364 {
3365         struct btrfs_key ins;
3366         int ret;
3367         struct extent_buffer *buf;
3368
3369         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3370                                  root_objectid, ref_generation, level,
3371                                  empty_size, hint, (u64)-1, &ins, 0);
3372         if (ret) {
3373                 BUG_ON(ret > 0);
3374                 return ERR_PTR(ret);
3375         }
3376
3377         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3378         return buf;
3379 }
3380
3381 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3382                         struct btrfs_root *root, struct extent_buffer *leaf)
3383 {
3384         u64 leaf_owner;
3385         u64 leaf_generation;
3386         struct btrfs_key key;
3387         struct btrfs_file_extent_item *fi;
3388         int i;
3389         int nritems;
3390         int ret;
3391
3392         BUG_ON(!btrfs_is_leaf(leaf));
3393         nritems = btrfs_header_nritems(leaf);
3394         leaf_owner = btrfs_header_owner(leaf);
3395         leaf_generation = btrfs_header_generation(leaf);
3396
3397         for (i = 0; i < nritems; i++) {
3398                 u64 disk_bytenr;
3399                 cond_resched();
3400
3401                 btrfs_item_key_to_cpu(leaf, &key, i);
3402                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3403                         continue;
3404                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3405                 if (btrfs_file_extent_type(leaf, fi) ==
3406                     BTRFS_FILE_EXTENT_INLINE)
3407                         continue;
3408                 /*
3409                  * FIXME make sure to insert a trans record that
3410                  * repeats the snapshot del on crash
3411                  */
3412                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3413                 if (disk_bytenr == 0)
3414                         continue;
3415
3416                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3417                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3418                                 leaf->start, leaf_owner, leaf_generation,
3419                                 key.objectid, 0);
3420                 BUG_ON(ret);
3421
3422                 atomic_inc(&root->fs_info->throttle_gen);
3423                 wake_up(&root->fs_info->transaction_throttle);
3424                 cond_resched();
3425         }
3426         return 0;
3427 }
3428
3429 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3430                                         struct btrfs_root *root,
3431                                         struct btrfs_leaf_ref *ref)
3432 {
3433         int i;
3434         int ret;
3435         struct btrfs_extent_info *info = ref->extents;
3436
3437         for (i = 0; i < ref->nritems; i++) {
3438                 ret = __btrfs_free_extent(trans, root, info->bytenr,
3439                                           info->num_bytes, ref->bytenr,
3440                                           ref->owner, ref->generation,
3441                                           info->objectid, 0);
3442
3443                 atomic_inc(&root->fs_info->throttle_gen);
3444                 wake_up(&root->fs_info->transaction_throttle);
3445                 cond_resched();
3446
3447                 BUG_ON(ret);
3448                 info++;
3449         }
3450
3451         return 0;
3452 }
3453
3454 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3455                                      u64 len, u32 *refs)
3456 {
3457         int ret;
3458
3459         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3460         BUG_ON(ret);
3461
3462 #if 0 /* some debugging code in case we see problems here */
3463         /* if the refs count is one, it won't get increased again.  But
3464          * if the ref count is > 1, someone may be decreasing it at
3465          * the same time we are.
3466          */
3467         if (*refs != 1) {
3468                 struct extent_buffer *eb = NULL;
3469                 eb = btrfs_find_create_tree_block(root, start, len);
3470                 if (eb)
3471                         btrfs_tree_lock(eb);
3472
3473                 mutex_lock(&root->fs_info->alloc_mutex);
3474                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3475                 BUG_ON(ret);
3476                 mutex_unlock(&root->fs_info->alloc_mutex);
3477
3478                 if (eb) {
3479                         btrfs_tree_unlock(eb);
3480                         free_extent_buffer(eb);
3481                 }
3482                 if (*refs == 1) {
3483                         printk(KERN_ERR "btrfs block %llu went down to one "
3484                                "during drop_snap\n", (unsigned long long)start);
3485                 }
3486
3487         }
3488 #endif
3489
3490         cond_resched();
3491         return ret;
3492 }
3493
3494 /*
3495  * helper function for drop_snapshot, this walks down the tree dropping ref
3496  * counts as it goes.
3497  */
3498 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3499                                    struct btrfs_root *root,
3500                                    struct btrfs_path *path, int *level)
3501 {
3502         u64 root_owner;
3503         u64 root_gen;
3504         u64 bytenr;
3505         u64 ptr_gen;
3506         struct extent_buffer *next;
3507         struct extent_buffer *cur;
3508         struct extent_buffer *parent;
3509         struct btrfs_leaf_ref *ref;
3510         u32 blocksize;
3511         int ret;
3512         u32 refs;
3513
3514         WARN_ON(*level < 0);
3515         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3516         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3517                                 path->nodes[*level]->len, &refs);
3518         BUG_ON(ret);
3519         if (refs > 1)
3520                 goto out;
3521
3522         /*
3523          * walk down to the last node level and free all the leaves
3524          */
3525         while (*level >= 0) {
3526                 WARN_ON(*level < 0);
3527                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3528                 cur = path->nodes[*level];
3529
3530                 if (btrfs_header_level(cur) != *level)
3531                         WARN_ON(1);
3532
3533                 if (path->slots[*level] >=
3534                     btrfs_header_nritems(cur))
3535                         break;
3536                 if (*level == 0) {
3537                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3538                         BUG_ON(ret);
3539                         break;
3540                 }
3541                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3542                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3543                 blocksize = btrfs_level_size(root, *level - 1);
3544
3545                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3546                 BUG_ON(ret);
3547                 if (refs != 1) {
3548                         parent = path->nodes[*level];
3549                         root_owner = btrfs_header_owner(parent);
3550                         root_gen = btrfs_header_generation(parent);
3551                         path->slots[*level]++;
3552
3553                         ret = __btrfs_free_extent(trans, root, bytenr,
3554                                                 blocksize, parent->start,
3555                                                 root_owner, root_gen,
3556                                                 *level - 1, 1);
3557                         BUG_ON(ret);
3558
3559                         atomic_inc(&root->fs_info->throttle_gen);
3560                         wake_up(&root->fs_info->transaction_throttle);
3561                         cond_resched();
3562
3563                         continue;
3564                 }
3565                 /*
3566                  * at this point, we have a single ref, and since the
3567                  * only place referencing this extent is a dead root
3568                  * the reference count should never go higher.
3569                  * So, we don't need to check it again
3570                  */
3571                 if (*level == 1) {
3572                         ref = btrfs_lookup_leaf_ref(root, bytenr);
3573                         if (ref && ref->generation != ptr_gen) {
3574                                 btrfs_free_leaf_ref(root, ref);
3575                                 ref = NULL;
3576                         }
3577                         if (ref) {
3578                                 ret = cache_drop_leaf_ref(trans, root, ref);
3579                                 BUG_ON(ret);
3580                                 btrfs_remove_leaf_ref(root, ref);
3581                                 btrfs_free_leaf_ref(root, ref);
3582                                 *level = 0;
3583                                 break;
3584                         }
3585                 }
3586                 next = btrfs_find_tree_block(root, bytenr, blocksize);
3587                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3588                         free_extent_buffer(next);
3589
3590                         next = read_tree_block(root, bytenr, blocksize,
3591                                                ptr_gen);
3592                         cond_resched();
3593 #if 0
3594                         /*
3595                          * this is a debugging check and can go away
3596                          * the ref should never go all the way down to 1
3597                          * at this point
3598                          */
3599                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3600                                                 &refs);
3601                         BUG_ON(ret);
3602                         WARN_ON(refs != 1);
3603 #endif
3604                 }
3605                 WARN_ON(*level <= 0);
3606                 if (path->nodes[*level-1])
3607                         free_extent_buffer(path->nodes[*level-1]);
3608                 path->nodes[*level-1] = next;
3609                 *level = btrfs_header_level(next);
3610                 path->slots[*level] = 0;
3611                 cond_resched();
3612         }
3613 out:
3614         WARN_ON(*level < 0);
3615         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3616
3617         if (path->nodes[*level] == root->node) {
3618                 parent = path->nodes[*level];
3619                 bytenr = path->nodes[*level]->start;
3620         } else {
3621                 parent = path->nodes[*level + 1];
3622                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3623         }
3624
3625         blocksize = btrfs_level_size(root, *level);
3626         root_owner = btrfs_header_owner(parent);
3627         root_gen = btrfs_header_generation(parent);
3628
3629         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3630                                   parent->start, root_owner, root_gen,
3631                                   *level, 1);
3632         free_extent_buffer(path->nodes[*level]);
3633         path->nodes[*level] = NULL;
3634         *level += 1;
3635         BUG_ON(ret);
3636
3637         cond_resched();
3638         return 0;
3639 }
3640
3641 /*
3642  * helper function for drop_subtree, this function is similar to
3643  * walk_down_tree. The main difference is that it checks reference
3644  * counts while tree blocks are locked.
