fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include <linux/sort.h>
  23 #include "compat.h"
  24 #include "hash.h"
  25 #include "crc32c.h"
  26 #include "ctree.h"
  27 #include "disk-io.h"
  28 #include "print-tree.h"
  29 #include "transaction.h"
  30 #include "volumes.h"
  31 #include "locking.h"
  32 #include "ref-cache.h"
  33
  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 /* when a block goes through cow, we update the reference counts of
1526  * everything that block points to.  The internal pointers of the block
1527  * can be in just about any order, and it is likely to have clusters of
1528  * things that are close together and clusters of things that are not.
1529  *
1530  * To help reduce the seeks that come with updating all of these reference
1531  * counts, sort them by byte number before actual updates are done.
1532  *
1533  * struct refsort is used to match byte number to slot in the btree block.
1534  * we sort based on the byte number and then use the slot to actually
1535  * find the item.
1536  */
1537 struct refsort {
1538         u64 bytenr;
1539         u32 slot;
1540 };
1541
1542 /*
1543  * for passing into sort()
1544  */
1545 static int refsort_cmp(const void *a_void, const void *b_void)
1546 {
1547         const struct refsort *a = a_void;
1548         const struct refsort *b = b_void;
1549
1550         if (a->bytenr < b->bytenr)
1551                 return -1;
1552         if (a->bytenr > b->bytenr)
1553                 return 1;
1554         return 0;
1555 }
1556
1557
1558 noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1559                            struct btrfs_root *root,
1560                            struct extent_buffer *orig_buf,
1561                            struct extent_buffer *buf, u32 *nr_extents)
1562 {
1563         u64 bytenr;
1564         u64 ref_root;
1565         u64 orig_root;
1566         u64 ref_generation;
1567         u64 orig_generation;
1568         struct refsort *sorted;
1569         u32 nritems;
1570         u32 nr_file_extents = 0;
1571         struct btrfs_key key;
1572         struct btrfs_file_extent_item *fi;
1573         int i;
1574         int level;
1575         int ret = 0;
1576         int faili = 0;
1577         int refi = 0;
1578         int slot;
1579         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1580                             u64, u64, u64, u64, u64, u64, u64, u64);
1581
1582         ref_root = btrfs_header_owner(buf);
1583         ref_generation = btrfs_header_generation(buf);
1584         orig_root = btrfs_header_owner(orig_buf);
1585         orig_generation = btrfs_header_generation(orig_buf);
1586
1587         nritems = btrfs_header_nritems(buf);
1588         level = btrfs_header_level(buf);
1589
1590         sorted = kmalloc(sizeof(struct refsort) * nritems, GFP_NOFS);
1591         BUG_ON(!sorted);
1592
1593         if (root->ref_cows) {
1594                 process_func = __btrfs_inc_extent_ref;
1595         } else {
1596                 if (level == 0 &&
1597                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1598                         goto out;
1599                 if (level != 0 &&
1600                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1601                         goto out;
1602                 process_func = __btrfs_update_extent_ref;
1603         }
1604
1605         /*
1606          * we make two passes through the items.  In the first pass we
1607          * only record the byte number and slot.  Then we sort based on
1608          * byte number and do the actual work based on the sorted results
1609          */
1610         for (i = 0; i < nritems; i++) {
1611                 cond_resched();
1612                 if (level == 0) {
1613                         btrfs_item_key_to_cpu(buf, &key, i);
1614                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1615                                 continue;
1616                         fi = btrfs_item_ptr(buf, i,
1617                                             struct btrfs_file_extent_item);
1618                         if (btrfs_file_extent_type(buf, fi) ==
1619                             BTRFS_FILE_EXTENT_INLINE)
1620                                 continue;
1621                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1622                         if (bytenr == 0)
1623                                 continue;
1624
1625                         nr_file_extents++;
1626                         sorted[refi].bytenr = bytenr;
1627                         sorted[refi].slot = i;
1628                         refi++;
1629                 } else {
1630                         bytenr = btrfs_node_blockptr(buf, i);
1631                         sorted[refi].bytenr = bytenr;
1632                         sorted[refi].slot = i;
1633                         refi++;
1634                 }
1635         }
1636         /*
1637          * if refi == 0, we didn't actually put anything into the sorted
1638          * array and we're done
1639          */
1640         if (refi == 0)
1641                 goto out;
1642
1643         sort(sorted, refi, sizeof(struct refsort), refsort_cmp, NULL);
1644
1645         for (i = 0; i < refi; i++) {
1646                 cond_resched();
1647                 slot = sorted[i].slot;
1648                 bytenr = sorted[i].bytenr;
1649
1650                 if (level == 0) {
1651                         btrfs_item_key_to_cpu(buf, &key, slot);
1652
1653                         ret = process_func(trans, root, bytenr,
1654                                            orig_buf->start, buf->start,
1655                                            orig_root, ref_root,
1656                                            orig_generation, ref_generation,
1657                                            key.objectid);
1658
1659                         if (ret) {
1660                                 faili = slot;
1661                                 WARN_ON(1);
1662                                 goto fail;
1663                         }
1664                 } else {
1665                         ret = process_func(trans, root, bytenr,
1666                                            orig_buf->start, buf->start,
1667                                            orig_root, ref_root,
1668                                            orig_generation, ref_generation,
1669                                            level - 1);
1670                         if (ret) {
1671                                 faili = slot;
1672                                 WARN_ON(1);
1673                                 goto fail;
1674                         }
1675                 }
1676         }
1677 out:
1678         kfree(sorted);
1679         if (nr_extents) {
1680                 if (level == 0)
1681                         *nr_extents = nr_file_extents;
1682                 else
1683                         *nr_extents = nritems;
1684         }
1685         return 0;
1686 fail:
1687         kfree(sorted);
1688         WARN_ON(1);
1689         return ret;
1690 }
1691
1692 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1693                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1694                      struct extent_buffer *buf, int start_slot, int nr)
1695
1696 {
1697         u64 bytenr;
1698         u64 ref_root;
1699         u64 orig_root;
1700         u64 ref_generation;
1701         u64 orig_generation;
1702         struct btrfs_key key;
1703         struct btrfs_file_extent_item *fi;
1704         int i;
1705         int ret;
1706         int slot;
1707         int level;
1708
1709         BUG_ON(start_slot < 0);
1710         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1711
1712         ref_root = btrfs_header_owner(buf);
1713         ref_generation = btrfs_header_generation(buf);
1714         orig_root = btrfs_header_owner(orig_buf);
1715         orig_generation = btrfs_header_generation(orig_buf);
1716         level = btrfs_header_level(buf);
1717
1718         if (!root->ref_cows) {
1719                 if (level == 0 &&
1720                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1721                         return 0;
1722                 if (level != 0 &&
1723                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1724                         return 0;
1725         }
1726
1727         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1728                 cond_resched();
1729                 if (level == 0) {
1730                         btrfs_item_key_to_cpu(buf, &key, slot);
1731                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1732                                 continue;
1733                         fi = btrfs_item_ptr(buf, slot,
1734                                             struct btrfs_file_extent_item);
1735                         if (btrfs_file_extent_type(buf, fi) ==
1736                             BTRFS_FILE_EXTENT_INLINE)
1737                                 continue;
1738                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1739                         if (bytenr == 0)
1740                                 continue;
1741                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1742                                             orig_buf->start, buf->start,
1743                                             orig_root, ref_root,
1744                                             orig_generation, ref_generation,
1745                                             key.objectid);
1746                         if (ret)
1747                                 goto fail;
1748                 } else {
1749                         bytenr = btrfs_node_blockptr(buf, slot);
1750                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1751                                             orig_buf->start, buf->start,
1752                                             orig_root, ref_root,
1753                                             orig_generation, ref_generation,
1754                                             level - 1);
1755                         if (ret)
1756                                 goto fail;
1757                 }
1758         }
1759         return 0;
1760 fail:
1761         WARN_ON(1);
1762         return -1;
1763 }
1764
1765 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1766                                  struct btrfs_root *root,
1767                                  struct btrfs_path *path,
1768                                  struct btrfs_block_group_cache *cache)
1769 {
1770         int ret;
1771         int pending_ret;
1772         struct btrfs_root *extent_root = root->fs_info->extent_root;
1773         unsigned long bi;
1774         struct extent_buffer *leaf;
1775
1776         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1777         if (ret < 0)
1778                 goto fail;
1779         BUG_ON(ret);
1780
1781         leaf = path->nodes[0];
1782         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1783         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1784         btrfs_mark_buffer_dirty(leaf);
1785         btrfs_release_path(extent_root, path);
1786 fail:
1787         finish_current_insert(trans, extent_root, 0);
1788         pending_ret = del_pending_extents(trans, extent_root, 0);
1789         if (ret)
1790                 return ret;
1791         if (pending_ret)
1792                 return pending_ret;
1793         return 0;
1794
1795 }
1796
1797 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1798                                    struct btrfs_root *root)
1799 {
1800         struct btrfs_block_group_cache *cache, *entry;
1801         struct rb_node *n;
1802         int err = 0;
1803         int werr = 0;
1804         struct btrfs_path *path;
1805         u64 last = 0;
1806
1807         path = btrfs_alloc_path();
1808         if (!path)
1809                 return -ENOMEM;
1810
1811         while (1) {
1812                 cache = NULL;
1813                 spin_lock(&root->fs_info->block_group_cache_lock);
1814                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1815                      n; n = rb_next(n)) {
1816                         entry = rb_entry(n, struct btrfs_block_group_cache,
1817                                          cache_node);
1818                         if (entry->dirty) {
1819                                 cache = entry;
1820                                 break;
1821                         }
1822                 }
1823                 spin_unlock(&root->fs_info->block_group_cache_lock);
1824
1825                 if (!cache)
1826                         break;
1827
1828                 cache->dirty = 0;
1829                 last += cache->key.offset;
1830
1831                 err = write_one_cache_group(trans, root,
1832                                             path, cache);
1833                 /*
1834                  * if we fail to write the cache group, we want
1835                  * to keep it marked dirty in hopes that a later
1836                  * write will work
1837                  */
1838                 if (err) {
1839                         werr = err;
1840                         continue;
1841                 }
1842         }
1843         btrfs_free_path(path);
1844         return werr;
1845 }
1846
1847 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1848 {
1849         struct btrfs_block_group_cache *block_group;
1850         int readonly = 0;
1851
1852         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1853         if (!block_group || block_group->ro)
1854                 readonly = 1;
1855         if (block_group)
1856                 put_block_group(block_group);
1857         return readonly;
1858 }
1859
1860 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1861                              u64 total_bytes, u64 bytes_used,
1862                              struct btrfs_space_info **space_info)
1863 {
1864         struct btrfs_space_info *found;
1865
1866         found = __find_space_info(info, flags);
1867         if (found) {
1868                 spin_lock(&found->lock);
1869                 found->total_bytes += total_bytes;
1870                 found->bytes_used += bytes_used;
1871                 found->full = 0;
1872                 spin_unlock(&found->lock);
1873                 *space_info = found;
1874                 return 0;
1875         }
1876         found = kzalloc(sizeof(*found), GFP_NOFS);
1877         if (!found)
1878                 return -ENOMEM;
1879
1880         list_add(&found->list, &info->space_info);
1881         INIT_LIST_HEAD(&found->block_groups);
1882         init_rwsem(&found->groups_sem);
1883         spin_lock_init(&found->lock);
1884         found->flags = flags;
1885         found->total_bytes = total_bytes;
1886         found->bytes_used = bytes_used;
1887         found->bytes_pinned = 0;
1888         found->bytes_reserved = 0;
1889         found->bytes_readonly = 0;
1890         found->full = 0;
1891         found->force_alloc = 0;
1892         *space_info = found;
1893         return 0;
1894 }
1895
1896 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1897 {
1898         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1899                                    BTRFS_BLOCK_GROUP_RAID1 |
1900                                    BTRFS_BLOCK_GROUP_RAID10 |
1901                                    BTRFS_BLOCK_GROUP_DUP);
1902         if (extra_flags) {
1903                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1904                         fs_info->avail_data_alloc_bits |= extra_flags;
1905                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1906                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1907                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1908                         fs_info->avail_system_alloc_bits |= extra_flags;
1909         }
1910 }
1911
1912 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1913 {
1914         spin_lock(&cache->space_info->lock);
1915         spin_lock(&cache->lock);
1916         if (!cache->ro) {
1917                 cache->space_info->bytes_readonly += cache->key.offset -
1918                                         btrfs_block_group_used(&cache->item);
1919                 cache->ro = 1;
1920         }
1921         spin_unlock(&cache->lock);
1922         spin_unlock(&cache->space_info->lock);
1923 }
1924
1925 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1926 {
1927         u64 num_devices = root->fs_info->fs_devices->rw_devices;
1928
1929         if (num_devices == 1)
1930                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1931         if (num_devices < 4)
1932                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1933
1934         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1935             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1936                       BTRFS_BLOCK_GROUP_RAID10))) {
1937                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1938         }
1939
1940         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1941             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1942                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1943         }
1944
1945         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1946             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1947              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1948              (flags & BTRFS_BLOCK_GROUP_DUP)))
1949                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1950         return flags;
1951 }
1952
1953 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1954                           struct btrfs_root *extent_root, u64 alloc_bytes,
1955                           u64 flags, int force)
1956 {
1957         struct btrfs_space_info *space_info;
1958         u64 thresh;
1959         int ret = 0;
1960
1961         mutex_lock(&extent_root->fs_info->chunk_mutex);
1962
1963         flags = btrfs_reduce_alloc_profile(extent_root, flags);
1964
1965         space_info = __find_space_info(extent_root->fs_info, flags);
1966         if (!space_info) {
1967                 ret = update_space_info(extent_root->fs_info, flags,
1968                                         0, 0, &space_info);
1969                 BUG_ON(ret);
1970         }
1971         BUG_ON(!space_info);
1972
1973         spin_lock(&space_info->lock);
1974         if (space_info->force_alloc) {
1975                 force = 1;
1976                 space_info->force_alloc = 0;
1977         }
1978         if (space_info->full) {
1979                 spin_unlock(&space_info->lock);
1980                 goto out;
1981         }
1982
1983         thresh = space_info->total_bytes - space_info->bytes_readonly;
1984         thresh = div_factor(thresh, 6);
1985         if (!force &&
1986            (space_info->bytes_used + space_info->bytes_pinned +
1987             space_info->bytes_reserved + alloc_bytes) < thresh) {
1988                 spin_unlock(&space_info->lock);
1989                 goto out;
1990         }
1991         spin_unlock(&space_info->lock);
1992
1993         ret = btrfs_alloc_chunk(trans, extent_root, flags);
1994         if (ret)
1995                 space_info->full = 1;
1996 out:
1997         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1998         return ret;
1999 }
2000
2001 static int update_block_group(struct btrfs_trans_handle *trans,
2002                               struct btrfs_root *root,
2003                               u64 bytenr, u64 num_bytes, int alloc,
2004                               int mark_free)
2005 {
2006         struct btrfs_block_group_cache *cache;
2007         struct btrfs_fs_info *info = root->fs_info;
2008         u64 total = num_bytes;
2009         u64 old_val;
2010         u64 byte_in_group;
2011
2012         while (total) {
2013                 cache = btrfs_lookup_block_group(info, bytenr);
2014                 if (!cache)
2015                         return -1;
2016                 byte_in_group = bytenr - cache->key.objectid;
2017                 WARN_ON(byte_in_group > cache->key.offset);
2018
2019                 spin_lock(&cache->space_info->lock);
2020                 spin_lock(&cache->lock);
2021                 cache->dirty = 1;
2022                 old_val = btrfs_block_group_used(&cache->item);
2023                 num_bytes = min(total, cache->key.offset - byte_in_group);
2024                 if (alloc) {
2025                         old_val += num_bytes;
2026                         cache->space_info->bytes_used += num_bytes;
2027                         if (cache->ro)
2028                                 cache->space_info->bytes_readonly -= num_bytes;
2029                         btrfs_set_block_group_used(&cache->item, old_val);
2030                         spin_unlock(&cache->lock);
2031                         spin_unlock(&cache->space_info->lock);
2032                 } else {
2033                         old_val -= num_bytes;
2034                         cache->space_info->bytes_used -= num_bytes;
2035                         if (cache->ro)
2036                                 cache->space_info->bytes_readonly += num_bytes;
2037                         btrfs_set_block_group_used(&cache->item, old_val);
2038                         spin_unlock(&cache->lock);
2039                         spin_unlock(&cache->space_info->lock);
2040                         if (mark_free) {
2041                                 int ret;
2042
2043                                 ret = btrfs_discard_extent(root, bytenr,
2044                                                            num_bytes);
2045                                 WARN_ON(ret);
2046
2047                                 ret = btrfs_add_free_space(cache, bytenr,
2048                                                            num_bytes);
2049                                 WARN_ON(ret);
2050                         }
2051                 }
2052                 put_block_group(cache);
2053                 total -= num_bytes;
2054                 bytenr += num_bytes;
2055         }
2056         return 0;
2057 }
2058
2059 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
2060 {
2061         struct btrfs_block_group_cache *cache;
2062         u64 bytenr;
2063
2064         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
2065         if (!cache)
2066                 return 0;
2067
2068         bytenr = cache->key.objectid;
2069         put_block_group(cache);
2070
2071         return bytenr;
2072 }
2073
2074 int btrfs_update_pinned_extents(struct btrfs_root *root,
2075                                 u64 bytenr, u64 num, int pin)
2076 {
2077         u64 len;
2078         struct btrfs_block_group_cache *cache;
2079         struct btrfs_fs_info *fs_info = root->fs_info;
2080
2081         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2082         if (pin) {
2083                 set_extent_dirty(&fs_info->pinned_extents,
2084                                 bytenr, bytenr + num - 1, GFP_NOFS);
2085         } else {
2086                 clear_extent_dirty(&fs_info->pinned_extents,
2087                                 bytenr, bytenr + num - 1, GFP_NOFS);
2088         }
2089         while (num > 0) {
2090                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2091                 BUG_ON(!cache);
2092                 len = min(num, cache->key.offset -
2093                           (bytenr - cache->key.objectid));
2094                 if (pin) {
2095                         spin_lock(&cache->space_info->lock);
2096                         spin_lock(&cache->lock);
2097                         cache->pinned += len;
2098                         cache->space_info->bytes_pinned += len;
2099                         spin_unlock(&cache->lock);
2100                         spin_unlock(&cache->space_info->lock);
2101                         fs_info->total_pinned += len;
2102                 } else {
2103                         spin_lock(&cache->space_info->lock);
2104                         spin_lock(&cache->lock);
2105                         cache->pinned -= len;
2106                         cache->space_info->bytes_pinned -= len;
2107                         spin_unlock(&cache->lock);
2108                         spin_unlock(&cache->space_info->lock);
2109                         fs_info->total_pinned -= len;
2110                         if (cache->cached)
2111                                 btrfs_add_free_space(cache, bytenr, len);
2112                 }
2113                 put_block_group(cache);
2114                 bytenr += len;
2115                 num -= len;
2116         }
2117         return 0;
2118 }
2119
2120 static int update_reserved_extents(struct btrfs_root *root,
2121                                    u64 bytenr, u64 num, int reserve)
2122 {
2123         u64 len;
2124         struct btrfs_block_group_cache *cache;
2125         struct btrfs_fs_info *fs_info = root->fs_info;
2126
2127         while (num > 0) {
2128                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2129                 BUG_ON(!cache);
2130                 len = min(num, cache->key.offset -
2131                           (bytenr - cache->key.objectid));
2132
2133                 spin_lock(&cache->space_info->lock);
2134                 spin_lock(&cache->lock);
2135                 if (reserve) {
2136                         cache->reserved += len;
2137                         cache->space_info->bytes_reserved += len;
2138                 } else {
2139                         cache->reserved -= len;
2140                         cache->space_info->bytes_reserved -= len;
2141                 }
2142                 spin_unlock(&cache->lock);
2143                 spin_unlock(&cache->space_info->lock);
2144                 put_block_group(cache);
2145                 bytenr += len;
2146                 num -= len;
2147         }
2148         return 0;
2149 }
2150
2151 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2152 {
2153         u64 last = 0;
2154         u64 start;
2155         u64 end;
2156         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2157         int ret;
2158
2159         mutex_lock(&root->fs_info->pinned_mutex);
2160         while (1) {
2161                 ret = find_first_extent_bit(pinned_extents, last,
2162                                             &start, &end, EXTENT_DIRTY);
2163                 if (ret)
2164                         break;
2165                 set_extent_dirty(copy, start, end, GFP_NOFS);
2166                 last = end + 1;
2167         }
2168         mutex_unlock(&root->fs_info->pinned_mutex);
2169         return 0;
2170 }
2171
2172 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2173                                struct btrfs_root *root,
2174                                struct extent_io_tree *unpin)
2175 {
2176         u64 start;
2177         u64 end;
2178         int ret;
2179
2180         mutex_lock(&root->fs_info->pinned_mutex);
2181         while (1) {
2182                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2183                                             EXTENT_DIRTY);
2184                 if (ret)
2185                         break;
2186
2187                 ret = btrfs_discard_extent(root, start, end + 1 - start);
2188
2189                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2190                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2191
2192                 if (need_resched()) {
2193                         mutex_unlock(&root->fs_info->pinned_mutex);
2194                         cond_resched();
2195                         mutex_lock(&root->fs_info->pinned_mutex);
2196                 }
2197         }
2198         mutex_unlock(&root->fs_info->pinned_mutex);
2199         return ret;
2200 }
2201
2202 static int finish_current_insert(struct btrfs_trans_handle *trans,
2203                                  struct btrfs_root *extent_root, int all)
2204 {
2205         u64 start;
2206         u64 end;
2207         u64 priv;
2208         u64 search = 0;
2209         u64 skipped = 0;
2210         struct btrfs_fs_info *info = extent_root->fs_info;
2211         struct btrfs_path *path;
2212         struct pending_extent_op *extent_op, *tmp;
2213         struct list_head insert_list, update_list;
2214         int ret;
2215         int num_inserts = 0, max_inserts;
2216
2217         path = btrfs_alloc_path();
2218         INIT_LIST_HEAD(&insert_list);
2219         INIT_LIST_HEAD(&update_list);
2220
2221         max_inserts = extent_root->leafsize /
2222                 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2223                  sizeof(struct btrfs_extent_ref) +
2224                  sizeof(struct btrfs_extent_item));
2225 again:
2226         mutex_lock(&info->extent_ins_mutex);
2227         while (1) {
2228                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2229                                             &end, EXTENT_WRITEBACK);
2230                 if (ret) {
2231                         if (skipped && all && !num_inserts &&
2232                             list_empty(&update_list)) {
2233                                 skipped = 0;
2234                                 search = 0;
2235                                 continue;
2236                         }
2237                         mutex_unlock(&info->extent_ins_mutex);
2238                         break;
2239                 }
2240
2241                 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2242                 if (!ret) {
2243                         skipped = 1;
2244                         search = end + 1;
2245                         if (need_resched()) {
2246                                 mutex_unlock(&info->extent_ins_mutex);
2247                                 cond_resched();
2248                                 mutex_lock(&info->extent_ins_mutex);
2249                         }
2250                         continue;
2251                 }
2252
2253                 ret = get_state_private(&info->extent_ins, start, &priv);
2254                 BUG_ON(ret);
2255                 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2256
2257                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2258                         num_inserts++;
2259                         list_add_tail(&extent_op->list, &insert_list);
2260                         search = end + 1;
2261                         if (num_inserts == max_inserts) {
2262                                 mutex_unlock(&info->extent_ins_mutex);
2263                                 break;
2264                         }
2265                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2266                         list_add_tail(&extent_op->list, &update_list);
2267                         search = end + 1;
2268                 } else {
2269                         BUG();
2270                 }
2271         }
2272
2273         /*
2274          * process the update list, clear the writeback bit for it, and if
2275          * somebody marked this thing for deletion then just unlock it and be
2276          * done, the free_extents will handle it
2277          */
2278         mutex_lock(&info->extent_ins_mutex);
2279         list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2280                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2281                                   extent_op->bytenr + extent_op->num_bytes - 1,
2282                                   EXTENT_WRITEBACK, GFP_NOFS);
2283                 if (extent_op->del) {
2284                         list_del_init(&extent_op->list);
2285                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2286                                       extent_op->bytenr + extent_op->num_bytes
2287                                       - 1, GFP_NOFS);
2288                         kfree(extent_op);
2289                 }
2290         }
2291         mutex_unlock(&info->extent_ins_mutex);
2292
2293         /*
2294          * still have things left on the update list, go ahead an update
2295          * everything
2296          */
2297         if (!list_empty(&update_list)) {
2298                 ret = update_backrefs(trans, extent_root, path, &update_list);
2299                 BUG_ON(ret);
2300         }
2301
2302         /*
2303          * if no inserts need to be done, but we skipped some extents and we
2304          * need to make sure everything is cleaned then reset everything and
2305          * go back to the beginning
2306          */
2307         if (!num_inserts && all && skipped) {
2308                 search = 0;
2309                 skipped = 0;
2310                 INIT_LIST_HEAD(&update_list);
2311                 INIT_LIST_HEAD(&insert_list);
2312                 goto again;
2313         } else if (!num_inserts) {
2314                 goto out;
2315         }
2316
2317         /*
2318          * process the insert extents list.  Again if we are deleting this
2319          * extent, then just unlock it, pin down the bytes if need be, and be
2320          * done with it.  Saves us from having to actually insert the extent
2321          * into the tree and then subsequently come along and delete it
2322          */
2323         mutex_lock(&info->extent_ins_mutex);
2324         list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2325                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2326                                   extent_op->bytenr + extent_op->num_bytes - 1,
2327                                   EXTENT_WRITEBACK, GFP_NOFS);
2328                 if (extent_op->del) {
2329                         u64 used;
2330                         list_del_init(&extent_op->list);
2331                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2332                                       extent_op->bytenr + extent_op->num_bytes
2333                                       - 1, GFP_NOFS);
2334
2335                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2336                         ret = pin_down_bytes(trans, extent_root,
2337                                              extent_op->bytenr,
2338                                              extent_op->num_bytes, 0);
2339                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2340
2341                         spin_lock(&info->delalloc_lock);
2342                         used = btrfs_super_bytes_used(&info->super_copy);
2343                         btrfs_set_super_bytes_used(&info->super_copy,
2344                                         used - extent_op->num_bytes);
2345                         used = btrfs_root_used(&extent_root->root_item);
2346                         btrfs_set_root_used(&extent_root->root_item,
2347                                         used - extent_op->num_bytes);
2348                         spin_unlock(&info->delalloc_lock);
2349
2350                         ret = update_block_group(trans, extent_root,
2351                                                  extent_op->bytenr,
2352                                                  extent_op->num_bytes,
2353                                                  0, ret > 0);
2354                         BUG_ON(ret);
2355                         kfree(extent_op);
2356                         num_inserts--;
2357                 }
2358         }
2359         mutex_unlock(&info->extent_ins_mutex);
2360
2361         ret = insert_extents(trans, extent_root, path, &insert_list,
2362                              num_inserts);
2363         BUG_ON(ret);
2364
2365         /*
2366          * if we broke out of the loop in order to insert stuff because we hit
2367          * the maximum number of inserts at a time we can handle, then loop
2368          * back and pick up where we left off
2369          */
2370         if (num_inserts == max_inserts) {
2371                 INIT_LIST_HEAD(&insert_list);
2372                 INIT_LIST_HEAD(&update_list);
2373                 num_inserts = 0;
2374                 goto again;
2375         }
2376
2377         /*
2378          * again, if we need to make absolutely sure there are no more pending
2379          * extent operations left and we know that we skipped some, go back to
2380          * the beginning and do it all again
2381          */
2382         if (all && skipped) {
2383                 INIT_LIST_HEAD(&insert_list);
2384                 INIT_LIST_HEAD(&update_list);
2385                 search = 0;
2386                 skipped = 0;
2387                 num_inserts = 0;
2388                 goto again;
2389         }
2390 out:
2391         btrfs_free_path(path);
2392         return 0;
2393 }
2394
2395 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2396                           struct btrfs_root *root,
2397                           u64 bytenr, u64 num_bytes, int is_data)
2398 {
2399         int err = 0;
2400         struct extent_buffer *buf;
2401
2402         if (is_data)
2403                 goto pinit;
2404
2405         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2406         if (!buf)
2407                 goto pinit;
2408
2409         /* we can reuse a block if it hasn't been written
2410          * and it is from this transaction.  We can't
2411          * reuse anything from the tree log root because
2412          * it has tiny sub-transactions.