3645  */
3646 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3647                                       struct btrfs_root *root,
3648                                       struct btrfs_path *path, int *level)
3649 {
3650         struct extent_buffer *next;
3651         struct extent_buffer *cur;
3652         struct extent_buffer *parent;
3653         u64 bytenr;
3654         u64 ptr_gen;
3655         u32 blocksize;
3656         u32 refs;
3657         int ret;
3658
3659         cur = path->nodes[*level];
3660         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3661                                       &refs);
3662         BUG_ON(ret);
3663         if (refs > 1)
3664                 goto out;
3665
3666         while (*level >= 0) {
3667                 cur = path->nodes[*level];
3668                 if (*level == 0) {
3669                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3670                         BUG_ON(ret);
3671                         clean_tree_block(trans, root, cur);
3672                         break;
3673                 }
3674                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3675                         clean_tree_block(trans, root, cur);
3676                         break;
3677                 }
3678
3679                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3680                 blocksize = btrfs_level_size(root, *level - 1);
3681                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3682
3683                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3684                 btrfs_tree_lock(next);
3685
3686                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3687                                               &refs);
3688                 BUG_ON(ret);
3689                 if (refs > 1) {
3690                         parent = path->nodes[*level];
3691                         ret = btrfs_free_extent(trans, root, bytenr,
3692                                         blocksize, parent->start,
3693                                         btrfs_header_owner(parent),
3694                                         btrfs_header_generation(parent),
3695                                         *level - 1, 1);
3696                         BUG_ON(ret);
3697                         path->slots[*level]++;
3698                         btrfs_tree_unlock(next);
3699                         free_extent_buffer(next);
3700                         continue;
3701                 }
3702
3703                 *level = btrfs_header_level(next);
3704                 path->nodes[*level] = next;
3705                 path->slots[*level] = 0;
3706                 path->locks[*level] = 1;
3707                 cond_resched();
3708         }
3709 out:
3710         parent = path->nodes[*level + 1];
3711         bytenr = path->nodes[*level]->start;
3712         blocksize = path->nodes[*level]->len;
3713
3714         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3715                         parent->start, btrfs_header_owner(parent),
3716                         btrfs_header_generation(parent), *level, 1);
3717         BUG_ON(ret);
3718
3719         if (path->locks[*level]) {
3720                 btrfs_tree_unlock(path->nodes[*level]);
3721                 path->locks[*level] = 0;
3722         }
3723         free_extent_buffer(path->nodes[*level]);
3724         path->nodes[*level] = NULL;
3725         *level += 1;
3726         cond_resched();
3727         return 0;
3728 }
3729
3730 /*
3731  * helper for dropping snapshots.  This walks back up the tree in the path
3732  * to find the first node higher up where we haven't yet gone through
3733  * all the slots
3734  */
3735 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3736                                  struct btrfs_root *root,
3737                                  struct btrfs_path *path,
3738                                  int *level, int max_level)
3739 {
3740         u64 root_owner;
3741         u64 root_gen;
3742         struct btrfs_root_item *root_item = &root->root_item;
3743         int i;
3744         int slot;
3745         int ret;
3746
3747         for (i = *level; i < max_level && path->nodes[i]; i++) {
3748                 slot = path->slots[i];
3749                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3750                         struct extent_buffer *node;
3751                         struct btrfs_disk_key disk_key;
3752                         node = path->nodes[i];
3753                         path->slots[i]++;
3754                         *level = i;
3755                         WARN_ON(*level == 0);
3756                         btrfs_node_key(node, &disk_key, path->slots[i]);
3757                         memcpy(&root_item->drop_progress,
3758                                &disk_key, sizeof(disk_key));
3759                         root_item->drop_level = i;
3760                         return 0;
3761                 } else {
3762                         struct extent_buffer *parent;
3763                         if (path->nodes[*level] == root->node)
3764                                 parent = path->nodes[*level];
3765                         else
3766                                 parent = path->nodes[*level + 1];
3767
3768                         root_owner = btrfs_header_owner(parent);
3769                         root_gen = btrfs_header_generation(parent);
3770
3771                         clean_tree_block(trans, root, path->nodes[*level]);
3772                         ret = btrfs_free_extent(trans, root,
3773                                                 path->nodes[*level]->start,
3774                                                 path->nodes[*level]->len,
3775                                                 parent->start, root_owner,
3776                                                 root_gen, *level, 1);
3777                         BUG_ON(ret);
3778                         if (path->locks[*level]) {
3779                                 btrfs_tree_unlock(path->nodes[*level]);
3780                                 path->locks[*level] = 0;
3781                         }
3782                         free_extent_buffer(path->nodes[*level]);
3783                         path->nodes[*level] = NULL;
3784                         *level = i + 1;
3785                 }
3786         }
3787         return 1;
3788 }
3789
3790 /*
3791  * drop the reference count on the tree rooted at 'snap'.  This traverses
3792  * the tree freeing any blocks that have a ref count of zero after being
3793  * decremented.
3794  */
3795 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3796                         *root)
3797 {
3798         int ret = 0;
3799         int wret;
3800         int level;
3801         struct btrfs_path *path;
3802         int i;
3803         int orig_level;
3804         struct btrfs_root_item *root_item = &root->root_item;
3805
3806         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3807         path = btrfs_alloc_path();
3808         BUG_ON(!path);
3809
3810         level = btrfs_header_level(root->node);
3811         orig_level = level;
3812         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3813                 path->nodes[level] = root->node;
3814                 extent_buffer_get(root->node);
3815                 path->slots[level] = 0;
3816         } else {
3817                 struct btrfs_key key;
3818                 struct btrfs_disk_key found_key;
3819                 struct extent_buffer *node;
3820
3821                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3822                 level = root_item->drop_level;
3823                 path->lowest_level = level;
3824                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3825                 if (wret < 0) {
3826                         ret = wret;
3827                         goto out;
3828                 }
3829                 node = path->nodes[level];
3830                 btrfs_node_key(node, &found_key, path->slots[level]);
3831                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3832                                sizeof(found_key)));
3833                 /*
3834                  * unlock our path, this is safe because only this
3835                  * function is allowed to delete this snapshot
3836                  */
3837                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3838                         if (path->nodes[i] && path->locks[i]) {
3839                                 path->locks[i] = 0;
3840                                 btrfs_tree_unlock(path->nodes[i]);
3841                         }
3842                 }
3843         }
3844         while (1) {
3845                 wret = walk_down_tree(trans, root, path, &level);
3846                 if (wret > 0)
3847                         break;
3848                 if (wret < 0)
3849                         ret = wret;
3850
3851                 wret = walk_up_tree(trans, root, path, &level,
3852                                     BTRFS_MAX_LEVEL);
3853                 if (wret > 0)
3854                         break;
3855                 if (wret < 0)
3856                         ret = wret;
3857                 if (trans->transaction->in_commit) {
3858                         ret = -EAGAIN;
3859                         break;
3860                 }
3861                 atomic_inc(&root->fs_info->throttle_gen);
3862                 wake_up(&root->fs_info->transaction_throttle);
3863         }
3864         for (i = 0; i <= orig_level; i++) {
3865                 if (path->nodes[i]) {
3866                         free_extent_buffer(path->nodes[i]);
3867                         path->nodes[i] = NULL;
3868                 }
3869         }
3870 out:
3871         btrfs_free_path(path);
3872         return ret;
3873 }
3874
3875 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3876                         struct btrfs_root *root,
3877                         struct extent_buffer *node,
3878                         struct extent_buffer *parent)
3879 {
3880         struct btrfs_path *path;
3881         int level;
3882         int parent_level;
3883         int ret = 0;
3884         int wret;
3885
3886         path = btrfs_alloc_path();
3887         BUG_ON(!path);
3888
3889         BUG_ON(!btrfs_tree_locked(parent));
3890         parent_level = btrfs_header_level(parent);
3891         extent_buffer_get(parent);
3892         path->nodes[parent_level] = parent;
3893         path->slots[parent_level] = btrfs_header_nritems(parent);
3894
3895         BUG_ON(!btrfs_tree_locked(node));
3896         level = btrfs_header_level(node);
3897         extent_buffer_get(node);
3898         path->nodes[level] = node;
3899         path->slots[level] = 0;
3900
3901         while (1) {
3902                 wret = walk_down_subtree(trans, root, path, &level);
3903                 if (wret < 0)
3904                         ret = wret;
3905                 if (wret != 0)
3906                         break;
3907
3908                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3909                 if (wret < 0)
3910                         ret = wret;
3911                 if (wret != 0)
3912                         break;
3913         }
3914
3915         btrfs_free_path(path);
3916         return ret;
3917 }
3918
3919 static unsigned long calc_ra(unsigned long start, unsigned long last,
3920                              unsigned long nr)
3921 {
3922         return min(last, start + nr - 1);
3923 }
3924
3925 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3926                                          u64 len)
3927 {
3928         u64 page_start;
3929         u64 page_end;
3930         unsigned long first_index;
3931         unsigned long last_index;
3932         unsigned long i;
3933         struct page *page;
3934         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3935         struct file_ra_state *ra;
3936         struct btrfs_ordered_extent *ordered;
3937         unsigned int total_read = 0;
3938         unsigned int total_dirty = 0;
3939         int ret = 0;
3940
3941         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3942
3943         mutex_lock(&inode->i_mutex);
3944         first_index = start >> PAGE_CACHE_SHIFT;
3945         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3946
3947         /* make sure the dirty trick played by the caller work */
3948         ret = invalidate_inode_pages2_range(inode->i_mapping,
3949                                             first_index, last_index);
3950         if (ret)
3951                 goto out_unlock;
3952
3953         file_ra_state_init(ra, inode->i_mapping);
3954
3955         for (i = first_index ; i <= last_index; i++) {
3956                 if (total_read % ra->ra_pages == 0) {
3957                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3958                                        calc_ra(i, last_index, ra->ra_pages));