2413          */
2414         if (btrfs_buffer_uptodate(buf, 0) &&
2415             btrfs_try_tree_lock(buf)) {
2416                 u64 header_owner = btrfs_header_owner(buf);
2417                 u64 header_transid = btrfs_header_generation(buf);
2418                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2419                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2420                     header_transid == trans->transid &&
2421                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2422                         clean_tree_block(NULL, root, buf);
2423                         btrfs_tree_unlock(buf);
2424                         free_extent_buffer(buf);
2425                         return 1;
2426                 }
2427                 btrfs_tree_unlock(buf);
2428         }
2429         free_extent_buffer(buf);
2430 pinit:
2431         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2432
2433         BUG_ON(err < 0);
2434         return 0;
2435 }
2436
2437 /*
2438  * remove an extent from the root, returns 0 on success
2439  */
2440 static int __free_extent(struct btrfs_trans_handle *trans,
2441                          struct btrfs_root *root,
2442                          u64 bytenr, u64 num_bytes, u64 parent,
2443                          u64 root_objectid, u64 ref_generation,
2444                          u64 owner_objectid, int pin, int mark_free)
2445 {
2446         struct btrfs_path *path;
2447         struct btrfs_key key;
2448         struct btrfs_fs_info *info = root->fs_info;
2449         struct btrfs_root *extent_root = info->extent_root;
2450         struct extent_buffer *leaf;
2451         int ret;
2452         int extent_slot = 0;
2453         int found_extent = 0;
2454         int num_to_del = 1;
2455         struct btrfs_extent_item *ei;
2456         u32 refs;
2457
2458         key.objectid = bytenr;
2459         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2460         key.offset = num_bytes;
2461         path = btrfs_alloc_path();
2462         if (!path)
2463                 return -ENOMEM;
2464
2465         path->reada = 1;
2466         ret = lookup_extent_backref(trans, extent_root, path,
2467                                     bytenr, parent, root_objectid,
2468                                     ref_generation, owner_objectid, 1);
2469         if (ret == 0) {
2470                 struct btrfs_key found_key;
2471                 extent_slot = path->slots[0];
2472                 while (extent_slot > 0) {
2473                         extent_slot--;
2474                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2475                                               extent_slot);
2476                         if (found_key.objectid != bytenr)
2477                                 break;
2478                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2479                             found_key.offset == num_bytes) {
2480                                 found_extent = 1;
2481                                 break;
2482                         }
2483                         if (path->slots[0] - extent_slot > 5)
2484                                 break;
2485                 }
2486                 if (!found_extent) {
2487                         ret = remove_extent_backref(trans, extent_root, path);
2488                         BUG_ON(ret);
2489                         btrfs_release_path(extent_root, path);
2490                         ret = btrfs_search_slot(trans, extent_root,
2491                                                 &key, path, -1, 1);
2492                         if (ret) {
2493                                 printk(KERN_ERR "umm, got %d back from search"
2494                                        ", was looking for %llu\n", ret,
2495                                        (unsigned long long)bytenr);
2496                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2497                         }
2498                         BUG_ON(ret);
2499                         extent_slot = path->slots[0];
2500                 }
2501         } else {
2502                 btrfs_print_leaf(extent_root, path->nodes[0]);
2503                 WARN_ON(1);
2504                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2505                        "root %llu gen %llu owner %llu\n",
2506                        (unsigned long long)bytenr,
2507                        (unsigned long long)root_objectid,
2508                        (unsigned long long)ref_generation,
2509                        (unsigned long long)owner_objectid);
2510         }
2511
2512         leaf = path->nodes[0];
2513         ei = btrfs_item_ptr(leaf, extent_slot,
2514                             struct btrfs_extent_item);
2515         refs = btrfs_extent_refs(leaf, ei);
2516         BUG_ON(refs == 0);
2517         refs -= 1;
2518         btrfs_set_extent_refs(leaf, ei, refs);
2519
2520         btrfs_mark_buffer_dirty(leaf);
2521
2522         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2523                 struct btrfs_extent_ref *ref;
2524                 ref = btrfs_item_ptr(leaf, path->slots[0],
2525                                      struct btrfs_extent_ref);
2526                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2527                 /* if the back ref and the extent are next to each other
2528                  * they get deleted below in one shot
2529                  */
2530                 path->slots[0] = extent_slot;
2531                 num_to_del = 2;
2532         } else if (found_extent) {
2533                 /* otherwise delete the extent back ref */
2534                 ret = remove_extent_backref(trans, extent_root, path);
2535                 BUG_ON(ret);
2536                 /* if refs are 0, we need to setup the path for deletion */
2537                 if (refs == 0) {
2538                         btrfs_release_path(extent_root, path);
2539                         ret = btrfs_search_slot(trans, extent_root, &key, path,
2540                                                 -1, 1);
2541                         BUG_ON(ret);
2542                 }
2543         }
2544
2545         if (refs == 0) {
2546                 u64 super_used;
2547                 u64 root_used;
2548
2549                 if (pin) {
2550                         mutex_lock(&root->fs_info->pinned_mutex);
2551                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2552                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2553                         mutex_unlock(&root->fs_info->pinned_mutex);
2554                         if (ret > 0)
2555                                 mark_free = 1;
2556                         BUG_ON(ret < 0);
2557                 }
2558                 /* block accounting for super block */
2559                 spin_lock(&info->delalloc_lock);
2560                 super_used = btrfs_super_bytes_used(&info->super_copy);
2561                 btrfs_set_super_bytes_used(&info->super_copy,
2562                                            super_used - num_bytes);
2563
2564                 /* block accounting for root item */
2565                 root_used = btrfs_root_used(&root->root_item);
2566                 btrfs_set_root_used(&root->root_item,
2567                                            root_used - num_bytes);
2568                 spin_unlock(&info->delalloc_lock);
2569                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2570                                       num_to_del);
2571                 BUG_ON(ret);
2572                 btrfs_release_path(extent_root, path);
2573
2574                 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2575                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2576                         BUG_ON(ret);
2577                 }
2578
2579                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2580                                          mark_free);
2581                 BUG_ON(ret);
2582         }
2583         btrfs_free_path(path);
2584         finish_current_insert(trans, extent_root, 0);
2585         return ret;
2586 }
2587
2588 /*
2589  * find all the blocks marked as pending in the radix tree and remove
2590  * them from the extent map
2591  */
2592 static int del_pending_extents(struct btrfs_trans_handle *trans,
2593                                struct btrfs_root *extent_root, int all)
2594 {
2595         int ret;
2596         int err = 0;
2597         u64 start;
2598         u64 end;
2599         u64 priv;
2600         u64 search = 0;
2601         int nr = 0, skipped = 0;
2602         struct extent_io_tree *pending_del;
2603         struct extent_io_tree *extent_ins;
2604         struct pending_extent_op *extent_op;
2605         struct btrfs_fs_info *info = extent_root->fs_info;
2606         struct list_head delete_list;
2607
2608         INIT_LIST_HEAD(&delete_list);
2609         extent_ins = &extent_root->fs_info->extent_ins;
2610         pending_del = &extent_root->fs_info->pending_del;
2611
2612 again:
2613         mutex_lock(&info->extent_ins_mutex);
2614         while (1) {
2615                 ret = find_first_extent_bit(pending_del, search, &start, &end,
2616                                             EXTENT_WRITEBACK);
2617                 if (ret) {
2618                         if (all && skipped && !nr) {
2619                                 search = 0;
2620                                 skipped = 0;
2621                                 continue;
2622                         }
2623                         mutex_unlock(&info->extent_ins_mutex);
2624                         break;
2625                 }
2626
2627                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2628                 if (!ret) {
2629                         search = end+1;
2630                         skipped = 1;
2631
2632                         if (need_resched()) {
2633                                 mutex_unlock(&info->extent_ins_mutex);
2634                                 cond_resched();
2635                                 mutex_lock(&info->extent_ins_mutex);
2636                         }
2637
2638                         continue;
2639                 }
2640                 BUG_ON(ret < 0);
2641
2642                 ret = get_state_private(pending_del, start, &priv);
2643                 BUG_ON(ret);
2644                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2645
2646                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2647                                   GFP_NOFS);
2648                 if (!test_range_bit(extent_ins, start, end,
2649                                     EXTENT_WRITEBACK, 0)) {
2650                         list_add_tail(&extent_op->list, &delete_list);
2651                         nr++;
2652                 } else {
2653                         kfree(extent_op);
2654
2655                         ret = get_state_private(&info->extent_ins, start,
2656                                                 &priv);
2657                         BUG_ON(ret);
2658                         extent_op = (struct pending_extent_op *)
2659                                                 (unsigned long)priv;
2660
2661                         clear_extent_bits(&info->extent_ins, start, end,
2662                                           EXTENT_WRITEBACK, GFP_NOFS);
2663
2664                         if (extent_op->type == PENDING_BACKREF_UPDATE) {
2665                                 list_add_tail(&extent_op->list, &delete_list);
2666                                 search = end + 1;
2667                                 nr++;
2668                                 continue;
2669                         }
2670
2671                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2672                         ret = pin_down_bytes(trans, extent_root, start,
2673                                              end + 1 - start, 0);
2674                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2675
2676                         ret = update_block_group(trans, extent_root, start,
2677                                                 end + 1 - start, 0, ret > 0);
2678
2679                         unlock_extent(extent_ins, start, end, GFP_NOFS);
2680                         BUG_ON(ret);
2681                         kfree(extent_op);
2682                 }
2683                 if (ret)
2684                         err = ret;
2685
2686                 search = end + 1;
2687
2688                 if (need_resched()) {
2689                         mutex_unlock(&info->extent_ins_mutex);
2690                         cond_resched();
2691                         mutex_lock(&info->extent_ins_mutex);
2692                 }
2693         }
2694
2695         if (nr) {
2696                 ret = free_extents(trans, extent_root, &delete_list);
2697                 BUG_ON(ret);
2698         }
2699
2700         if (all && skipped) {
2701                 INIT_LIST_HEAD(&delete_list);
2702                 search = 0;
2703                 nr = 0;
2704                 goto again;
2705         }
2706
2707         return err;
2708 }
2709
2710 /*
2711  * remove an extent from the root, returns 0 on success
2712  */
2713 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2714                                struct btrfs_root *root,
2715                                u64 bytenr, u64 num_bytes, u64 parent,
2716                                u64 root_objectid, u64 ref_generation,
2717                                u64 owner_objectid, int pin)
2718 {
2719         struct btrfs_root *extent_root = root->fs_info->extent_root;
2720         int pending_ret;
2721         int ret;
2722
2723         WARN_ON(num_bytes < root->sectorsize);
2724         if (root == extent_root) {
2725                 struct pending_extent_op *extent_op = NULL;
2726
2727                 mutex_lock(&root->fs_info->extent_ins_mutex);
2728                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2729                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2730                         u64 priv;
2731                         ret = get_state_private(&root->fs_info->extent_ins,
2732                                                 bytenr, &priv);
2733                         BUG_ON(ret);
2734                         extent_op = (struct pending_extent_op *)
2735                                                 (unsigned long)priv;
2736
2737                         extent_op->del = 1;
2738                         if (extent_op->type == PENDING_EXTENT_INSERT) {
2739                                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2740                                 return 0;
2741                         }
2742                 }
2743
2744                 if (extent_op) {
2745                         ref_generation = extent_op->orig_generation;
2746                         parent = extent_op->orig_parent;
2747                 }
2748
2749                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2750                 BUG_ON(!extent_op);
2751
2752                 extent_op->type = PENDING_EXTENT_DELETE;
2753                 extent_op->bytenr = bytenr;
2754                 extent_op->num_bytes = num_bytes;
2755                 extent_op->parent = parent;
2756                 extent_op->orig_parent = parent;
2757                 extent_op->generation = ref_generation;
2758                 extent_op->orig_generation = ref_generation;
2759                 extent_op->level = (int)owner_objectid;
2760                 INIT_LIST_HEAD(&extent_op->list);
2761                 extent_op->del = 0;
2762
2763                 set_extent_bits(&root->fs_info->pending_del,
2764                                 bytenr, bytenr + num_bytes - 1,
2765                                 EXTENT_WRITEBACK, GFP_NOFS);
2766                 set_state_private(&root->fs_info->pending_del,
2767                                   bytenr, (unsigned long)extent_op);
2768                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2769                 return 0;
2770         }
2771         /* if metadata always pin */
2772         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2773                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2774                         mutex_lock(&root->fs_info->pinned_mutex);
2775                         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2776                         mutex_unlock(&root->fs_info->pinned_mutex);
2777                         update_reserved_extents(root, bytenr, num_bytes, 0);
2778                         return 0;
2779                 }
2780                 pin = 1;
2781         }
2782
2783         /* if data pin when any transaction has committed this */
2784         if (ref_generation != trans->transid)
2785                 pin = 1;
2786
2787         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2788                             root_objectid, ref_generation,
2789                             owner_objectid, pin, pin == 0);
2790
2791         finish_current_insert(trans, root->fs_info->extent_root, 0);
2792         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2793         return ret ? ret : pending_ret;
2794 }
2795
2796 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2797                       struct btrfs_root *root,
2798                       u64 bytenr, u64 num_bytes, u64 parent,
2799                       u64 root_objectid, u64 ref_generation,
2800                       u64 owner_objectid, int pin)
2801 {
2802         int ret;
2803
2804         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2805                                   root_objectid, ref_generation,
2806                                   owner_objectid, pin);
2807         return ret;
2808 }
2809
2810 static u64 stripe_align(struct btrfs_root *root, u64 val)
2811 {
2812         u64 mask = ((u64)root->stripesize - 1);
2813         u64 ret = (val + mask) & ~mask;
2814         return ret;
2815 }
2816
2817 /*
2818  * walks the btree of allocated extents and find a hole of a given size.