3959                 }
3960                 total_read++;
3961 again:
3962                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3963                         BUG_ON(1);
3964                 page = grab_cache_page(inode->i_mapping, i);
3965                 if (!page) {
3966                         ret = -ENOMEM;
3967                         goto out_unlock;
3968                 }
3969                 if (!PageUptodate(page)) {
3970                         btrfs_readpage(NULL, page);
3971                         lock_page(page);
3972                         if (!PageUptodate(page)) {
3973                                 unlock_page(page);
3974                                 page_cache_release(page);
3975                                 ret = -EIO;
3976                                 goto out_unlock;
3977                         }
3978                 }
3979                 wait_on_page_writeback(page);
3980
3981                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3982                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3983                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3984
3985                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3986                 if (ordered) {
3987                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3988                         unlock_page(page);
3989                         page_cache_release(page);
3990                         btrfs_start_ordered_extent(inode, ordered, 1);
3991                         btrfs_put_ordered_extent(ordered);
3992                         goto again;
3993                 }
3994                 set_page_extent_mapped(page);
3995
3996                 if (i == first_index)
3997                         set_extent_bits(io_tree, page_start, page_end,
3998                                         EXTENT_BOUNDARY, GFP_NOFS);
3999                 btrfs_set_extent_delalloc(inode, page_start, page_end);
4000
4001                 set_page_dirty(page);
4002                 total_dirty++;
4003
4004                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4005                 unlock_page(page);
4006                 page_cache_release(page);
4007         }
4008
4009 out_unlock:
4010         kfree(ra);
4011         mutex_unlock(&inode->i_mutex);
4012         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4013         return ret;
4014 }
4015
4016 static noinline int relocate_data_extent(struct inode *reloc_inode,
4017                                          struct btrfs_key *extent_key,
4018                                          u64 offset)
4019 {
4020         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4021         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4022         struct extent_map *em;
4023         u64 start = extent_key->objectid - offset;
4024         u64 end = start + extent_key->offset - 1;
4025
4026         em = alloc_extent_map(GFP_NOFS);
4027         BUG_ON(!em || IS_ERR(em));
4028
4029         em->start = start;
4030         em->len = extent_key->offset;
4031         em->block_len = extent_key->offset;
4032         em->block_start = extent_key->objectid;
4033         em->bdev = root->fs_info->fs_devices->latest_bdev;
4034         set_bit(EXTENT_FLAG_PINNED, &em->flags);
4035
4036         /* setup extent map to cheat btrfs_readpage */
4037         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4038         while (1) {
4039                 int ret;
4040                 spin_lock(&em_tree->lock);
4041                 ret = add_extent_mapping(em_tree, em);
4042                 spin_unlock(&em_tree->lock);
4043                 if (ret != -EEXIST) {
4044                         free_extent_map(em);
4045                         break;
4046                 }
4047                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4048         }
4049         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4050
4051         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4052 }
4053
4054 struct btrfs_ref_path {
4055         u64 extent_start;
4056         u64 nodes[BTRFS_MAX_LEVEL];
4057         u64 root_objectid;
4058         u64 root_generation;
4059         u64 owner_objectid;
4060         u32 num_refs;
4061         int lowest_level;
4062         int current_level;
4063         int shared_level;
4064
4065         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4066         u64 new_nodes[BTRFS_MAX_LEVEL];
4067 };
4068
4069 struct disk_extent {
4070         u64 ram_bytes;
4071         u64 disk_bytenr;
4072         u64 disk_num_bytes;
4073         u64 offset;
4074         u64 num_bytes;
4075         u8 compression;
4076         u8 encryption;
4077         u16 other_encoding;
4078 };
4079
4080 static int is_cowonly_root(u64 root_objectid)
4081 {
4082         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4083             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4084             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4085             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4086             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4087             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4088                 return 1;
4089         return 0;
4090 }
4091
4092 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4093                                     struct btrfs_root *extent_root,
4094                                     struct btrfs_ref_path *ref_path,
4095                                     int first_time)
4096 {
4097         struct extent_buffer *leaf;
4098         struct btrfs_path *path;
4099         struct btrfs_extent_ref *ref;
4100         struct btrfs_key key;
4101         struct btrfs_key found_key;
4102         u64 bytenr;
4103         u32 nritems;
4104         int level;
4105         int ret = 1;
4106
4107         path = btrfs_alloc_path();
4108         if (!path)
4109                 return -ENOMEM;
4110
4111         if (first_time) {
4112                 ref_path->lowest_level = -1;
4113                 ref_path->current_level = -1;
4114                 ref_path->shared_level = -1;
4115                 goto walk_up;
4116         }
4117 walk_down:
4118         level = ref_path->current_level - 1;
4119         while (level >= -1) {
4120                 u64 parent;
4121                 if (level < ref_path->lowest_level)
4122                         break;
4123
4124                 if (level >= 0)
4125                         bytenr = ref_path->nodes[level];
4126                 else
4127                         bytenr = ref_path->extent_start;
4128                 BUG_ON(bytenr == 0);
4129
4130                 parent = ref_path->nodes[level + 1];
4131                 ref_path->nodes[level + 1] = 0;
4132                 ref_path->current_level = level;
4133                 BUG_ON(parent == 0);
4134
4135                 key.objectid = bytenr;
4136                 key.offset = parent + 1;
4137                 key.type = BTRFS_EXTENT_REF_KEY;
4138
4139                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4140                 if (ret < 0)
4141                         goto out;
4142                 BUG_ON(ret == 0);
4143
4144                 leaf = path->nodes[0];
4145                 nritems = btrfs_header_nritems(leaf);
4146                 if (path->slots[0] >= nritems) {
4147                         ret = btrfs_next_leaf(extent_root, path);
4148                         if (ret < 0)
4149                                 goto out;
4150                         if (ret > 0)
4151                                 goto next;
4152                         leaf = path->nodes[0];
4153                 }
4154
4155                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4156                 if (found_key.objectid == bytenr &&
4157                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4158                         if (level < ref_path->shared_level)
4159                                 ref_path->shared_level = level;
4160                         goto found;
4161                 }
4162 next:
4163                 level--;
4164                 btrfs_release_path(extent_root, path);
4165                 cond_resched();
4166         }
4167         /* reached lowest level */
4168         ret = 1;
4169         goto out;
4170 walk_up:
4171         level = ref_path->current_level;
4172         while (level < BTRFS_MAX_LEVEL - 1) {
4173                 u64 ref_objectid;
4174
4175                 if (level >= 0)
4176                         bytenr = ref_path->nodes[level];
4177                 else
4178                         bytenr = ref_path->extent_start;
4179
4180                 BUG_ON(bytenr == 0);
4181
4182                 key.objectid = bytenr;
4183                 key.offset = 0;
4184                 key.type = BTRFS_EXTENT_REF_KEY;
4185
4186                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4187                 if (ret < 0)
4188                         goto out;
4189
4190                 leaf = path->nodes[0];
4191                 nritems = btrfs_header_nritems(leaf);
4192                 if (path->slots[0] >= nritems) {
4193                         ret = btrfs_next_leaf(extent_root, path);
4194                         if (ret < 0)
4195                                 goto out;
4196                         if (ret > 0) {
4197                                 /* the extent was freed by someone */
4198                                 if (ref_path->lowest_level == level)
4199                                         goto out;
4200                                 btrfs_release_path(extent_root, path);
4201                                 goto walk_down;
4202                         }
4203                         leaf = path->nodes[0];
4204                 }
4205
4206                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4207                 if (found_key.objectid != bytenr ||
4208                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4209                         /* the extent was freed by someone */
4210                         if (ref_path->lowest_level == level) {
4211                                 ret = 1;
4212                                 goto out;
4213                         }
4214                         btrfs_release_path(extent_root, path);
4215                         goto walk_down;
4216                 }
4217 found:
4218                 ref = btrfs_item_ptr(leaf, path->slots[0],
4219                                 struct btrfs_extent_ref);
4220                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4221                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4222                         if (first_time) {
4223                                 level = (int)ref_objectid;
4224                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4225                                 ref_path->lowest_level = level;
4226                                 ref_path->current_level = level;
4227                                 ref_path->nodes[level] = bytenr;
4228                         } else {
4229                                 WARN_ON(ref_objectid != level);
4230                         }
4231                 } else {
4232                         WARN_ON(level != -1);
4233                 }
4234                 first_time = 0;
4235
4236                 if (ref_path->lowest_level == level) {
4237                         ref_path->owner_objectid = ref_objectid;
4238                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4239                 }
4240
4241                 /*
4242                  * the block is tree root or the block isn't in reference
4243                  * counted tree.
4244                  */
4245                 if (found_key.objectid == found_key.offset ||
4246                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4247                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4248                         ref_path->root_generation =
4249                                 btrfs_ref_generation(leaf, ref);
4250                         if (level < 0) {
4251                                 /* special reference from the tree log */
4252                                 ref_path->nodes[0] = found_key.offset;
4253                                 ref_path->current_level = 0;
4254                         }
4255                         ret = 0;
4256                         goto out;
4257                 }
4258
4259                 level++;
4260                 BUG_ON(ref_path->nodes[level] != 0);
4261                 ref_path->nodes[level] = found_key.offset;
4262                 ref_path->current_level = level;
4263
4264                 /*
4265                  * the reference was created in the running transaction,
4266                  * no need to continue walking up.
4267                  */
4268                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4269                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4270                         ref_path->root_generation =
4271                                 btrfs_ref_generation(leaf, ref);
4272                         ret = 0;
4273                         goto out;
4274                 }
4275
4276                 btrfs_release_path(extent_root, path);
4277                 cond_resched();
4278         }
4279         /* reached max tree level, but no tree root found. */
4280         BUG();
4281 out:
4282         btrfs_free_path(path);
4283         return ret;
4284 }
4285
4286 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4287                                 struct btrfs_root *extent_root,
4288                                 struct btrfs_ref_path *ref_path,
4289                                 u64 extent_start)
4290 {
4291         memset(ref_path, 0, sizeof(*ref_path));
4292         ref_path->extent_start = extent_start;
4293
4294         return __next_ref_path(trans, extent_root, ref_path, 1);
4295 }
4296
4297 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4298                                struct btrfs_root *extent_root,
4299                                struct btrfs_ref_path *ref_path)
4300 {
4301         return __next_ref_path(trans, extent_root, ref_path, 0);
4302 }
4303
4304 static noinline int get_new_locations(struct inode *reloc_inode,
4305                                       struct btrfs_key *extent_key,
4306                                       u64 offset, int no_fragment,
4307                                       struct disk_extent **extents,
4308                                       int *nr_extents)
4309 {
4310         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4311         struct btrfs_path *path;
4312         struct btrfs_file_extent_item *fi;
4313         struct extent_buffer *leaf;
4314         struct disk_extent *exts = *extents;
4315         struct btrfs_key found_key;
4316         u64 cur_pos;
4317         u64 last_byte;
4318         u32 nritems;
4319         int nr = 0;
4320         int max = *nr_extents;
4321         int ret;
4322
4323         WARN_ON(!no_fragment && *extents);
4324         if (!exts) {
4325                 max = 1;
4326                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4327                 if (!exts)
4328                         return -ENOMEM;
4329         }
4330
4331         path = btrfs_alloc_path();
4332         BUG_ON(!path);
4333
4334         cur_pos = extent_key->objectid - offset;
4335         last_byte = extent_key->objectid + extent_key->offset;
4336         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4337                                        cur_pos, 0);
4338         if (ret < 0)
4339                 goto out;
4340         if (ret > 0) {
4341                 ret = -ENOENT;
4342                 goto out;
4343         }
4344
4345         while (1) {
4346                 leaf = path->nodes[0];
4347                 nritems = btrfs_header_nritems(leaf);
4348                 if (path->slots[0] >= nritems) {
4349                         ret = btrfs_next_leaf(root, path);
4350                         if (ret < 0)
4351                                 goto out;
4352                         if (ret > 0)
4353                                 break;
4354                         leaf = path->nodes[0];
4355                 }
4356
4357                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4358                 if (found_key.offset != cur_pos ||
4359                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4360                     found_key.objectid != reloc_inode->i_ino)
4361                         break;
4362
4363                 fi = btrfs_item_ptr(leaf, path->slots[0],
4364                                     struct btrfs_file_extent_item);
4365                 if (btrfs_file_extent_type(leaf, fi) !=
4366                     BTRFS_FILE_EXTENT_REG ||
4367                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4368                         break;
4369
4370                 if (nr == max) {
4371                         struct disk_extent *old = exts;
4372                         max *= 2;
4373                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4374                         memcpy(exts, old, sizeof(*exts) * nr);
4375                         if (old != *extents)
4376                                 kfree(old);
4377                 }
4378
4379                 exts[nr].disk_bytenr =
4380                         btrfs_file_extent_disk_bytenr(leaf, fi);
4381                 exts[nr].disk_num_bytes =
4382                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4383                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4384                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4385                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4386                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4387                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4388                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4389                                                                            fi);
4390                 BUG_ON(exts[nr].offset > 0);
4391                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4392                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4393
4394                 cur_pos += exts[nr].num_bytes;
4395                 nr++;
4396
4397                 if (cur_pos + offset >= last_byte)
4398                         break;
4399
4400                 if (no_fragment) {
4401                         ret = 1;
4402                         goto out;
4403                 }
4404                 path->slots[0]++;
4405         }
4406
4407         BUG_ON(cur_pos + offset > last_byte);
4408         if (cur_pos + offset < last_byte) {
4409                 ret = -ENOENT;
4410                 goto out;
4411         }
4412         ret = 0;
4413 out:
4414         btrfs_free_path(path);
4415         if (ret) {
4416                 if (exts != *extents)
4417                         kfree(exts);
4418         } else {
4419                 *extents = exts;
4420                 *nr_extents = nr;
4421         }
4422         return ret;
4423 }
4424
4425 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4426                                         struct btrfs_root *root,
4427                                         struct btrfs_path *path,
4428                                         struct btrfs_key *extent_key,
4429                                         struct btrfs_key *leaf_key,
4430                                         struct btrfs_ref_path *ref_path,
4431                                         struct disk_extent *new_extents,
4432                                         int nr_extents)
4433 {
4434         struct extent_buffer *leaf;
4435         struct btrfs_file_extent_item *fi;
4436         struct inode *inode = NULL;
4437         struct btrfs_key key;
4438         u64 lock_start = 0;
4439         u64 lock_end = 0;
4440         u64 num_bytes;
4441         u64 ext_offset;
4442         u64 first_pos;
4443         u32 nritems;
4444         int nr_scaned = 0;
4445         int extent_locked = 0;
4446         int extent_type;
4447         int ret;
4448
4449         memcpy(&key, leaf_key, sizeof(key));
4450         first_pos = INT_LIMIT(loff_t) - extent_key->offset;
4451         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4452                 if (key.objectid < ref_path->owner_objectid ||
4453                     (key.objectid == ref_path->owner_objectid &&
4454                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4455                         key.objectid = ref_path->owner_objectid;
4456                         key.type = BTRFS_EXTENT_DATA_KEY;
4457                         key.offset = 0;
4458                 }
4459         }
4460
4461         while (1) {
4462                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4463                 if (ret < 0)
4464                         goto out;
4465
4466                 leaf = path->nodes[0];
4467                 nritems = btrfs_header_nritems(leaf);
4468 next:
4469                 if (extent_locked && ret > 0) {
4470                         /*
4471                          * the file extent item was modified by someone
4472                          * before the extent got locked.
4473                          */
4474                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4475                                       lock_end, GFP_NOFS);
4476                         extent_locked = 0;
4477                 }
4478
4479                 if (path->slots[0] >= nritems) {
4480                         if (++nr_scaned > 2)
4481                                 break;
4482
4483                         BUG_ON(extent_locked);
4484                         ret = btrfs_next_leaf(root, path);
4485                         if (ret < 0)
4486                                 goto out;
4487                         if (ret > 0)
4488                                 break;
4489                         leaf = path->nodes[0];
4490                         nritems = btrfs_header_nritems(leaf);
4491                 }
4492
4493                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4494
4495                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4496                         if ((key.objectid > ref_path->owner_objectid) ||
4497                             (key.objectid == ref_path->owner_objectid &&
4498                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4499                             (key.offset >= first_pos + extent_key->offset))
4500                                 break;
4501                 }
4502
4503                 if (inode && key.objectid != inode->i_ino) {
4504                         BUG_ON(extent_locked);
4505                         btrfs_release_path(root, path);
4506                         mutex_unlock(&inode->i_mutex);
4507                         iput(inode);
4508                         inode = NULL;
4509                         continue;
4510                 }
4511
4512                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4513                         path->slots[0]++;
4514                         ret = 1;
4515                         goto next;
4516                 }
4517                 fi = btrfs_item_ptr(leaf, path->slots[0],
4518                                     struct btrfs_file_extent_item);
4519                 extent_type = btrfs_file_extent_type(leaf, fi);
4520                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4521                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4522                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4523                      extent_key->objectid)) {
4524                         path->slots[0]++;
4525                         ret = 1;
4526                         goto next;
4527                 }
4528
4529                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4530                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4531
4532                 if (first_pos > key.offset - ext_offset)
4533                         first_pos = key.offset - ext_offset;
4534
4535                 if (!extent_locked) {
4536                         lock_start = key.offset;
4537                         lock_end = lock_start + num_bytes - 1;
4538                 } else {
4539                         if (lock_start > key.offset ||
4540                             lock_end + 1 < key.offset + num_bytes) {
4541                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4542                                               lock_start, lock_end, GFP_NOFS);
4543                                 extent_locked = 0;
4544                         }
4545                 }
4546
4547                 if (!inode) {
4548                         btrfs_release_path(root, path);
4549
4550                         inode = btrfs_iget_locked(root->fs_info->sb,
4551                                                   key.objectid, root);
4552                         if (inode->i_state & I_NEW) {
4553                                 BTRFS_I(inode)->root = root;
4554                                 BTRFS_I(inode)->location.objectid =
4555                                         key.objectid;
4556                                 BTRFS_I(inode)->location.type =
4557                                         BTRFS_INODE_ITEM_KEY;
4558                                 BTRFS_I(inode)->location.offset = 0;
4559                                 btrfs_read_locked_inode(inode);
4560                                 unlock_new_inode(inode);
4561                         }
4562                         /*
4563                          * some code call btrfs_commit_transaction while
4564                          * holding the i_mutex, so we can't use mutex_lock
4565                          * here.
4566                          */
4567                         if (is_bad_inode(inode) ||
4568                             !mutex_trylock(&inode->i_mutex)) {
4569                                 iput(inode);
4570                                 inode = NULL;
4571                                 key.offset = (u64)-1;
4572                                 goto skip;
4573                         }
4574                 }
4575
4576                 if (!extent_locked) {
4577                         struct btrfs_ordered_extent *ordered;
4578
4579                         btrfs_release_path(root, path);
4580
4581                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4582                                     lock_end, GFP_NOFS);
4583                         ordered = btrfs_lookup_first_ordered_extent(inode,
4584                                                                     lock_end);
4585                         if (ordered &&
4586                             ordered->file_offset <= lock_end &&
4587                             ordered->file_offset + ordered->len > lock_start) {
4588                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4589                                               lock_start, lock_end, GFP_NOFS);
4590                                 btrfs_start_ordered_extent(inode, ordered, 1);
4591                                 btrfs_put_ordered_extent(ordered);
4592                                 key.offset += num_bytes;
4593                                 goto skip;
4594                         }
4595                         if (ordered)
4596                                 btrfs_put_ordered_extent(ordered);
4597
4598                         extent_locked = 1;
4599                         continue;
4600                 }
4601
4602                 if (nr_extents == 1) {
4603                         /* update extent pointer in place */
4604                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4605                                                 new_extents[0].disk_bytenr);
4606                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4607                                                 new_extents[0].disk_num_bytes);
4608                         btrfs_mark_buffer_dirty(leaf);
4609
4610                         btrfs_drop_extent_cache(inode, key.offset,