2819  * The key ins is changed to record the hole:
2820  * ins->objectid == block start
2821  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2822  * ins->offset == number of blocks
2823  * Any available blocks before search_start are skipped.
2824  */
2825 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2826                                      struct btrfs_root *orig_root,
2827                                      u64 num_bytes, u64 empty_size,
2828                                      u64 search_start, u64 search_end,
2829                                      u64 hint_byte, struct btrfs_key *ins,
2830                                      u64 exclude_start, u64 exclude_nr,
2831                                      int data)
2832 {
2833         int ret = 0;
2834         struct btrfs_root *root = orig_root->fs_info->extent_root;
2835         u64 total_needed = num_bytes;
2836         u64 *last_ptr = NULL;
2837         u64 last_wanted = 0;
2838         struct btrfs_block_group_cache *block_group = NULL;
2839         int chunk_alloc_done = 0;
2840         int empty_cluster = 2 * 1024 * 1024;
2841         int allowed_chunk_alloc = 0;
2842         struct list_head *head = NULL, *cur = NULL;
2843         int loop = 0;
2844         int extra_loop = 0;
2845         struct btrfs_space_info *space_info;
2846
2847         WARN_ON(num_bytes < root->sectorsize);
2848         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2849         ins->objectid = 0;
2850         ins->offset = 0;
2851
2852         if (orig_root->ref_cows || empty_size)
2853                 allowed_chunk_alloc = 1;
2854
2855         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2856                 last_ptr = &root->fs_info->last_alloc;
2857                 empty_cluster = 64 * 1024;
2858         }
2859
2860         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2861                 last_ptr = &root->fs_info->last_data_alloc;
2862
2863         if (last_ptr) {
2864                 if (*last_ptr) {
2865                         hint_byte = *last_ptr;
2866                         last_wanted = *last_ptr;
2867                 } else
2868                         empty_size += empty_cluster;
2869         } else {
2870                 empty_cluster = 0;
2871         }
2872         search_start = max(search_start, first_logical_byte(root, 0));
2873         search_start = max(search_start, hint_byte);
2874
2875         if (last_wanted && search_start != last_wanted) {
2876                 last_wanted = 0;
2877                 empty_size += empty_cluster;
2878         }
2879
2880         total_needed += empty_size;
2881         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2882         if (!block_group)
2883                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2884                                                              search_start);
2885         space_info = __find_space_info(root->fs_info, data);
2886
2887         down_read(&space_info->groups_sem);
2888         while (1) {
2889                 struct btrfs_free_space *free_space;
2890                 /*
2891                  * the only way this happens if our hint points to a block
2892                  * group thats not of the proper type, while looping this
2893                  * should never happen
2894                  */
2895                 if (empty_size)
2896                         extra_loop = 1;
2897
2898                 if (!block_group)
2899                         goto new_group_no_lock;
2900
2901                 if (unlikely(!block_group->cached)) {
2902                         mutex_lock(&block_group->cache_mutex);
2903                         ret = cache_block_group(root, block_group);
2904                         mutex_unlock(&block_group->cache_mutex);
2905                         if (ret)
2906                                 break;
2907                 }
2908
2909                 mutex_lock(&block_group->alloc_mutex);
2910                 if (unlikely(!block_group_bits(block_group, data)))
2911                         goto new_group;
2912
2913                 if (unlikely(block_group->ro))
2914                         goto new_group;
2915
2916                 free_space = btrfs_find_free_space(block_group, search_start,
2917                                                    total_needed);
2918                 if (free_space) {
2919                         u64 start = block_group->key.objectid;
2920                         u64 end = block_group->key.objectid +
2921                                 block_group->key.offset;
2922
2923                         search_start = stripe_align(root, free_space->offset);
2924
2925                         /* move on to the next group */
2926                         if (search_start + num_bytes >= search_end)
2927                                 goto new_group;
2928
2929                         /* move on to the next group */
2930                         if (search_start + num_bytes > end)
2931                                 goto new_group;
2932
2933                         if (last_wanted && search_start != last_wanted) {
2934                                 total_needed += empty_cluster;
2935                                 empty_size += empty_cluster;
2936                                 last_wanted = 0;
2937                                 /*
2938                                  * if search_start is still in this block group
2939                                  * then we just re-search this block group
2940                                  */
2941                                 if (search_start >= start &&
2942                                     search_start < end) {
2943                                         mutex_unlock(&block_group->alloc_mutex);
2944                                         continue;
2945                                 }
2946
2947                                 /* else we go to the next block group */
2948                                 goto new_group;
2949                         }
2950
2951                         if (exclude_nr > 0 &&
2952                             (search_start + num_bytes > exclude_start &&
2953                              search_start < exclude_start + exclude_nr)) {
2954                                 search_start = exclude_start + exclude_nr;
2955                                 /*
2956                                  * if search_start is still in this block group
2957                                  * then we just re-search this block group
2958                                  */
2959                                 if (search_start >= start &&
2960                                     search_start < end) {
2961                                         mutex_unlock(&block_group->alloc_mutex);
2962                                         last_wanted = 0;
2963                                         continue;
2964                                 }
2965
2966                                 /* else we go to the next block group */
2967                                 goto new_group;
2968                         }
2969
2970                         ins->objectid = search_start;
2971                         ins->offset = num_bytes;
2972
2973                         btrfs_remove_free_space_lock(block_group, search_start,
2974                                                      num_bytes);
2975                         /* we are all good, lets return */
2976                         mutex_unlock(&block_group->alloc_mutex);
2977                         break;
2978                 }
2979 new_group:
2980                 mutex_unlock(&block_group->alloc_mutex);
2981                 put_block_group(block_group);
2982                 block_group = NULL;
2983 new_group_no_lock:
2984                 /* don't try to compare new allocations against the
2985                  * last allocation any more
2986                  */
2987                 last_wanted = 0;
2988
2989                 /*
2990                  * Here's how this works.
2991                  * loop == 0: we were searching a block group via a hint
2992                  *              and didn't find anything, so we start at
2993                  *              the head of the block groups and keep searching
2994                  * loop == 1: we're searching through all of the block groups
2995                  *              if we hit the head again we have searched
2996                  *              all of the block groups for this space and we
2997                  *              need to try and allocate, if we cant error out.
2998                  * loop == 2: we allocated more space and are looping through
2999                  *              all of the block groups again.
3000                  */
3001                 if (loop == 0) {
3002                         head = &space_info->block_groups;
3003                         cur = head->next;
3004                         loop++;
3005                 } else if (loop == 1 && cur == head) {
3006                         int keep_going;
3007
3008                         /* at this point we give up on the empty_size
3009                          * allocations and just try to allocate the min
3010                          * space.
3011                          *
3012                          * The extra_loop field was set if an empty_size
3013                          * allocation was attempted above, and if this
3014                          * is try we need to try the loop again without
3015                          * the additional empty_size.
3016                          */
3017                         total_needed -= empty_size;
3018                         empty_size = 0;
3019                         keep_going = extra_loop;
3020                         loop++;
3021
3022                         if (allowed_chunk_alloc && !chunk_alloc_done) {
3023                                 up_read(&space_info->groups_sem);
3024                                 ret = do_chunk_alloc(trans, root, num_bytes +
3025                                                      2 * 1024 * 1024, data, 1);
3026                                 down_read(&space_info->groups_sem);
3027                                 if (ret < 0)
3028                                         goto loop_check;
3029                                 head = &space_info->block_groups;
3030                                 /*
3031                                  * we've allocated a new chunk, keep
3032                                  * trying
3033                                  */
3034                                 keep_going = 1;
3035                                 chunk_alloc_done = 1;
3036                         } else if (!allowed_chunk_alloc) {
3037                                 space_info->force_alloc = 1;
3038                         }
3039 loop_check:
3040                         if (keep_going) {
3041                                 cur = head->next;
3042                                 extra_loop = 0;
3043                         } else {
3044                                 break;
3045                         }
3046                 } else if (cur == head) {
3047                         break;
3048                 }
3049
3050                 block_group = list_entry(cur, struct btrfs_block_group_cache,
3051                                          list);
3052                 atomic_inc(&block_group->count);
3053
3054                 search_start = block_group->key.objectid;
3055                 cur = cur->next;
3056         }
3057
3058         /* we found what we needed */
3059         if (ins->objectid) {
3060                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
3061                         trans->block_group = block_group->key.objectid;
3062
3063                 if (last_ptr)
3064                         *last_ptr = ins->objectid + ins->offset;
3065                 ret = 0;
3066         } else if (!ret) {
3067                 printk(KERN_ERR "btrfs searching for %llu bytes, "
3068                        "num_bytes %llu, loop %d, allowed_alloc %d\n",
3069                        (unsigned long long)total_needed,
3070                        (unsigned long long)num_bytes,
3071                        loop, allowed_chunk_alloc);
3072                 ret = -ENOSPC;
3073         }
3074         if (block_group)
3075                 put_block_group(block_group);
3076
3077         up_read(&space_info->groups_sem);
3078         return ret;
3079 }
3080
3081 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3082 {
3083         struct btrfs_block_group_cache *cache;
3084
3085         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3086                (unsigned long long)(info->total_bytes - info->bytes_used -
3087                                     info->bytes_pinned - info->bytes_reserved),
3088                (info->full) ? "" : "not ");
3089
3090         down_read(&info->groups_sem);
3091         list_for_each_entry(cache, &info->block_groups, list) {
3092                 spin_lock(&cache->lock);
3093                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3094                        "%llu pinned %llu reserved\n",
3095                        (unsigned long long)cache->key.objectid,
3096                        (unsigned long long)cache->key.offset,
3097                        (unsigned long long)btrfs_block_group_used(&cache->item),
3098                        (unsigned long long)cache->pinned,
3099                        (unsigned long long)cache->reserved);
3100                 btrfs_dump_free_space(cache, bytes);
3101                 spin_unlock(&cache->lock);
3102         }
3103         up_read(&info->groups_sem);
3104 }
3105
3106 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3107                                   struct btrfs_root *root,
3108                                   u64 num_bytes, u64 min_alloc_size,
3109                                   u64 empty_size, u64 hint_byte,
3110                                   u64 search_end, struct btrfs_key *ins,
3111                                   u64 data)
3112 {
3113         int ret;
3114         u64 search_start = 0;
3115         u64 alloc_profile;
3116         struct btrfs_fs_info *info = root->fs_info;
3117
3118         if (data) {
3119                 alloc_profile = info->avail_data_alloc_bits &
3120                         info->data_alloc_profile;
3121                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3122         } else if (root == root->fs_info->chunk_root) {
3123                 alloc_profile = info->avail_system_alloc_bits &
3124                         info->system_alloc_profile;
3125                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3126         } else {
3127                 alloc_profile = info->avail_metadata_alloc_bits &
3128                         info->metadata_alloc_profile;
3129                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3130         }
3131 again:
3132         data = btrfs_reduce_alloc_profile(root, data);
3133         /*
3134          * the only place that sets empty_size is btrfs_realloc_node, which
3135          * is not called recursively on allocations
3136          */
3137         if (empty_size || root->ref_cows) {
3138                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3139                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3140                                      2 * 1024 * 1024,
3141                                      BTRFS_BLOCK_GROUP_METADATA |
3142                                      (info->metadata_alloc_profile &
3143                                       info->avail_metadata_alloc_bits), 0);
3144                 }
3145                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3146                                      num_bytes + 2 * 1024 * 1024, data, 0);
3147         }
3148
3149         WARN_ON(num_bytes < root->sectorsize);
3150         ret = find_free_extent(trans, root, num_bytes, empty_size,
3151                                search_start, search_end, hint_byte, ins,
3152                                trans->alloc_exclude_start,
3153                                trans->alloc_exclude_nr, data);
3154
3155         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3156                 num_bytes = num_bytes >> 1;
3157                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3158                 num_bytes = max(num_bytes, min_alloc_size);
3159                 do_chunk_alloc(trans, root->fs_info->extent_root,
3160                                num_bytes, data, 1);
3161                 goto again;
3162         }
3163         if (ret) {
3164                 struct btrfs_space_info *sinfo;
3165
3166                 sinfo = __find_space_info(root->fs_info, data);
3167                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3168                        "wanted %llu\n", (unsigned long long)data,
3169                        (unsigned long long)num_bytes);
3170                 dump_space_info(sinfo, num_bytes);
3171                 BUG();
3172         }
3173
3174         return ret;
3175 }
3176
3177 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3178 {
3179         struct btrfs_block_group_cache *cache;
3180         int ret = 0;
3181
3182         cache = btrfs_lookup_block_group(root->fs_info, start);
3183         if (!cache) {
3184                 printk(KERN_ERR "Unable to find block group for %llu\n",
3185                        (unsigned long long)start);
3186                 return -ENOSPC;
3187         }
3188
3189         ret = btrfs_discard_extent(root, start, len);
3190
3191         btrfs_add_free_space(cache, start, len);
3192         put_block_group(cache);
3193         update_reserved_extents(root, start, len, 0);
3194
3195         return ret;
3196 }
3197
3198 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3199                                   struct btrfs_root *root,
3200                                   u64 num_bytes, u64 min_alloc_size,
3201                                   u64 empty_size, u64 hint_byte,
3202                                   u64 search_end, struct btrfs_key *ins,
3203                                   u64 data)
3204 {
3205         int ret;
3206         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3207                                      empty_size, hint_byte, search_end, ins,
3208                                      data);
3209         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3210         return ret;
3211 }
3212
3213 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3214                                          struct btrfs_root *root, u64 parent,
3215                                          u64 root_objectid, u64 ref_generation,
3216                                          u64 owner, struct btrfs_key *ins)
3217 {
3218         int ret;
3219         int pending_ret;
3220         u64 super_used;
3221         u64 root_used;
3222         u64 num_bytes = ins->offset;
3223         u32 sizes[2];
3224         struct btrfs_fs_info *info = root->fs_info;
3225         struct btrfs_root *extent_root = info->extent_root;
3226         struct btrfs_extent_item *extent_item;
3227         struct btrfs_extent_ref *ref;
3228         struct btrfs_path *path;
3229         struct btrfs_key keys[2];
3230
3231         if (parent == 0)
3232                 parent = ins->objectid;
3233
3234         /* block accounting for super block */
3235         spin_lock(&info->delalloc_lock);
3236         super_used = btrfs_super_bytes_used(&info->super_copy);
3237         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3238
3239         /* block accounting for root item */
3240         root_used = btrfs_root_used(&root->root_item);
3241         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3242         spin_unlock(&info->delalloc_lock);
3243
3244         if (root == extent_root) {
3245                 struct pending_extent_op *extent_op;
3246
3247                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3248                 BUG_ON(!extent_op);
3249
3250                 extent_op->type = PENDING_EXTENT_INSERT;
3251                 extent_op->bytenr = ins->objectid;
3252                 extent_op->num_bytes = ins->offset;
3253                 extent_op->parent = parent;
3254                 extent_op->orig_parent = 0;
3255                 extent_op->generation = ref_generation;
3256                 extent_op->orig_generation = 0;
3257                 extent_op->level = (int)owner;
3258                 INIT_LIST_HEAD(&extent_op->list);
3259                 extent_op->del = 0;
3260
3261                 mutex_lock(&root->fs_info->extent_ins_mutex);
3262                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3263                                 ins->objectid + ins->offset - 1,
3264                                 EXTENT_WRITEBACK, GFP_NOFS);
3265                 set_state_private(&root->fs_info->extent_ins,
3266                                   ins->objectid, (unsigned long)extent_op);
3267                 mutex_unlock(&root->fs_info->extent_ins_mutex);
3268                 goto update_block;
3269         }
3270
3271         memcpy(&keys[0], ins, sizeof(*ins));
3272         keys[1].objectid = ins->objectid;
3273         keys[1].type = BTRFS_EXTENT_REF_KEY;
3274         keys[1].offset = parent;
3275         sizes[0] = sizeof(*extent_item);
3276         sizes[1] = sizeof(*ref);
3277
3278         path = btrfs_alloc_path();
3279         BUG_ON(!path);
3280
3281         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3282                                        sizes, 2);
3283         BUG_ON(ret);
3284
3285         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3286                                      struct btrfs_extent_item);
3287         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3288         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3289                              struct btrfs_extent_ref);
3290
3291         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3292         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3293         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3294         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3295
3296         btrfs_mark_buffer_dirty(path->nodes[0]);
3297
3298         trans->alloc_exclude_start = 0;
3299         trans->alloc_exclude_nr = 0;
3300         btrfs_free_path(path);
3301         finish_current_insert(trans, extent_root, 0);
3302         pending_ret = del_pending_extents(trans, extent_root, 0);
3303
3304         if (ret)
3305                 goto out;
3306         if (pending_ret) {
3307                 ret = pending_ret;
3308                 goto out;
3309         }
3310
3311 update_block:
3312         ret = update_block_group(trans, root, ins->objectid,
3313                                  ins->offset, 1, 0);
3314         if (ret) {
3315                 printk(KERN_ERR "btrfs update block group failed for %llu "
3316                        "%llu\n", (unsigned long long)ins->objectid,
3317                        (unsigned long long)ins->offset);
3318                 BUG();
3319         }
3320 out:
3321         return ret;
3322 }
3323
3324 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3325                                 struct btrfs_root *root, u64 parent,
3326                                 u64 root_objectid, u64 ref_generation,
3327                                 u64 owner, struct btrfs_key *ins)
3328 {
3329         int ret;
3330
3331         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3332                 return 0;
3333         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3334                                             ref_generation, owner, ins);
3335         update_reserved_extents(root, ins->objectid, ins->offset, 0);
3336         return ret;
3337 }
3338
3339 /*
3340  * this is used by the tree logging recovery code.  It records that
3341  * an extent has been allocated and makes sure to clear the free
3342  * space cache bits as well
3343  */
3344 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3345                                 struct btrfs_root *root, u64 parent,
3346                                 u64 root_objectid, u64 ref_generation,
3347                                 u64 owner, struct btrfs_key *ins)
3348 {
3349         int ret;
3350         struct btrfs_block_group_cache *block_group;
3351
3352         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3353         mutex_lock(&block_group->cache_mutex);
3354         cache_block_group(root, block_group);
3355         mutex_unlock(&block_group->cache_mutex);
3356
3357         ret = btrfs_remove_free_space(block_group, ins->objectid,
3358                                       ins->offset);
3359         BUG_ON(ret);
3360         put_block_group(block_group);
3361         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3362                                             ref_generation, owner, ins);
3363         return ret;
3364 }
3365
3366 /*
3367  * finds a free extent and does all the dirty work required for allocation
3368  * returns the key for the extent through ins, and a tree buffer for
3369  * the first block of the extent through buf.
3370  *
3371  * returns 0 if everything worked, non-zero otherwise.
3372  */
3373 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3374                        struct btrfs_root *root,
3375                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3376                        u64 root_objectid, u64 ref_generation,
3377                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3378                        u64 search_end, struct btrfs_key *ins, u64 data)
3379 {
3380         int ret;
3381
3382         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3383                                      min_alloc_size, empty_size, hint_byte,
3384                                      search_end, ins, data);
3385         BUG_ON(ret);
3386         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3387                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3388                                         root_objectid, ref_generation,
3389                                         owner_objectid, ins);
3390                 BUG_ON(ret);
3391
3392         } else {
3393                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3394         }
3395         return ret;
3396 }
3397
3398 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3399                                             struct btrfs_root *root,
3400                                             u64 bytenr, u32 blocksize)
3401 {
3402         struct extent_buffer *buf;
3403
3404         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3405         if (!buf)
3406                 return ERR_PTR(-ENOMEM);
3407         btrfs_set_header_generation(buf, trans->transid);
3408         btrfs_tree_lock(buf);
3409         clean_tree_block(trans, root, buf);
3410         btrfs_set_buffer_uptodate(buf);
3411         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3412                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3413                          buf->start + buf->len - 1, GFP_NOFS);
3414         } else {
3415                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3416                          buf->start + buf->len - 1, GFP_NOFS);
3417         }
3418         trans->blocks_used++;
3419         return buf;
3420 }
3421
3422 /*
3423  * helper function to allocate a block for a given tree
3424  * returns the tree buffer or NULL.
3425  */
3426 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3427                                              struct btrfs_root *root,
3428                                              u32 blocksize, u64 parent,
3429                                              u64 root_objectid,
3430                                              u64 ref_generation,
3431                                              int level,
3432                                              u64 hint,
3433                                              u64 empty_size)
3434 {
3435         struct btrfs_key ins;
3436         int ret;
3437         struct extent_buffer *buf;
3438
3439         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3440                                  root_objectid, ref_generation, level,
3441                                  empty_size, hint, (u64)-1, &ins, 0);
3442         if (ret) {
3443                 BUG_ON(ret > 0);
3444                 return ERR_PTR(ret);
3445         }
3446
3447         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3448         return buf;
3449 }
3450
3451 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3452                         struct btrfs_root *root, struct extent_buffer *leaf)
3453 {
3454         u64 leaf_owner;
3455         u64 leaf_generation;
3456         struct btrfs_key key;
3457         struct btrfs_file_extent_item *fi;
3458         int i;
3459         int nritems;
3460         int ret;
3461
3462         BUG_ON(!btrfs_is_leaf(leaf));
3463         nritems = btrfs_header_nritems(leaf);
3464         leaf_owner = btrfs_header_owner(leaf);
3465         leaf_generation = btrfs_header_generation(leaf);
3466
3467         for (i = 0; i < nritems; i++) {
3468                 u64 disk_bytenr;
3469                 cond_resched();
3470
3471                 btrfs_item_key_to_cpu(leaf, &key, i);
3472                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3473                         continue;
3474                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3475                 if (btrfs_file_extent_type(leaf, fi) ==
3476                     BTRFS_FILE_EXTENT_INLINE)
3477                         continue;
3478                 /*
3479                  * FIXME make sure to insert a trans record that
3480                  * repeats the snapshot del on crash
3481                  */
3482                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3483                 if (disk_bytenr == 0)
3484                         continue;
3485
3486                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3487                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3488                                 leaf->start, leaf_owner, leaf_generation,
3489                                 key.objectid, 0);
3490                 BUG_ON(ret);
3491
3492                 atomic_inc(&root->fs_info->throttle_gen);
3493                 wake_up(&root->fs_info->transaction_throttle);
3494                 cond_resched();
3495         }
3496         return 0;
3497 }
3498
3499 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3500                                         struct btrfs_root *root,
3501                                         struct btrfs_leaf_ref *ref)
3502 {
3503         int i;
3504         int ret;
3505         struct btrfs_extent_info *info = ref->extents;
3506
3507         for (i = 0; i < ref->nritems; i++) {
3508                 ret = __btrfs_free_extent(trans, root, info->bytenr,
3509                                           info->num_bytes, ref->bytenr,
3510                                           ref->owner, ref->generation,
3511                                           info->objectid, 0);
3512
3513                 atomic_inc(&root->fs_info->throttle_gen);
3514                 wake_up(&root->fs_info->transaction_throttle);
3515                 cond_resched();
3516
3517                 BUG_ON(ret);
3518                 info++;
3519         }
3520
3521         return 0;
3522 }
3523
3524 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3525                                      u64 len, u32 *refs)
3526 {
3527         int ret;
3528
3529         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3530         BUG_ON(ret);
3531
3532 #if 0 /* some debugging code in case we see problems here */
3533         /* if the refs count is one, it won't get increased again.  But
3534          * if the ref count is > 1, someone may be decreasing it at
3535          * the same time we are.
3536          */
3537         if (*refs != 1) {
3538                 struct extent_buffer *eb = NULL;
3539                 eb = btrfs_find_create_tree_block(root, start, len);
3540                 if (eb)
3541                         btrfs_tree_lock(eb);
3542
3543                 mutex_lock(&root->fs_info->alloc_mutex);
3544                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3545                 BUG_ON(ret);
3546                 mutex_unlock(&root->fs_info->alloc_mutex);
3547
3548                 if (eb) {
3549                         btrfs_tree_unlock(eb);
3550                         free_extent_buffer(eb);
3551                 }
3552                 if (*refs == 1) {
3553                         printk(KERN_ERR "btrfs block %llu went down to one "
3554                                "during drop_snap\n", (unsigned long long)start);
3555                 }
3556
3557         }
3558 #endif
3559
3560         cond_resched();
3561         return ret;
3562 }
3563
3564 /*
3565  * helper function for drop_snapshot, this walks down the tree dropping ref
3566  * counts as it goes.