4611                                                 key.offset + num_bytes - 1, 0);
4612
4613                         ret = btrfs_inc_extent_ref(trans, root,
4614                                                 new_extents[0].disk_bytenr,
4615                                                 new_extents[0].disk_num_bytes,
4616                                                 leaf->start,
4617                                                 root->root_key.objectid,
4618                                                 trans->transid,
4619                                                 key.objectid);
4620                         BUG_ON(ret);
4621
4622                         ret = btrfs_free_extent(trans, root,
4623                                                 extent_key->objectid,
4624                                                 extent_key->offset,
4625                                                 leaf->start,
4626                                                 btrfs_header_owner(leaf),
4627                                                 btrfs_header_generation(leaf),
4628                                                 key.objectid, 0);
4629                         BUG_ON(ret);
4630
4631                         btrfs_release_path(root, path);
4632                         key.offset += num_bytes;
4633                 } else {
4634                         BUG_ON(1);
4635 #if 0
4636                         u64 alloc_hint;
4637                         u64 extent_len;
4638                         int i;
4639                         /*
4640                          * drop old extent pointer at first, then insert the
4641                          * new pointers one bye one
4642                          */
4643                         btrfs_release_path(root, path);
4644                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4645                                                  key.offset + num_bytes,
4646                                                  key.offset, &alloc_hint);
4647                         BUG_ON(ret);
4648
4649                         for (i = 0; i < nr_extents; i++) {
4650                                 if (ext_offset >= new_extents[i].num_bytes) {
4651                                         ext_offset -= new_extents[i].num_bytes;
4652                                         continue;
4653                                 }
4654                                 extent_len = min(new_extents[i].num_bytes -
4655                                                  ext_offset, num_bytes);
4656
4657                                 ret = btrfs_insert_empty_item(trans, root,
4658                                                               path, &key,
4659                                                               sizeof(*fi));
4660                                 BUG_ON(ret);
4661
4662                                 leaf = path->nodes[0];
4663                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4664                                                 struct btrfs_file_extent_item);
4665                                 btrfs_set_file_extent_generation(leaf, fi,
4666                                                         trans->transid);
4667                                 btrfs_set_file_extent_type(leaf, fi,
4668                                                         BTRFS_FILE_EXTENT_REG);
4669                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4670                                                 new_extents[i].disk_bytenr);
4671                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4672                                                 new_extents[i].disk_num_bytes);
4673                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4674                                                 new_extents[i].ram_bytes);
4675
4676                                 btrfs_set_file_extent_compression(leaf, fi,
4677                                                 new_extents[i].compression);
4678                                 btrfs_set_file_extent_encryption(leaf, fi,
4679                                                 new_extents[i].encryption);
4680                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4681                                                 new_extents[i].other_encoding);
4682
4683                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4684                                                         extent_len);
4685                                 ext_offset += new_extents[i].offset;
4686                                 btrfs_set_file_extent_offset(leaf, fi,
4687                                                         ext_offset);
4688                                 btrfs_mark_buffer_dirty(leaf);
4689
4690                                 btrfs_drop_extent_cache(inode, key.offset,
4691                                                 key.offset + extent_len - 1, 0);
4692
4693                                 ret = btrfs_inc_extent_ref(trans, root,
4694                                                 new_extents[i].disk_bytenr,
4695                                                 new_extents[i].disk_num_bytes,
4696                                                 leaf->start,
4697                                                 root->root_key.objectid,
4698                                                 trans->transid, key.objectid);
4699                                 BUG_ON(ret);
4700                                 btrfs_release_path(root, path);
4701
4702                                 inode_add_bytes(inode, extent_len);
4703
4704                                 ext_offset = 0;
4705                                 num_bytes -= extent_len;
4706                                 key.offset += extent_len;
4707
4708                                 if (num_bytes == 0)
4709                                         break;
4710                         }
4711                         BUG_ON(i >= nr_extents);
4712 #endif
4713                 }
4714
4715                 if (extent_locked) {
4716                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4717                                       lock_end, GFP_NOFS);
4718                         extent_locked = 0;
4719                 }
4720 skip:
4721                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4722                     key.offset >= first_pos + extent_key->offset)
4723                         break;
4724
4725                 cond_resched();
4726         }
4727         ret = 0;
4728 out:
4729         btrfs_release_path(root, path);
4730         if (inode) {
4731                 mutex_unlock(&inode->i_mutex);
4732                 if (extent_locked) {
4733                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4734                                       lock_end, GFP_NOFS);
4735                 }
4736                 iput(inode);
4737         }
4738         return ret;
4739 }
4740
4741 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4742                                struct btrfs_root *root,
4743                                struct extent_buffer *buf, u64 orig_start)
4744 {
4745         int level;
4746         int ret;
4747
4748         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4749         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4750
4751         level = btrfs_header_level(buf);
4752         if (level == 0) {
4753                 struct btrfs_leaf_ref *ref;
4754                 struct btrfs_leaf_ref *orig_ref;
4755
4756                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4757                 if (!orig_ref)
4758                         return -ENOENT;
4759
4760                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4761                 if (!ref) {
4762                         btrfs_free_leaf_ref(root, orig_ref);
4763                         return -ENOMEM;
4764                 }
4765
4766                 ref->nritems = orig_ref->nritems;
4767                 memcpy(ref->extents, orig_ref->extents,
4768                         sizeof(ref->extents[0]) * ref->nritems);
4769
4770                 btrfs_free_leaf_ref(root, orig_ref);
4771
4772                 ref->root_gen = trans->transid;
4773                 ref->bytenr = buf->start;
4774                 ref->owner = btrfs_header_owner(buf);
4775                 ref->generation = btrfs_header_generation(buf);
4776                 ret = btrfs_add_leaf_ref(root, ref, 0);
4777                 WARN_ON(ret);
4778                 btrfs_free_leaf_ref(root, ref);
4779         }
4780         return 0;
4781 }
4782
4783 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4784                                         struct extent_buffer *leaf,
4785                                         struct btrfs_block_group_cache *group,
4786                                         struct btrfs_root *target_root)
4787 {
4788         struct btrfs_key key;
4789         struct inode *inode = NULL;
4790         struct btrfs_file_extent_item *fi;
4791         u64 num_bytes;
4792         u64 skip_objectid = 0;
4793         u32 nritems;
4794         u32 i;
4795
4796         nritems = btrfs_header_nritems(leaf);
4797         for (i = 0; i < nritems; i++) {
4798                 btrfs_item_key_to_cpu(leaf, &key, i);
4799                 if (key.objectid == skip_objectid ||
4800                     key.type != BTRFS_EXTENT_DATA_KEY)
4801                         continue;
4802                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4803                 if (btrfs_file_extent_type(leaf, fi) ==
4804                     BTRFS_FILE_EXTENT_INLINE)
4805                         continue;
4806                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4807                         continue;
4808                 if (!inode || inode->i_ino != key.objectid) {
4809                         iput(inode);
4810                         inode = btrfs_ilookup(target_root->fs_info->sb,
4811                                               key.objectid, target_root, 1);
4812                 }
4813                 if (!inode) {
4814                         skip_objectid = key.objectid;
4815                         continue;
4816                 }
4817                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4818
4819                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4820                             key.offset + num_bytes - 1, GFP_NOFS);
4821                 btrfs_drop_extent_cache(inode, key.offset,
4822                                         key.offset + num_bytes - 1, 1);
4823                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4824                               key.offset + num_bytes - 1, GFP_NOFS);
4825                 cond_resched();
4826         }
4827         iput(inode);
4828         return 0;
4829 }
4830
4831 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4832                                         struct btrfs_root *root,
4833                                         struct extent_buffer *leaf,
4834                                         struct btrfs_block_group_cache *group,
4835                                         struct inode *reloc_inode)
4836 {
4837         struct btrfs_key key;
4838         struct btrfs_key extent_key;
4839         struct btrfs_file_extent_item *fi;
4840         struct btrfs_leaf_ref *ref;
4841         struct disk_extent *new_extent;
4842         u64 bytenr;
4843         u64 num_bytes;
4844         u32 nritems;
4845         u32 i;
4846         int ext_index;
4847         int nr_extent;
4848         int ret;
4849
4850         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4851         BUG_ON(!new_extent);
4852
4853         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4854         BUG_ON(!ref);
4855
4856         ext_index = -1;
4857         nritems = btrfs_header_nritems(leaf);
4858         for (i = 0; i < nritems; i++) {
4859                 btrfs_item_key_to_cpu(leaf, &key, i);
4860                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4861                         continue;
4862                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4863                 if (btrfs_file_extent_type(leaf, fi) ==
4864                     BTRFS_FILE_EXTENT_INLINE)
4865                         continue;
4866                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4867                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4868                 if (bytenr == 0)
4869                         continue;
4870
4871                 ext_index++;
4872                 if (bytenr >= group->key.objectid + group->key.offset ||
4873                     bytenr + num_bytes <= group->key.objectid)
4874                         continue;
4875
4876                 extent_key.objectid = bytenr;
4877                 extent_key.offset = num_bytes;
4878                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4879                 nr_extent = 1;
4880                 ret = get_new_locations(reloc_inode, &extent_key,
4881                                         group->key.objectid, 1,
4882                                         &new_extent, &nr_extent);
4883                 if (ret > 0)
4884                         continue;
4885                 BUG_ON(ret < 0);
4886
4887                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4888                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4889                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4890                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4891
4892                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4893                                                 new_extent->disk_bytenr);
4894                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4895                                                 new_extent->disk_num_bytes);
4896                 btrfs_mark_buffer_dirty(leaf);
4897
4898                 ret = btrfs_inc_extent_ref(trans, root,
4899                                         new_extent->disk_bytenr,
4900                                         new_extent->disk_num_bytes,
4901                                         leaf->start,
4902                                         root->root_key.objectid,
4903                                         trans->transid, key.objectid);
4904                 BUG_ON(ret);
4905                 ret = btrfs_free_extent(trans, root,
4906                                         bytenr, num_bytes, leaf->start,
4907                                         btrfs_header_owner(leaf),
4908                                         btrfs_header_generation(leaf),
4909                                         key.objectid, 0);
4910                 BUG_ON(ret);
4911                 cond_resched();
4912         }
4913         kfree(new_extent);
4914         BUG_ON(ext_index + 1 != ref->nritems);
4915         btrfs_free_leaf_ref(root, ref);
4916         return 0;
4917 }
4918
4919 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4920                           struct btrfs_root *root)
4921 {
4922         struct btrfs_root *reloc_root;
4923         int ret;
4924
4925         if (root->reloc_root) {
4926                 reloc_root = root->reloc_root;
4927                 root->reloc_root = NULL;
4928                 list_add(&reloc_root->dead_list,
4929                          &root->fs_info->dead_reloc_roots);
4930
4931                 btrfs_set_root_bytenr(&reloc_root->root_item,
4932                                       reloc_root->node->start);
4933                 btrfs_set_root_level(&root->root_item,
4934                                      btrfs_header_level(reloc_root->node));
4935                 memset(&reloc_root->root_item.drop_progress, 0,
4936                         sizeof(struct btrfs_disk_key));
4937                 reloc_root->root_item.drop_level = 0;
4938
4939                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4940                                         &reloc_root->root_key,
4941                                         &reloc_root->root_item);
4942                 BUG_ON(ret);
4943         }
4944         return 0;
4945 }
4946
4947 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4948 {
4949         struct btrfs_trans_handle *trans;
4950         struct btrfs_root *reloc_root;
4951         struct btrfs_root *prev_root = NULL;
4952         struct list_head dead_roots;
4953         int ret;
4954         unsigned long nr;
4955
4956         INIT_LIST_HEAD(&dead_roots);
4957         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4958
4959         while (!list_empty(&dead_roots)) {
4960                 reloc_root = list_entry(dead_roots.prev,
4961                                         struct btrfs_root, dead_list);
4962                 list_del_init(&reloc_root->dead_list);
4963
4964                 BUG_ON(reloc_root->commit_root != NULL);
4965                 while (1) {
4966                         trans = btrfs_join_transaction(root, 1);
4967                         BUG_ON(!trans);
4968
4969                         mutex_lock(&root->fs_info->drop_mutex);
4970                         ret = btrfs_drop_snapshot(trans, reloc_root);
4971                         if (ret != -EAGAIN)
4972                                 break;
4973                         mutex_unlock(&root->fs_info->drop_mutex);
4974
4975                         nr = trans->blocks_used;
4976                         ret = btrfs_end_transaction(trans, root);
4977                         BUG_ON(ret);
4978                         btrfs_btree_balance_dirty(root, nr);
4979                 }
4980
4981                 free_extent_buffer(reloc_root->node);
4982
4983                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4984                                      &reloc_root->root_key);
4985                 BUG_ON(ret);
4986                 mutex_unlock(&root->fs_info->drop_mutex);
4987
4988                 nr = trans->blocks_used;
4989                 ret = btrfs_end_transaction(trans, root);
4990                 BUG_ON(ret);
4991                 btrfs_btree_balance_dirty(root, nr);
4992
4993                 kfree(prev_root);
4994                 prev_root = reloc_root;
4995         }
4996         if (prev_root) {
4997                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
4998                 kfree(prev_root);
4999         }
5000         return 0;
5001 }
5002
5003 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5004 {
5005         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5006         return 0;
5007 }
5008
5009 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5010 {
5011         struct btrfs_root *reloc_root;
5012         struct btrfs_trans_handle *trans;
5013         struct btrfs_key location;
5014         int found;
5015         int ret;
5016
5017         mutex_lock(&root->fs_info->tree_reloc_mutex);
5018         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5019         BUG_ON(ret);
5020         found = !list_empty(&root->fs_info->dead_reloc_roots);
5021         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5022
5023         if (found) {
5024                 trans = btrfs_start_transaction(root, 1);
5025                 BUG_ON(!trans);
5026                 ret = btrfs_commit_transaction(trans, root);
5027                 BUG_ON(ret);
5028         }
5029
5030         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5031         location.offset = (u64)-1;
5032         location.type = BTRFS_ROOT_ITEM_KEY;
5033
5034         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5035         BUG_ON(!reloc_root);
5036         btrfs_orphan_cleanup(reloc_root);
5037         return 0;
5038 }
5039
5040 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5041                                     struct btrfs_root *root)
5042 {
5043         struct btrfs_root *reloc_root;
5044         struct extent_buffer *eb;
5045         struct btrfs_root_item *root_item;
5046         struct btrfs_key root_key;
5047         int ret;
5048
5049         BUG_ON(!root->ref_cows);
5050         if (root->reloc_root)
5051                 return 0;
5052
5053         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5054         BUG_ON(!root_item);
5055
5056         ret = btrfs_copy_root(trans, root, root->commit_root,
5057                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5058         BUG_ON(ret);
5059
5060         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5061         root_key.offset = root->root_key.objectid;
5062         root_key.type = BTRFS_ROOT_ITEM_KEY;
5063
5064         memcpy(root_item, &root->root_item, sizeof(root_item));
5065         btrfs_set_root_refs(root_item, 0);
5066         btrfs_set_root_bytenr(root_item, eb->start);
5067         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5068         btrfs_set_root_generation(root_item, trans->transid);
5069
5070         btrfs_tree_unlock(eb);
5071         free_extent_buffer(eb);
5072
5073         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5074                                 &root_key, root_item);
5075         BUG_ON(ret);
5076         kfree(root_item);
5077
5078         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5079                                                  &root_key);
5080         BUG_ON(!reloc_root);
5081         reloc_root->last_trans = trans->transid;
5082         reloc_root->commit_root = NULL;
5083         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5084
5085         root->reloc_root = reloc_root;
5086         return 0;
5087 }
5088
5089 /*
5090  * Core function of space balance.
5091  *
5092  * The idea is using reloc trees to relocate tree blocks in reference
5093  * counted roots. There is one reloc tree for each subvol, and all
5094  * reloc trees share same root key objectid. Reloc trees are snapshots
5095  * of the latest committed roots of subvols (root->commit_root).
5096  *
5097  * To relocate a tree block referenced by a subvol, there are two steps.
5098  * COW the block through subvol's reloc tree, then update block pointer
5099  * in the subvol to point to the new block. Since all reloc trees share
5100  * same root key objectid, doing special handing for tree blocks owned
5101  * by them is easy. Once a tree block has been COWed in one reloc tree,
5102  * we can use the resulting new block directly when the same block is
5103  * required to COW again through other reloc trees. By this way, relocated
5104  * tree blocks are shared between reloc trees, so they are also shared
5105  * between subvols.
5106  */
5107 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5108                                       struct btrfs_root *root,
5109                                       struct btrfs_path *path,
5110                                       struct btrfs_key *first_key,
5111                                       struct btrfs_ref_path *ref_path,
5112                                       struct btrfs_block_group_cache *group,
5113                                       struct inode *reloc_inode)
5114 {
5115         struct btrfs_root *reloc_root;
5116         struct extent_buffer *eb = NULL;
5117         struct btrfs_key *keys;
5118         u64 *nodes;
5119         int level;
5120         int shared_level;
5121         int lowest_level = 0;
5122         int ret;
5123
5124         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5125                 lowest_level = ref_path->owner_objectid;
5126
5127         if (!root->ref_cows) {
5128                 path->lowest_level = lowest_level;
5129                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5130                 BUG_ON(ret < 0);
5131                 path->lowest_level = 0;
5132                 btrfs_release_path(root, path);
5133                 return 0;
5134         }
5135
5136         mutex_lock(&root->fs_info->tree_reloc_mutex);
5137         ret = init_reloc_tree(trans, root);
5138         BUG_ON(ret);
5139         reloc_root = root->reloc_root;
5140
5141         shared_level = ref_path->shared_level;
5142         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5143
5144         keys = ref_path->node_keys;
5145         nodes = ref_path->new_nodes;
5146         memset(&keys[shared_level + 1], 0,
5147                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5148         memset(&nodes[shared_level + 1], 0,
5149                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5150
5151         if (nodes[lowest_level] == 0) {
5152                 path->lowest_level = lowest_level;
5153                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5154                                         0, 1);
5155                 BUG_ON(ret);
5156                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5157                         eb = path->nodes[level];
5158                         if (!eb || eb == reloc_root->node)
5159                                 break;
5160                         nodes[level] = eb->start;
5161                         if (level == 0)
5162                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5163                         else
5164                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5165                 }
5166                 if (nodes[0] &&
5167                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5168                         eb = path->nodes[0];
5169                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5170                                                       group, reloc_inode);
5171                         BUG_ON(ret);
5172                 }
5173                 btrfs_release_path(reloc_root, path);
5174         } else {
5175                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5176                                        lowest_level);
5177                 BUG_ON(ret);
5178         }
5179
5180         /*
5181          * replace tree blocks in the fs tree with tree blocks in
5182          * the reloc tree.
5183          */
5184         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5185         BUG_ON(ret < 0);
5186
5187         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5188                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5189                                         0, 0);
5190                 BUG_ON(ret);
5191                 extent_buffer_get(path->nodes[0]);
5192                 eb = path->nodes[0];
5193                 btrfs_release_path(reloc_root, path);
5194                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5195                 BUG_ON(ret);
5196                 free_extent_buffer(eb);
5197         }
5198
5199         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5200         path->lowest_level = 0;
5201         return 0;
5202 }
5203
5204 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5205                                         struct btrfs_root *root,
5206                                         struct btrfs_path *path,
5207                                         struct btrfs_key *first_key,
5208                                         struct btrfs_ref_path *ref_path)
5209 {
5210         int ret;
5211
5212         ret = relocate_one_path(trans, root, path, first_key,
5213                                 ref_path, NULL, NULL);
5214         BUG_ON(ret);
5215
5216         if (root == root->fs_info->extent_root)
5217                 btrfs_extent_post_op(trans, root);
5218
5219         return 0;
5220 }
5221
5222 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5223                                     struct btrfs_root *extent_root,
5224                                     struct btrfs_path *path,
5225                                     struct btrfs_key *extent_key)
5226 {
5227         int ret;
5228
5229         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5230         if (ret)
5231                 goto out;
5232         ret = btrfs_del_item(trans, extent_root, path);
5233 out:
5234         btrfs_release_path(extent_root, path);
5235         return ret;
5236 }
5237
5238 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5239                                                 struct btrfs_ref_path *ref_path)
5240 {
5241         struct btrfs_key root_key;
5242
5243         root_key.objectid = ref_path->root_objectid;
5244         root_key.type = BTRFS_ROOT_ITEM_KEY;
5245         if (is_cowonly_root(ref_path->root_objectid))
5246                 root_key.offset = 0;
5247         else
5248                 root_key.offset = (u64)-1;
5249
5250         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5251 }
5252
5253 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5254                                         struct btrfs_path *path,
5255                                         struct btrfs_key *extent_key,
5256                                         struct btrfs_block_group_cache *group,
5257                                         struct inode *reloc_inode, int pass)
5258 {
5259         struct btrfs_trans_handle *trans;
5260         struct btrfs_root *found_root;
5261         struct btrfs_ref_path *ref_path = NULL;
5262         struct disk_extent *new_extents = NULL;
5263         int nr_extents = 0;
5264         int loops;
5265         int ret;
5266         int level;
5267         struct btrfs_key first_key;
5268         u64 prev_block = 0;
5269
5270
5271         trans = btrfs_start_transaction(extent_root, 1);
5272         BUG_ON(!trans);
5273
5274         if (extent_key->objectid == 0) {
5275                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5276                 goto out;
5277         }
5278
5279         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5280         if (!ref_path) {
5281                 ret = -ENOMEM;
5282                 goto out;
5283         }
5284
5285         for (loops = 0; ; loops++) {
5286                 if (loops == 0) {
5287                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5288                                                    extent_key->objectid);
5289                 } else {
5290                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5291                 }
5292                 if (ret < 0)
5293                         goto out;
5294                 if (ret > 0)
5295                         break;
5296
5297                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5298                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5299                         continue;
5300
5301                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5302                 BUG_ON(!found_root);
5303                 /*
5304                  * for reference counted tree, only process reference paths
5305                  * rooted at the latest committed root.