3567  */
3568 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3569                                    struct btrfs_root *root,
3570                                    struct btrfs_path *path, int *level)
3571 {
3572         u64 root_owner;
3573         u64 root_gen;
3574         u64 bytenr;
3575         u64 ptr_gen;
3576         struct extent_buffer *next;
3577         struct extent_buffer *cur;
3578         struct extent_buffer *parent;
3579         struct btrfs_leaf_ref *ref;
3580         u32 blocksize;
3581         int ret;
3582         u32 refs;
3583
3584         WARN_ON(*level < 0);
3585         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3586         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3587                                 path->nodes[*level]->len, &refs);
3588         BUG_ON(ret);
3589         if (refs > 1)
3590                 goto out;
3591
3592         /*
3593          * walk down to the last node level and free all the leaves
3594          */
3595         while (*level >= 0) {
3596                 WARN_ON(*level < 0);
3597                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3598                 cur = path->nodes[*level];
3599
3600                 if (btrfs_header_level(cur) != *level)
3601                         WARN_ON(1);
3602
3603                 if (path->slots[*level] >=
3604                     btrfs_header_nritems(cur))
3605                         break;
3606                 if (*level == 0) {
3607                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3608                         BUG_ON(ret);
3609                         break;
3610                 }
3611                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3612                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3613                 blocksize = btrfs_level_size(root, *level - 1);
3614
3615                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3616                 BUG_ON(ret);
3617                 if (refs != 1) {
3618                         parent = path->nodes[*level];
3619                         root_owner = btrfs_header_owner(parent);
3620                         root_gen = btrfs_header_generation(parent);
3621                         path->slots[*level]++;
3622
3623                         ret = __btrfs_free_extent(trans, root, bytenr,
3624                                                 blocksize, parent->start,
3625                                                 root_owner, root_gen,
3626                                                 *level - 1, 1);
3627                         BUG_ON(ret);
3628
3629                         atomic_inc(&root->fs_info->throttle_gen);
3630                         wake_up(&root->fs_info->transaction_throttle);
3631                         cond_resched();
3632
3633                         continue;
3634                 }
3635                 /*
3636                  * at this point, we have a single ref, and since the
3637                  * only place referencing this extent is a dead root
3638                  * the reference count should never go higher.
3639                  * So, we don't need to check it again
3640                  */
3641                 if (*level == 1) {
3642                         ref = btrfs_lookup_leaf_ref(root, bytenr);
3643                         if (ref && ref->generation != ptr_gen) {
3644                                 btrfs_free_leaf_ref(root, ref);
3645                                 ref = NULL;
3646                         }
3647                         if (ref) {
3648                                 ret = cache_drop_leaf_ref(trans, root, ref);
3649                                 BUG_ON(ret);
3650                                 btrfs_remove_leaf_ref(root, ref);
3651                                 btrfs_free_leaf_ref(root, ref);
3652                                 *level = 0;
3653                                 break;
3654                         }
3655                 }
3656                 next = btrfs_find_tree_block(root, bytenr, blocksize);
3657                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3658                         free_extent_buffer(next);
3659
3660                         next = read_tree_block(root, bytenr, blocksize,
3661                                                ptr_gen);
3662                         cond_resched();
3663 #if 0
3664                         /*
3665                          * this is a debugging check and can go away
3666                          * the ref should never go all the way down to 1
3667                          * at this point
3668                          */
3669                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3670                                                 &refs);
3671                         BUG_ON(ret);
3672                         WARN_ON(refs != 1);
3673 #endif
3674                 }
3675                 WARN_ON(*level <= 0);
3676                 if (path->nodes[*level-1])
3677                         free_extent_buffer(path->nodes[*level-1]);
3678                 path->nodes[*level-1] = next;
3679                 *level = btrfs_header_level(next);
3680                 path->slots[*level] = 0;
3681                 cond_resched();
3682         }
3683 out:
3684         WARN_ON(*level < 0);
3685         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3686
3687         if (path->nodes[*level] == root->node) {
3688                 parent = path->nodes[*level];
3689                 bytenr = path->nodes[*level]->start;
3690         } else {
3691                 parent = path->nodes[*level + 1];
3692                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3693         }
3694
3695         blocksize = btrfs_level_size(root, *level);
3696         root_owner = btrfs_header_owner(parent);
3697         root_gen = btrfs_header_generation(parent);
3698
3699         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3700                                   parent->start, root_owner, root_gen,
3701                                   *level, 1);
3702         free_extent_buffer(path->nodes[*level]);
3703         path->nodes[*level] = NULL;
3704         *level += 1;
3705         BUG_ON(ret);
3706
3707         cond_resched();
3708         return 0;
3709 }
3710
3711 /*
3712  * helper function for drop_subtree, this function is similar to
3713  * walk_down_tree. The main difference is that it checks reference
3714  * counts while tree blocks are locked.
3715  */
3716 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3717                                       struct btrfs_root *root,
3718                                       struct btrfs_path *path, int *level)
3719 {
3720         struct extent_buffer *next;
3721         struct extent_buffer *cur;
3722         struct extent_buffer *parent;
3723         u64 bytenr;
3724         u64 ptr_gen;
3725         u32 blocksize;
3726         u32 refs;
3727         int ret;
3728
3729         cur = path->nodes[*level];
3730         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3731                                       &refs);
3732         BUG_ON(ret);
3733         if (refs > 1)
3734                 goto out;
3735
3736         while (*level >= 0) {
3737                 cur = path->nodes[*level];
3738                 if (*level == 0) {
3739                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3740                         BUG_ON(ret);
3741                         clean_tree_block(trans, root, cur);
3742                         break;
3743                 }
3744                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3745                         clean_tree_block(trans, root, cur);
3746                         break;
3747                 }
3748
3749                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3750                 blocksize = btrfs_level_size(root, *level - 1);
3751                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3752
3753                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3754                 btrfs_tree_lock(next);
3755
3756                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3757                                               &refs);
3758                 BUG_ON(ret);
3759                 if (refs > 1) {
3760                         parent = path->nodes[*level];
3761                         ret = btrfs_free_extent(trans, root, bytenr,
3762                                         blocksize, parent->start,
3763                                         btrfs_header_owner(parent),
3764                                         btrfs_header_generation(parent),
3765                                         *level - 1, 1);
3766                         BUG_ON(ret);
3767                         path->slots[*level]++;
3768                         btrfs_tree_unlock(next);
3769                         free_extent_buffer(next);
3770                         continue;
3771                 }
3772
3773                 *level = btrfs_header_level(next);
3774                 path->nodes[*level] = next;
3775                 path->slots[*level] = 0;
3776                 path->locks[*level] = 1;
3777                 cond_resched();
3778         }
3779 out:
3780         parent = path->nodes[*level + 1];
3781         bytenr = path->nodes[*level]->start;
3782         blocksize = path->nodes[*level]->len;
3783
3784         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3785                         parent->start, btrfs_header_owner(parent),
3786                         btrfs_header_generation(parent), *level, 1);
3787         BUG_ON(ret);
3788
3789         if (path->locks[*level]) {
3790                 btrfs_tree_unlock(path->nodes[*level]);
3791                 path->locks[*level] = 0;
3792         }
3793         free_extent_buffer(path->nodes[*level]);
3794         path->nodes[*level] = NULL;
3795         *level += 1;
3796         cond_resched();
3797         return 0;
3798 }
3799
3800 /*
3801  * helper for dropping snapshots.  This walks back up the tree in the path
3802  * to find the first node higher up where we haven't yet gone through
3803  * all the slots
3804  */
3805 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3806                                  struct btrfs_root *root,
3807                                  struct btrfs_path *path,
3808                                  int *level, int max_level)
3809 {
3810         u64 root_owner;
3811         u64 root_gen;
3812         struct btrfs_root_item *root_item = &root->root_item;
3813         int i;
3814         int slot;
3815         int ret;
3816
3817         for (i = *level; i < max_level && path->nodes[i]; i++) {
3818                 slot = path->slots[i];
3819                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3820                         struct extent_buffer *node;
3821                         struct btrfs_disk_key disk_key;
3822                         node = path->nodes[i];
3823                         path->slots[i]++;
3824                         *level = i;
3825                         WARN_ON(*level == 0);
3826                         btrfs_node_key(node, &disk_key, path->slots[i]);
3827                         memcpy(&root_item->drop_progress,
3828                                &disk_key, sizeof(disk_key));
3829                         root_item->drop_level = i;
3830                         return 0;
3831                 } else {
3832                         struct extent_buffer *parent;
3833                         if (path->nodes[*level] == root->node)
3834                                 parent = path->nodes[*level];
3835                         else
3836                                 parent = path->nodes[*level + 1];
3837
3838                         root_owner = btrfs_header_owner(parent);
3839                         root_gen = btrfs_header_generation(parent);
3840
3841                         clean_tree_block(trans, root, path->nodes[*level]);
3842                         ret = btrfs_free_extent(trans, root,
3843                                                 path->nodes[*level]->start,
3844                                                 path->nodes[*level]->len,
3845                                                 parent->start, root_owner,
3846                                                 root_gen, *level, 1);
3847                         BUG_ON(ret);
3848                         if (path->locks[*level]) {
3849                                 btrfs_tree_unlock(path->nodes[*level]);
3850                                 path->locks[*level] = 0;
3851                         }
3852                         free_extent_buffer(path->nodes[*level]);
3853                         path->nodes[*level] = NULL;
3854                         *level = i + 1;
3855                 }
3856         }
3857         return 1;
3858 }
3859
3860 /*
3861  * drop the reference count on the tree rooted at 'snap'.  This traverses
3862  * the tree freeing any blocks that have a ref count of zero after being
3863  * decremented.