5306                  */
5307                 if (found_root->ref_cows &&
5308                     ref_path->root_generation != found_root->root_key.offset)
5309                         continue;
5310
5311                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5312                         if (pass == 0) {
5313                                 /*
5314                                  * copy data extents to new locations
5315                                  */
5316                                 u64 group_start = group->key.objectid;
5317                                 ret = relocate_data_extent(reloc_inode,
5318                                                            extent_key,
5319                                                            group_start);
5320                                 if (ret < 0)
5321                                         goto out;
5322                                 break;
5323                         }
5324                         level = 0;
5325                 } else {
5326                         level = ref_path->owner_objectid;
5327                 }
5328
5329                 if (prev_block != ref_path->nodes[level]) {
5330                         struct extent_buffer *eb;
5331                         u64 block_start = ref_path->nodes[level];
5332                         u64 block_size = btrfs_level_size(found_root, level);
5333
5334                         eb = read_tree_block(found_root, block_start,
5335                                              block_size, 0);
5336                         btrfs_tree_lock(eb);
5337                         BUG_ON(level != btrfs_header_level(eb));
5338
5339                         if (level == 0)
5340                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5341                         else
5342                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5343
5344                         btrfs_tree_unlock(eb);
5345                         free_extent_buffer(eb);
5346                         prev_block = block_start;
5347                 }
5348
5349                 btrfs_record_root_in_trans(found_root);
5350                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5351                         /*
5352                          * try to update data extent references while
5353                          * keeping metadata shared between snapshots.
5354                          */
5355                         if (pass == 1) {
5356                                 ret = relocate_one_path(trans, found_root,
5357                                                 path, &first_key, ref_path,
5358                                                 group, reloc_inode);
5359                                 if (ret < 0)
5360                                         goto out;
5361                                 continue;
5362                         }
5363                         /*
5364                          * use fallback method to process the remaining
5365                          * references.
5366                          */
5367                         if (!new_extents) {
5368                                 u64 group_start = group->key.objectid;
5369                                 new_extents = kmalloc(sizeof(*new_extents),
5370                                                       GFP_NOFS);
5371                                 nr_extents = 1;
5372                                 ret = get_new_locations(reloc_inode,
5373                                                         extent_key,
5374                                                         group_start, 1,
5375                                                         &new_extents,
5376                                                         &nr_extents);
5377                                 if (ret)
5378                                         goto out;
5379                         }
5380                         ret = replace_one_extent(trans, found_root,
5381                                                 path, extent_key,
5382                                                 &first_key, ref_path,
5383                                                 new_extents, nr_extents);
5384                 } else {
5385                         ret = relocate_tree_block(trans, found_root, path,
5386                                                   &first_key, ref_path);
5387                 }
5388                 if (ret < 0)
5389                         goto out;
5390         }
5391         ret = 0;
5392 out:
5393         btrfs_end_transaction(trans, extent_root);
5394         kfree(new_extents);
5395         kfree(ref_path);
5396         return ret;
5397 }
5398
5399 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5400 {
5401         u64 num_devices;
5402         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5403                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5404
5405         num_devices = root->fs_info->fs_devices->rw_devices;
5406         if (num_devices == 1) {
5407                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5408                 stripped = flags & ~stripped;
5409
5410                 /* turn raid0 into single device chunks */
5411                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5412                         return stripped;
5413
5414                 /* turn mirroring into duplication */
5415                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5416                              BTRFS_BLOCK_GROUP_RAID10))
5417                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5418                 return flags;
5419         } else {
5420                 /* they already had raid on here, just return */
5421                 if (flags & stripped)
5422                         return flags;
5423
5424                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5425                 stripped = flags & ~stripped;
5426
5427                 /* switch duplicated blocks with raid1 */
5428                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5429                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5430
5431                 /* turn single device chunks into raid0 */
5432                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5433         }
5434         return flags;
5435 }
5436
5437 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5438                      struct btrfs_block_group_cache *shrink_block_group,
5439                      int force)
5440 {
5441         struct btrfs_trans_handle *trans;
5442         u64 new_alloc_flags;
5443         u64 calc;
5444
5445         spin_lock(&shrink_block_group->lock);
5446         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5447                 spin_unlock(&shrink_block_group->lock);
5448
5449                 trans = btrfs_start_transaction(root, 1);
5450                 spin_lock(&shrink_block_group->lock);
5451
5452                 new_alloc_flags = update_block_group_flags(root,
5453                                                    shrink_block_group->flags);
5454                 if (new_alloc_flags != shrink_block_group->flags) {
5455                         calc =
5456                              btrfs_block_group_used(&shrink_block_group->item);
5457                 } else {
5458                         calc = shrink_block_group->key.offset;
5459                 }
5460                 spin_unlock(&shrink_block_group->lock);
5461
5462                 do_chunk_alloc(trans, root->fs_info->extent_root,
5463                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5464
5465                 btrfs_end_transaction(trans, root);
5466         } else
5467                 spin_unlock(&shrink_block_group->lock);
5468         return 0;
5469 }
5470
5471 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5472                                  struct btrfs_root *root,
5473                                  u64 objectid, u64 size)
5474 {
5475         struct btrfs_path *path;
5476         struct btrfs_inode_item *item;
5477         struct extent_buffer *leaf;
5478         int ret;
5479
5480         path = btrfs_alloc_path();
5481         if (!path)
5482                 return -ENOMEM;
5483
5484         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5485         if (ret)
5486                 goto out;
5487
5488         leaf = path->nodes[0];
5489         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5490         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5491         btrfs_set_inode_generation(leaf, item, 1);
5492         btrfs_set_inode_size(leaf, item, size);
5493         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5494         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5495         btrfs_mark_buffer_dirty(leaf);
5496         btrfs_release_path(root, path);
5497 out:
5498         btrfs_free_path(path);
5499         return ret;
5500 }
5501
5502 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5503                                         struct btrfs_block_group_cache *group)
5504 {
5505         struct inode *inode = NULL;
5506         struct btrfs_trans_handle *trans;
5507         struct btrfs_root *root;
5508         struct btrfs_key root_key;
5509         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5510         int err = 0;
5511
5512         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5513         root_key.type = BTRFS_ROOT_ITEM_KEY;
5514         root_key.offset = (u64)-1;
5515         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5516         if (IS_ERR(root))
5517                 return ERR_CAST(root);
5518
5519         trans = btrfs_start_transaction(root, 1);
5520         BUG_ON(!trans);
5521
5522         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5523         if (err)
5524                 goto out;
5525
5526         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5527         BUG_ON(err);
5528
5529         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5530                                        group->key.offset, 0, group->key.offset,
5531                                        0, 0, 0);
5532         BUG_ON(err);
5533
5534         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5535         if (inode->i_state & I_NEW) {
5536                 BTRFS_I(inode)->root = root;
5537                 BTRFS_I(inode)->location.objectid = objectid;
5538                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5539                 BTRFS_I(inode)->location.offset = 0;
5540                 btrfs_read_locked_inode(inode);
5541                 unlock_new_inode(inode);
5542                 BUG_ON(is_bad_inode(inode));
5543         } else {
5544                 BUG_ON(1);
5545         }
5546         BTRFS_I(inode)->index_cnt = group->key.objectid;
5547
5548         err = btrfs_orphan_add(trans, inode);
5549 out:
5550         btrfs_end_transaction(trans, root);
5551         if (err) {
5552                 if (inode)
5553                         iput(inode);
5554                 inode = ERR_PTR(err);
5555         }
5556         return inode;
5557 }
5558
5559 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5560 {
5561
5562         struct btrfs_ordered_sum *sums;
5563         struct btrfs_sector_sum *sector_sum;
5564         struct btrfs_ordered_extent *ordered;
5565         struct btrfs_root *root = BTRFS_I(inode)->root;
5566         struct list_head list;
5567         size_t offset;
5568         int ret;
5569         u64 disk_bytenr;
5570
5571         INIT_LIST_HEAD(&list);
5572
5573         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5574         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5575
5576         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5577         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5578                                        disk_bytenr + len - 1, &list);
5579
5580         while (!list_empty(&list)) {
5581                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5582                 list_del_init(&sums->list);
5583
5584                 sector_sum = sums->sums;
5585                 sums->bytenr = ordered->start;
5586
5587                 offset = 0;
5588                 while (offset < sums->len) {
5589                         sector_sum->bytenr += ordered->start - disk_bytenr;
5590                         sector_sum++;
5591                         offset += root->sectorsize;
5592                 }
5593
5594                 btrfs_add_ordered_sum(inode, ordered, sums);
5595         }
5596         btrfs_put_ordered_extent(ordered);
5597         return 0;
5598 }
5599
5600 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5601 {
5602         struct btrfs_trans_handle *trans;
5603         struct btrfs_path *path;
5604         struct btrfs_fs_info *info = root->fs_info;
5605         struct extent_buffer *leaf;
5606         struct inode *reloc_inode;
5607         struct btrfs_block_group_cache *block_group;
5608         struct btrfs_key key;
5609         u64 skipped;
5610         u64 cur_byte;
5611         u64 total_found;
5612         u32 nritems;
5613         int ret;
5614         int progress;
5615         int pass = 0;
5616
5617         root = root->fs_info->extent_root;
5618
5619         block_group = btrfs_lookup_block_group(info, group_start);
5620         BUG_ON(!block_group);
5621
5622         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5623                (unsigned long long)block_group->key.objectid,
5624                (unsigned long long)block_group->flags);
5625
5626         path = btrfs_alloc_path();
5627         BUG_ON(!path);
5628
5629         reloc_inode = create_reloc_inode(info, block_group);
5630         BUG_ON(IS_ERR(reloc_inode));
5631
5632         __alloc_chunk_for_shrink(root, block_group, 1);