3864  */
3865 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3866                         *root)
3867 {
3868         int ret = 0;
3869         int wret;
3870         int level;
3871         struct btrfs_path *path;
3872         int i;
3873         int orig_level;
3874         struct btrfs_root_item *root_item = &root->root_item;
3875
3876         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3877         path = btrfs_alloc_path();
3878         BUG_ON(!path);
3879
3880         level = btrfs_header_level(root->node);
3881         orig_level = level;
3882         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3883                 path->nodes[level] = root->node;
3884                 extent_buffer_get(root->node);
3885                 path->slots[level] = 0;
3886         } else {
3887                 struct btrfs_key key;
3888                 struct btrfs_disk_key found_key;
3889                 struct extent_buffer *node;
3890
3891                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3892                 level = root_item->drop_level;
3893                 path->lowest_level = level;
3894                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3895                 if (wret < 0) {
3896                         ret = wret;
3897                         goto out;
3898                 }
3899                 node = path->nodes[level];
3900                 btrfs_node_key(node, &found_key, path->slots[level]);
3901                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3902                                sizeof(found_key)));
3903                 /*
3904                  * unlock our path, this is safe because only this
3905                  * function is allowed to delete this snapshot
3906                  */
3907                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3908                         if (path->nodes[i] && path->locks[i]) {
3909                                 path->locks[i] = 0;
3910                                 btrfs_tree_unlock(path->nodes[i]);
3911                         }
3912                 }
3913         }
3914         while (1) {
3915                 wret = walk_down_tree(trans, root, path, &level);
3916                 if (wret > 0)
3917                         break;
3918                 if (wret < 0)
3919                         ret = wret;
3920
3921                 wret = walk_up_tree(trans, root, path, &level,
3922                                     BTRFS_MAX_LEVEL);
3923                 if (wret > 0)
3924                         break;
3925                 if (wret < 0)
3926                         ret = wret;
3927                 if (trans->transaction->in_commit) {
3928                         ret = -EAGAIN;
3929                         break;
3930                 }
3931                 atomic_inc(&root->fs_info->throttle_gen);
3932                 wake_up(&root->fs_info->transaction_throttle);
3933         }
3934         for (i = 0; i <= orig_level; i++) {
3935                 if (path->nodes[i]) {
3936                         free_extent_buffer(path->nodes[i]);
3937                         path->nodes[i] = NULL;
3938                 }
3939         }
3940 out:
3941         btrfs_free_path(path);
3942         return ret;
3943 }
3944
3945 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3946                         struct btrfs_root *root,
3947                         struct extent_buffer *node,
3948                         struct extent_buffer *parent)
3949 {
3950         struct btrfs_path *path;
3951         int level;
3952         int parent_level;
3953         int ret = 0;
3954         int wret;
3955
3956         path = btrfs_alloc_path();
3957         BUG_ON(!path);
3958
3959         BUG_ON(!btrfs_tree_locked(parent));
3960         parent_level = btrfs_header_level(parent);
3961         extent_buffer_get(parent);
3962         path->nodes[parent_level] = parent;
3963         path->slots[parent_level] = btrfs_header_nritems(parent);
3964
3965         BUG_ON(!btrfs_tree_locked(node));
3966         level = btrfs_header_level(node);
3967         extent_buffer_get(node);
3968         path->nodes[level] = node;
3969         path->slots[level] = 0;
3970
3971         while (1) {
3972                 wret = walk_down_subtree(trans, root, path, &level);
3973                 if (wret < 0)
3974                         ret = wret;
3975                 if (wret != 0)
3976                         break;
3977
3978                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3979                 if (wret < 0)
3980                         ret = wret;
3981                 if (wret != 0)
3982                         break;
3983         }
3984
3985         btrfs_free_path(path);
3986         return ret;
3987 }
3988
3989 static unsigned long calc_ra(unsigned long start, unsigned long last,
3990                              unsigned long nr)
3991 {
3992         return min(last, start + nr - 1);
3993 }
3994
3995 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3996                                          u64 len)
3997 {
3998         u64 page_start;
3999         u64 page_end;
4000         unsigned long first_index;
4001         unsigned long last_index;
4002         unsigned long i;
4003         struct page *page;
4004         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4005         struct file_ra_state *ra;
4006         struct btrfs_ordered_extent *ordered;
4007         unsigned int total_read = 0;
4008         unsigned int total_dirty = 0;
4009         int ret = 0;
4010
4011         ra = kzalloc(sizeof(*ra), GFP_NOFS);
4012
4013         mutex_lock(&inode->i_mutex);
4014         first_index = start >> PAGE_CACHE_SHIFT;
4015         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
4016
4017         /* make sure the dirty trick played by the caller work */
4018         ret = invalidate_inode_pages2_range(inode->i_mapping,
4019                                             first_index, last_index);
4020         if (ret)
4021                 goto out_unlock;
4022
4023         file_ra_state_init(ra, inode->i_mapping);
4024
4025         for (i = first_index ; i <= last_index; i++) {
4026                 if (total_read % ra->ra_pages == 0) {
4027                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
4028                                        calc_ra(i, last_index, ra->ra_pages));
4029                 }
4030                 total_read++;
4031 again:
4032                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
4033                         BUG_ON(1);
4034                 page = grab_cache_page(inode->i_mapping, i);
4035                 if (!page) {
4036                         ret = -ENOMEM;
4037                         goto out_unlock;
4038                 }
4039                 if (!PageUptodate(page)) {
4040                         btrfs_readpage(NULL, page);
4041                         lock_page(page);
4042                         if (!PageUptodate(page)) {
4043                                 unlock_page(page);
4044                                 page_cache_release(page);
4045                                 ret = -EIO;
4046                                 goto out_unlock;
4047                         }
4048                 }
4049                 wait_on_page_writeback(page);
4050
4051                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
4052                 page_end = page_start + PAGE_CACHE_SIZE - 1;
4053                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4054
4055                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4056                 if (ordered) {
4057                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4058                         unlock_page(page);
4059                         page_cache_release(page);
4060                         btrfs_start_ordered_extent(inode, ordered, 1);
4061                         btrfs_put_ordered_extent(ordered);
4062                         goto again;
4063                 }
4064                 set_page_extent_mapped(page);
4065
4066                 if (i == first_index)
4067                         set_extent_bits(io_tree, page_start, page_end,
4068                                         EXTENT_BOUNDARY, GFP_NOFS);
4069                 btrfs_set_extent_delalloc(inode, page_start, page_end);
4070
4071                 set_page_dirty(page);
4072                 total_dirty++;
4073
4074                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4075                 unlock_page(page);
4076                 page_cache_release(page);
4077         }
4078
4079 out_unlock:
4080         kfree(ra);
4081         mutex_unlock(&inode->i_mutex);
4082         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4083         return ret;
4084 }
4085
4086 static noinline int relocate_data_extent(struct inode *reloc_inode,
4087                                          struct btrfs_key *extent_key,
4088                                          u64 offset)
4089 {
4090         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4091         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4092         struct extent_map *em;
4093         u64 start = extent_key->objectid - offset;
4094         u64 end = start + extent_key->offset - 1;
4095
4096         em = alloc_extent_map(GFP_NOFS);
4097         BUG_ON(!em || IS_ERR(em));
4098
4099         em->start = start;
4100         em->len = extent_key->offset;
4101         em->block_len = extent_key->offset;
4102         em->block_start = extent_key->objectid;
4103         em->bdev = root->fs_info->fs_devices->latest_bdev;
4104         set_bit(EXTENT_FLAG_PINNED, &em->flags);
4105
4106         /* setup extent map to cheat btrfs_readpage */
4107         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4108         while (1) {
4109                 int ret;
4110                 spin_lock(&em_tree->lock);
4111                 ret = add_extent_mapping(em_tree, em);
4112                 spin_unlock(&em_tree->lock);
4113                 if (ret != -EEXIST) {
4114                         free_extent_map(em);
4115                         break;
4116                 }
4117                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4118         }
4119         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4120
4121         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4122 }
4123
4124 struct btrfs_ref_path {
4125         u64 extent_start;
4126         u64 nodes[BTRFS_MAX_LEVEL];
4127         u64 root_objectid;
4128         u64 root_generation;
4129         u64 owner_objectid;
4130         u32 num_refs;
4131         int lowest_level;
4132         int current_level;
4133         int shared_level;
4134
4135         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4136         u64 new_nodes[BTRFS_MAX_LEVEL];
4137 };
4138
4139 struct disk_extent {
4140         u64 ram_bytes;
4141         u64 disk_bytenr;
4142         u64 disk_num_bytes;
4143         u64 offset;
4144         u64 num_bytes;
4145         u8 compression;
4146         u8 encryption;
4147         u16 other_encoding;
4148 };
4149
4150 static int is_cowonly_root(u64 root_objectid)
4151 {
4152         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4153             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4154             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4155             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4156             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4157             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4158                 return 1;
4159         return 0;
4160 }
4161
4162 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4163                                     struct btrfs_root *extent_root,
4164                                     struct btrfs_ref_path *ref_path,
4165                                     int first_time)
4166 {
4167         struct extent_buffer *leaf;
4168         struct btrfs_path *path;
4169         struct btrfs_extent_ref *ref;
4170         struct btrfs_key key;
4171         struct btrfs_key found_key;
4172         u64 bytenr;
4173         u32 nritems;
4174         int level;
4175         int ret = 1;
4176
4177         path = btrfs_alloc_path();
4178         if (!path)
4179                 return -ENOMEM;
4180
4181         if (first_time) {
4182                 ref_path->lowest_level = -1;
4183                 ref_path->current_level = -1;
4184                 ref_path->shared_level = -1;
4185                 goto walk_up;
4186         }
4187 walk_down:
4188         level = ref_path->current_level - 1;
4189         while (level >= -1) {
4190                 u64 parent;
4191                 if (level < ref_path->lowest_level)
4192                         break;
4193
4194                 if (level >= 0)
4195                         bytenr = ref_path->nodes[level];
4196                 else
4197                         bytenr = ref_path->extent_start;
4198                 BUG_ON(bytenr == 0);
4199
4200                 parent = ref_path->nodes[level + 1];
4201                 ref_path->nodes[level + 1] = 0;
4202                 ref_path->current_level = level;
4203                 BUG_ON(parent == 0);
4204
4205                 key.objectid = bytenr;
4206                 key.offset = parent + 1;
4207                 key.type = BTRFS_EXTENT_REF_KEY;
4208
4209                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4210                 if (ret < 0)
4211                         goto out;
4212                 BUG_ON(ret == 0);
4213
4214                 leaf = path->nodes[0];
4215                 nritems = btrfs_header_nritems(leaf);
4216                 if (path->slots[0] >= nritems) {
4217                         ret = btrfs_next_leaf(extent_root, path);
4218                         if (ret < 0)
4219                                 goto out;
4220                         if (ret > 0)
4221                                 goto next;
4222                         leaf = path->nodes[0];
4223                 }
4224
4225                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4226                 if (found_key.objectid == bytenr &&
4227                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4228                         if (level < ref_path->shared_level)
4229                                 ref_path->shared_level = level;
4230                         goto found;
4231                 }
4232 next:
4233                 level--;
4234                 btrfs_release_path(extent_root, path);
4235                 cond_resched();
4236         }
4237         /* reached lowest level */
4238         ret = 1;
4239         goto out;
4240 walk_up:
4241         level = ref_path->current_level;
4242         while (level < BTRFS_MAX_LEVEL - 1) {
4243                 u64 ref_objectid;
4244
4245                 if (level >= 0)
4246                         bytenr = ref_path->nodes[level];
4247                 else
4248                         bytenr = ref_path->extent_start;
4249
4250                 BUG_ON(bytenr == 0);
4251
4252                 key.objectid = bytenr;
4253                 key.offset = 0;
4254                 key.type = BTRFS_EXTENT_REF_KEY;
4255
4256                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4257                 if (ret < 0)
4258                         goto out;
4259
4260                 leaf = path->nodes[0];
4261                 nritems = btrfs_header_nritems(leaf);
4262                 if (path->slots[0] >= nritems) {
4263                         ret = btrfs_next_leaf(extent_root, path);
4264                         if (ret < 0)
4265                                 goto out;
4266                         if (ret > 0) {
4267                                 /* the extent was freed by someone */
4268                                 if (ref_path->lowest_level == level)
4269                                         goto out;
4270                                 btrfs_release_path(extent_root, path);
4271                                 goto walk_down;
4272                         }
4273                         leaf = path->nodes[0];
4274                 }
4275
4276                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4277                 if (found_key.objectid != bytenr ||
4278                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4279                         /* the extent was freed by someone */
4280                         if (ref_path->lowest_level == level) {
4281                                 ret = 1;
4282                                 goto out;
4283                         }
4284                         btrfs_release_path(extent_root, path);
4285                         goto walk_down;
4286                 }
4287 found:
4288                 ref = btrfs_item_ptr(leaf, path->slots[0],
4289                                 struct btrfs_extent_ref);
4290                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4291                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4292                         if (first_time) {
4293                                 level = (int)ref_objectid;
4294                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4295                                 ref_path->lowest_level = level;
4296                                 ref_path->current_level = level;
4297                                 ref_path->nodes[level] = bytenr;
4298                         } else {
4299                                 WARN_ON(ref_objectid != level);
4300                         }
4301                 } else {
4302                         WARN_ON(level != -1);
4303                 }
4304                 first_time = 0;
4305
4306                 if (ref_path->lowest_level == level) {
4307                         ref_path->owner_objectid = ref_objectid;
4308                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4309                 }
4310
4311                 /*
4312                  * the block is tree root or the block isn't in reference
4313                  * counted tree.
4314                  */
4315                 if (found_key.objectid == found_key.offset ||
4316                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4317                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4318                         ref_path->root_generation =
4319                                 btrfs_ref_generation(leaf, ref);
4320                         if (level < 0) {
4321                                 /* special reference from the tree log */
4322                                 ref_path->nodes[0] = found_key.offset;
4323                                 ref_path->current_level = 0;
4324                         }
4325                         ret = 0;
4326                         goto out;
4327                 }
4328
4329                 level++;
4330                 BUG_ON(ref_path->nodes[level] != 0);
4331                 ref_path->nodes[level] = found_key.offset;
4332                 ref_path->current_level = level;
4333
4334                 /*
4335                  * the reference was created in the running transaction,
4336                  * no need to continue walking up.
4337                  */
4338                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4339                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4340                         ref_path->root_generation =
4341                                 btrfs_ref_generation(leaf, ref);
4342                         ret = 0;
4343                         goto out;
4344                 }
4345
4346                 btrfs_release_path(extent_root, path);
4347                 cond_resched();
4348         }
4349         /* reached max tree level, but no tree root found. */
4350         BUG();
4351 out:
4352         btrfs_free_path(path);
4353         return ret;
4354 }
4355
4356 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4357                                 struct btrfs_root *extent_root,
4358                                 struct btrfs_ref_path *ref_path,
4359                                 u64 extent_start)
4360 {
4361         memset(ref_path, 0, sizeof(*ref_path));
4362         ref_path->extent_start = extent_start;
4363
4364         return __next_ref_path(trans, extent_root, ref_path, 1);
4365 }
4366
4367 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4368                                struct btrfs_root *extent_root,
4369                                struct btrfs_ref_path *ref_path)
4370 {
4371         return __next_ref_path(trans, extent_root, ref_path, 0);
4372 }
4373
4374 static noinline int get_new_locations(struct inode *reloc_inode,
4375                                       struct btrfs_key *extent_key,
4376                                       u64 offset, int no_fragment,
4377                                       struct disk_extent **extents,
4378                                       int *nr_extents)
4379 {
4380         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4381         struct btrfs_path *path;
4382         struct btrfs_file_extent_item *fi;
4383         struct extent_buffer *leaf;
4384         struct disk_extent *exts = *extents;
4385         struct btrfs_key found_key;
4386         u64 cur_pos;
4387         u64 last_byte;
4388         u32 nritems;
4389         int nr = 0;
4390         int max = *nr_extents;
4391         int ret;
4392
4393         WARN_ON(!no_fragment && *extents);
4394         if (!exts) {
4395                 max = 1;
4396                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4397                 if (!exts)
4398                         return -ENOMEM;
4399         }
4400
4401         path = btrfs_alloc_path();
4402         BUG_ON(!path);
4403
4404         cur_pos = extent_key->objectid - offset;
4405         last_byte = extent_key->objectid + extent_key->offset;
4406         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4407                                        cur_pos, 0);
4408         if (ret < 0)
4409                 goto out;
4410         if (ret > 0) {
4411                 ret = -ENOENT;
4412                 goto out;
4413         }
4414
4415         while (1) {
4416                 leaf = path->nodes[0];
4417                 nritems = btrfs_header_nritems(leaf);
4418                 if (path->slots[0] >= nritems) {
4419                         ret = btrfs_next_leaf(root, path);
4420                         if (ret < 0)
4421                                 goto out;
4422                         if (ret > 0)
4423                                 break;
4424                         leaf = path->nodes[0];
4425                 }
4426
4427                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4428                 if (found_key.offset != cur_pos ||
4429                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4430                     found_key.objectid != reloc_inode->i_ino)
4431                         break;
4432
4433                 fi = btrfs_item_ptr(leaf, path->slots[0],
4434                                     struct btrfs_file_extent_item);
4435                 if (btrfs_file_extent_type(leaf, fi) !=
4436                     BTRFS_FILE_EXTENT_REG ||
4437                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4438                         break;
4439
4440                 if (nr == max) {
4441                         struct disk_extent *old = exts;
4442                         max *= 2;
4443                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4444                         memcpy(exts, old, sizeof(*exts) * nr);
4445                         if (old != *extents)
4446                                 kfree(old);
4447                 }
4448
4449                 exts[nr].disk_bytenr =
4450                         btrfs_file_extent_disk_bytenr(leaf, fi);
4451                 exts[nr].disk_num_bytes =
4452                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4453                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4454                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4455                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4456                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4457                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4458                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4459                                                                            fi);
4460                 BUG_ON(exts[nr].offset > 0);
4461                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4462                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4463
4464                 cur_pos += exts[nr].num_bytes;
4465                 nr++;
4466
4467                 if (cur_pos + offset >= last_byte)
4468                         break;
4469
4470                 if (no_fragment) {
4471                         ret = 1;
4472                         goto out;
4473                 }
4474                 path->slots[0]++;
4475         }
4476
4477         BUG_ON(cur_pos + offset > last_byte);
4478         if (cur_pos + offset < last_byte) {
4479                 ret = -ENOENT;
4480                 goto out;
4481         }
4482         ret = 0;
4483 out:
4484         btrfs_free_path(path);
4485         if (ret) {
4486                 if (exts != *extents)
4487                         kfree(exts);
4488         } else {
4489                 *extents = exts;
4490                 *nr_extents = nr;
4491         }
4492         return ret;
4493 }
4494
4495 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4496                                         struct btrfs_root *root,
4497                                         struct btrfs_path *path,
4498                                         struct btrfs_key *extent_key,
4499                                         struct btrfs_key *leaf_key,
4500                                         struct btrfs_ref_path *ref_path,
4501                                         struct disk_extent *new_extents,
4502                                         int nr_extents)
4503 {
4504         struct extent_buffer *leaf;
4505         struct btrfs_file_extent_item *fi;
4506         struct inode *inode = NULL;
4507         struct btrfs_key key;
4508         u64 lock_start = 0;
4509         u64 lock_end = 0;
4510         u64 num_bytes;
4511         u64 ext_offset;
4512         u64 search_end = (u64)-1;
4513         u32 nritems;
4514         int nr_scaned = 0;
4515         int extent_locked = 0;
4516         int extent_type;
4517         int ret;
4518
4519         memcpy(&key, leaf_key, sizeof(key));
4520         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4521                 if (key.objectid < ref_path->owner_objectid ||
4522                     (key.objectid == ref_path->owner_objectid &&
4523                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4524                         key.objectid = ref_path->owner_objectid;
4525                         key.type = BTRFS_EXTENT_DATA_KEY;
4526                         key.offset = 0;
4527                 }
4528         }
4529
4530         while (1) {
4531                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4532                 if (ret < 0)
4533                         goto out;
4534
4535                 leaf = path->nodes[0];
4536                 nritems = btrfs_header_nritems(leaf);
4537 next:
4538                 if (extent_locked && ret > 0) {
4539                         /*
4540                          * the file extent item was modified by someone
4541                          * before the extent got locked.