5633         set_block_group_readonly(block_group);
5634
5635         btrfs_start_delalloc_inodes(info->tree_root);
5636         btrfs_wait_ordered_extents(info->tree_root, 0);
5637 again:
5638         skipped = 0;
5639         total_found = 0;
5640         progress = 0;
5641         key.objectid = block_group->key.objectid;
5642         key.offset = 0;
5643         key.type = 0;
5644         cur_byte = key.objectid;
5645
5646         trans = btrfs_start_transaction(info->tree_root, 1);
5647         btrfs_commit_transaction(trans, info->tree_root);
5648
5649         mutex_lock(&root->fs_info->cleaner_mutex);
5650         btrfs_clean_old_snapshots(info->tree_root);
5651         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5652         mutex_unlock(&root->fs_info->cleaner_mutex);
5653
5654         while (1) {
5655                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5656                 if (ret < 0)
5657                         goto out;
5658 next:
5659                 leaf = path->nodes[0];
5660                 nritems = btrfs_header_nritems(leaf);
5661                 if (path->slots[0] >= nritems) {
5662                         ret = btrfs_next_leaf(root, path);
5663                         if (ret < 0)
5664                                 goto out;
5665                         if (ret == 1) {
5666                                 ret = 0;
5667                                 break;
5668                         }
5669                         leaf = path->nodes[0];
5670                         nritems = btrfs_header_nritems(leaf);
5671                 }
5672
5673                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5674
5675                 if (key.objectid >= block_group->key.objectid +
5676                     block_group->key.offset)
5677                         break;
5678
5679                 if (progress && need_resched()) {
5680                         btrfs_release_path(root, path);
5681                         cond_resched();
5682                         progress = 0;
5683                         continue;
5684                 }
5685                 progress = 1;
5686
5687                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5688                     key.objectid + key.offset <= cur_byte) {
5689                         path->slots[0]++;
5690                         goto next;
5691                 }
5692
5693                 total_found++;
5694                 cur_byte = key.objectid + key.offset;
5695                 btrfs_release_path(root, path);
5696
5697                 __alloc_chunk_for_shrink(root, block_group, 0);
5698                 ret = relocate_one_extent(root, path, &key, block_group,
5699                                           reloc_inode, pass);
5700                 BUG_ON(ret < 0);
5701                 if (ret > 0)
5702                         skipped++;
5703
5704                 key.objectid = cur_byte;
5705                 key.type = 0;
5706                 key.offset = 0;
5707         }
5708
5709         btrfs_release_path(root, path);
5710
5711         if (pass == 0) {
5712                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5713                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5714         }
5715
5716         if (total_found > 0) {
5717                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5718                        (unsigned long long)total_found, pass);
5719                 pass++;
5720                 if (total_found == skipped && pass > 2) {
5721                         iput(reloc_inode);
5722                         reloc_inode = create_reloc_inode(info, block_group);
5723                         pass = 0;
5724                 }
5725                 goto again;
5726         }
5727
5728         /* delete reloc_inode */
5729         iput(reloc_inode);
5730
5731         /* unpin extents in this range */
5732         trans = btrfs_start_transaction(info->tree_root, 1);
5733         btrfs_commit_transaction(trans, info->tree_root);
5734
5735         spin_lock(&block_group->lock);
5736         WARN_ON(block_group->pinned > 0);
5737         WARN_ON(block_group->reserved > 0);
5738         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5739         spin_unlock(&block_group->lock);
5740         put_block_group(block_group);
5741         ret = 0;
5742 out:
5743         btrfs_free_path(path);
5744         return ret;
5745 }
5746
5747 static int find_first_block_group(struct btrfs_root *root,
5748                 struct btrfs_path *path, struct btrfs_key *key)
5749 {
5750         int ret = 0;
5751         struct btrfs_key found_key;
5752         struct extent_buffer *leaf;
5753         int slot;
5754
5755         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5756         if (ret < 0)
5757                 goto out;
5758
5759         while (1) {
5760                 slot = path->slots[0];
5761                 leaf = path->nodes[0];
5762                 if (slot >= btrfs_header_nritems(leaf)) {
5763                         ret = btrfs_next_leaf(root, path);
5764                         if (ret == 0)
5765                                 continue;
5766                         if (ret < 0)
5767                                 goto out;
5768                         break;
5769                 }
5770                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5771
5772                 if (found_key.objectid >= key->objectid &&
5773                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5774                         ret = 0;
5775                         goto out;
5776                 }
5777                 path->slots[0]++;
5778         }
5779         ret = -ENOENT;
5780 out:
5781         return ret;
5782 }
5783
5784 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5785 {
5786         struct btrfs_block_group_cache *block_group;
5787         struct rb_node *n;
5788
5789         spin_lock(&info->block_group_cache_lock);
5790         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5791                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5792                                        cache_node);
5793                 rb_erase(&block_group->cache_node,
5794                          &info->block_group_cache_tree);
5795                 spin_unlock(&info->block_group_cache_lock);
5796
5797                 btrfs_remove_free_space_cache(block_group);
5798                 down_write(&block_group->space_info->groups_sem);
5799                 list_del(&block_group->list);
5800                 up_write(&block_group->space_info->groups_sem);
5801
5802                 WARN_ON(atomic_read(&block_group->count) != 1);
5803                 kfree(block_group);
5804
5805                 spin_lock(&info->block_group_cache_lock);
5806         }
5807         spin_unlock(&info->block_group_cache_lock);
5808         return 0;
5809 }
5810
5811 int btrfs_read_block_groups(struct btrfs_root *root)
5812 {
5813         struct btrfs_path *path;
5814         int ret;
5815         struct btrfs_block_group_cache *cache;
5816         struct btrfs_fs_info *info = root->fs_info;
5817         struct btrfs_space_info *space_info;
5818         struct btrfs_key key;
5819         struct btrfs_key found_key;
5820         struct extent_buffer *leaf;
5821
5822         root = info->extent_root;
5823         key.objectid = 0;
5824         key.offset = 0;
5825         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5826         path = btrfs_alloc_path();
5827         if (!path)
5828                 return -ENOMEM;
5829
5830         while (1) {
5831                 ret = find_first_block_group(root, path, &key);
5832                 if (ret > 0) {
5833                         ret = 0;
5834                         goto error;
5835                 }
5836                 if (ret != 0)
5837                         goto error;
5838
5839                 leaf = path->nodes[0];
5840                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5841                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5842                 if (!cache) {
5843                         ret = -ENOMEM;
5844                         break;
5845                 }
5846
5847                 atomic_set(&cache->count, 1);
5848                 spin_lock_init(&cache->lock);
5849                 mutex_init(&cache->alloc_mutex);
5850                 mutex_init(&cache->cache_mutex);
5851                 INIT_LIST_HEAD(&cache->list);
5852                 read_extent_buffer(leaf, &cache->item,
5853                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5854                                    sizeof(cache->item));
5855                 memcpy(&cache->key, &found_key, sizeof(found_key));
5856
5857                 key.objectid = found_key.objectid + found_key.offset;
5858                 btrfs_release_path(root, path);
5859                 cache->flags = btrfs_block_group_flags(&cache->item);
5860
5861                 ret = update_space_info(info, cache->flags, found_key.offset,
5862                                         btrfs_block_group_used(&cache->item),
5863                                         &space_info);
5864                 BUG_ON(ret);
5865                 cache->space_info = space_info;
5866                 down_write(&space_info->groups_sem);
5867                 list_add_tail(&cache->list, &space_info->block_groups);
5868                 up_write(&space_info->groups_sem);
5869
5870                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5871                 BUG_ON(ret);
5872
5873                 set_avail_alloc_bits(root->fs_info, cache->flags);
5874                 if (btrfs_chunk_readonly(root, cache->key.objectid))
5875                         set_block_group_readonly(cache);
5876         }
5877         ret = 0;
5878 error:
5879         btrfs_free_path(path);
5880         return ret;
5881 }
5882
5883 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5884                            struct btrfs_root *root, u64 bytes_used,
5885                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5886                            u64 size)
5887 {
5888         int ret;
5889         struct btrfs_root *extent_root;
5890         struct btrfs_block_group_cache *cache;
5891
5892         extent_root = root->fs_info->extent_root;
5893
5894         root->fs_info->last_trans_new_blockgroup = trans->transid;
5895
5896         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5897         if (!cache)
5898                 return -ENOMEM;
5899
5900         cache->key.objectid = chunk_offset;
5901         cache->key.offset = size;
5902         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5903         atomic_set(&cache->count, 1);
5904         spin_lock_init(&cache->lock);
5905         mutex_init(&cache->alloc_mutex);
5906         mutex_init(&cache->cache_mutex);
5907         INIT_LIST_HEAD(&cache->list);
5908
5909         btrfs_set_block_group_used(&cache->item, bytes_used);
5910         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5911         cache->flags = type;
5912         btrfs_set_block_group_flags(&cache->item, type);
5913
5914         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5915                                 &cache->space_info);
5916         BUG_ON(ret);
5917         down_write(&cache->space_info->groups_sem);
5918         list_add_tail(&cache->list, &cache->space_info->block_groups);
5919         up_write(&cache->space_info->groups_sem);
5920
5921         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5922         BUG_ON(ret);
5923
5924         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5925                                 sizeof(cache->item));
5926         BUG_ON(ret);
5927
5928         finish_current_insert(trans, extent_root, 0);
5929         ret = del_pending_extents(trans, extent_root, 0);
5930         BUG_ON(ret);
5931         set_avail_alloc_bits(extent_root->fs_info, type);
5932
5933         return 0;
5934 }
5935
5936 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5937                              struct btrfs_root *root, u64 group_start)
5938 {
5939         struct btrfs_path *path;
5940         struct btrfs_block_group_cache *block_group;
5941         struct btrfs_key key;
5942         int ret;
5943
5944         root = root->fs_info->extent_root;
5945
5946         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5947         BUG_ON(!block_group);
5948         BUG_ON(!block_group->ro);
5949
5950         memcpy(&key, &block_group->key, sizeof(key));
5951
5952         path = btrfs_alloc_path();
5953         BUG_ON(!path);
5954
5955         spin_lock(&root->fs_info->block_group_cache_lock);
5956         rb_erase(&block_group->cache_node,
5957                  &root->fs_info->block_group_cache_tree);
5958         spin_unlock(&root->fs_info->block_group_cache_lock);
5959         btrfs_remove_free_space_cache(block_group);
5960         down_write(&block_group->space_info->groups_sem);
5961         list_del(&block_group->list);
5962         up_write(&block_group->space_info->groups_sem);
5963
5964         spin_lock(&block_group->space_info->lock);
5965         block_group->space_info->total_bytes -= block_group->key.offset;
5966         block_group->space_info->bytes_readonly -= block_group->key.offset;
5967         spin_unlock(&block_group->space_info->lock);
5968         block_group->space_info->full = 0;
5969
5970         put_block_group(block_group);
5971         put_block_group(block_group);
5972
5973         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5974         if (ret > 0)
5975                 ret = -EIO;
5976         if (ret < 0)
5977                 goto out;
5978
5979         ret = btrfs_del_item(trans, root, path);
5980 out:
5981         btrfs_free_path(path);
5982         return ret;
5983 }