4542                          */
4543                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4544                                       lock_end, GFP_NOFS);
4545                         extent_locked = 0;
4546                 }
4547
4548                 if (path->slots[0] >= nritems) {
4549                         if (++nr_scaned > 2)
4550                                 break;
4551
4552                         BUG_ON(extent_locked);
4553                         ret = btrfs_next_leaf(root, path);
4554                         if (ret < 0)
4555                                 goto out;
4556                         if (ret > 0)
4557                                 break;
4558                         leaf = path->nodes[0];
4559                         nritems = btrfs_header_nritems(leaf);
4560                 }
4561
4562                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4563
4564                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4565                         if ((key.objectid > ref_path->owner_objectid) ||
4566                             (key.objectid == ref_path->owner_objectid &&
4567                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4568                             key.offset >= search_end)
4569                                 break;
4570                 }
4571
4572                 if (inode && key.objectid != inode->i_ino) {
4573                         BUG_ON(extent_locked);
4574                         btrfs_release_path(root, path);
4575                         mutex_unlock(&inode->i_mutex);
4576                         iput(inode);
4577                         inode = NULL;
4578                         continue;
4579                 }
4580
4581                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4582                         path->slots[0]++;
4583                         ret = 1;
4584                         goto next;
4585                 }
4586                 fi = btrfs_item_ptr(leaf, path->slots[0],
4587                                     struct btrfs_file_extent_item);
4588                 extent_type = btrfs_file_extent_type(leaf, fi);
4589                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4590                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4591                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4592                      extent_key->objectid)) {
4593                         path->slots[0]++;
4594                         ret = 1;
4595                         goto next;
4596                 }
4597
4598                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4599                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4600
4601                 if (search_end == (u64)-1) {
4602                         search_end = key.offset - ext_offset +
4603                                 btrfs_file_extent_ram_bytes(leaf, fi);
4604                 }
4605
4606                 if (!extent_locked) {
4607                         lock_start = key.offset;
4608                         lock_end = lock_start + num_bytes - 1;
4609                 } else {
4610                         if (lock_start > key.offset ||
4611                             lock_end + 1 < key.offset + num_bytes) {
4612                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4613                                               lock_start, lock_end, GFP_NOFS);
4614                                 extent_locked = 0;
4615                         }
4616                 }
4617
4618                 if (!inode) {
4619                         btrfs_release_path(root, path);
4620
4621                         inode = btrfs_iget_locked(root->fs_info->sb,
4622                                                   key.objectid, root);
4623                         if (inode->i_state & I_NEW) {
4624                                 BTRFS_I(inode)->root = root;
4625                                 BTRFS_I(inode)->location.objectid =
4626                                         key.objectid;
4627                                 BTRFS_I(inode)->location.type =
4628                                         BTRFS_INODE_ITEM_KEY;
4629                                 BTRFS_I(inode)->location.offset = 0;
4630                                 btrfs_read_locked_inode(inode);
4631                                 unlock_new_inode(inode);
4632                         }
4633                         /*
4634                          * some code call btrfs_commit_transaction while
4635                          * holding the i_mutex, so we can't use mutex_lock
4636                          * here.
4637                          */
4638                         if (is_bad_inode(inode) ||
4639                             !mutex_trylock(&inode->i_mutex)) {
4640                                 iput(inode);
4641                                 inode = NULL;
4642                                 key.offset = (u64)-1;
4643                                 goto skip;
4644                         }
4645                 }
4646
4647                 if (!extent_locked) {
4648                         struct btrfs_ordered_extent *ordered;
4649
4650                         btrfs_release_path(root, path);
4651
4652                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4653                                     lock_end, GFP_NOFS);
4654                         ordered = btrfs_lookup_first_ordered_extent(inode,
4655                                                                     lock_end);
4656                         if (ordered &&
4657                             ordered->file_offset <= lock_end &&
4658                             ordered->file_offset + ordered->len > lock_start) {
4659                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4660                                               lock_start, lock_end, GFP_NOFS);
4661                                 btrfs_start_ordered_extent(inode, ordered, 1);
4662                                 btrfs_put_ordered_extent(ordered);
4663                                 key.offset += num_bytes;
4664                                 goto skip;
4665                         }
4666                         if (ordered)
4667                                 btrfs_put_ordered_extent(ordered);
4668
4669                         extent_locked = 1;
4670                         continue;
4671                 }
4672
4673                 if (nr_extents == 1) {
4674                         /* update extent pointer in place */
4675                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4676                                                 new_extents[0].disk_bytenr);
4677                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4678                                                 new_extents[0].disk_num_bytes);
4679                         btrfs_mark_buffer_dirty(leaf);
4680
4681                         btrfs_drop_extent_cache(inode, key.offset,
4682                                                 key.offset + num_bytes - 1, 0);
4683
4684                         ret = btrfs_inc_extent_ref(trans, root,
4685                                                 new_extents[0].disk_bytenr,
4686                                                 new_extents[0].disk_num_bytes,
4687                                                 leaf->start,
4688                                                 root->root_key.objectid,
4689                                                 trans->transid,
4690                                                 key.objectid);
4691                         BUG_ON(ret);
4692
4693                         ret = btrfs_free_extent(trans, root,
4694                                                 extent_key->objectid,
4695                                                 extent_key->offset,
4696                                                 leaf->start,
4697                                                 btrfs_header_owner(leaf),
4698                                                 btrfs_header_generation(leaf),
4699                                                 key.objectid, 0);
4700                         BUG_ON(ret);
4701
4702                         btrfs_release_path(root, path);
4703                         key.offset += num_bytes;
4704                 } else {
4705                         BUG_ON(1);
4706 #if 0
4707                         u64 alloc_hint;
4708                         u64 extent_len;
4709                         int i;
4710                         /*
4711                          * drop old extent pointer at first, then insert the
4712                          * new pointers one bye one
4713                          */
4714                         btrfs_release_path(root, path);
4715                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4716                                                  key.offset + num_bytes,
4717                                                  key.offset, &alloc_hint);
4718                         BUG_ON(ret);
4719
4720                         for (i = 0; i < nr_extents; i++) {
4721                                 if (ext_offset >= new_extents[i].num_bytes) {
4722                                         ext_offset -= new_extents[i].num_bytes;
4723                                         continue;
4724                                 }
4725                                 extent_len = min(new_extents[i].num_bytes -
4726                                                  ext_offset, num_bytes);
4727
4728                                 ret = btrfs_insert_empty_item(trans, root,
4729                                                               path, &key,
4730                                                               sizeof(*fi));
4731                                 BUG_ON(ret);
4732
4733                                 leaf = path->nodes[0];
4734                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4735                                                 struct btrfs_file_extent_item);
4736                                 btrfs_set_file_extent_generation(leaf, fi,
4737                                                         trans->transid);
4738                                 btrfs_set_file_extent_type(leaf, fi,
4739                                                         BTRFS_FILE_EXTENT_REG);
4740                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4741                                                 new_extents[i].disk_bytenr);
4742                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4743                                                 new_extents[i].disk_num_bytes);
4744                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4745                                                 new_extents[i].ram_bytes);
4746
4747                                 btrfs_set_file_extent_compression(leaf, fi,
4748                                                 new_extents[i].compression);
4749                                 btrfs_set_file_extent_encryption(leaf, fi,
4750                                                 new_extents[i].encryption);
4751                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4752                                                 new_extents[i].other_encoding);
4753
4754                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4755                                                         extent_len);
4756                                 ext_offset += new_extents[i].offset;
4757                                 btrfs_set_file_extent_offset(leaf, fi,
4758                                                         ext_offset);
4759                                 btrfs_mark_buffer_dirty(leaf);
4760
4761                                 btrfs_drop_extent_cache(inode, key.offset,
4762                                                 key.offset + extent_len - 1, 0);
4763
4764                                 ret = btrfs_inc_extent_ref(trans, root,
4765                                                 new_extents[i].disk_bytenr,
4766                                                 new_extents[i].disk_num_bytes,
4767                                                 leaf->start,
4768                                                 root->root_key.objectid,
4769                                                 trans->transid, key.objectid);
4770                                 BUG_ON(ret);
4771                                 btrfs_release_path(root, path);
4772
4773                                 inode_add_bytes(inode, extent_len);
4774
4775                                 ext_offset = 0;
4776                                 num_bytes -= extent_len;
4777                                 key.offset += extent_len;
4778
4779                                 if (num_bytes == 0)
4780                                         break;
4781                         }
4782                         BUG_ON(i >= nr_extents);
4783 #endif
4784                 }
4785
4786                 if (extent_locked) {
4787                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4788                                       lock_end, GFP_NOFS);
4789                         extent_locked = 0;
4790                 }
4791 skip:
4792                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4793                     key.offset >= search_end)
4794                         break;
4795
4796                 cond_resched();
4797         }
4798         ret = 0;
4799 out:
4800         btrfs_release_path(root, path);
4801         if (inode) {
4802                 mutex_unlock(&inode->i_mutex);
4803                 if (extent_locked) {
4804                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4805                                       lock_end, GFP_NOFS);
4806                 }
4807                 iput(inode);
4808         }
4809         return ret;
4810 }
4811
4812 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4813                                struct btrfs_root *root,
4814                                struct extent_buffer *buf, u64 orig_start)
4815 {
4816         int level;
4817         int ret;
4818
4819         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4820         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4821
4822         level = btrfs_header_level(buf);
4823         if (level == 0) {
4824                 struct btrfs_leaf_ref *ref;
4825                 struct btrfs_leaf_ref *orig_ref;
4826
4827                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4828                 if (!orig_ref)
4829                         return -ENOENT;
4830
4831                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4832                 if (!ref) {
4833                         btrfs_free_leaf_ref(root, orig_ref);
4834                         return -ENOMEM;
4835                 }
4836
4837                 ref->nritems = orig_ref->nritems;
4838                 memcpy(ref->extents, orig_ref->extents,
4839                         sizeof(ref->extents[0]) * ref->nritems);
4840
4841                 btrfs_free_leaf_ref(root, orig_ref);
4842
4843                 ref->root_gen = trans->transid;
4844                 ref->bytenr = buf->start;
4845                 ref->owner = btrfs_header_owner(buf);
4846                 ref->generation = btrfs_header_generation(buf);
4847                 ret = btrfs_add_leaf_ref(root, ref, 0);
4848                 WARN_ON(ret);
4849                 btrfs_free_leaf_ref(root, ref);
4850         }
4851         return 0;
4852 }
4853
4854 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4855                                         struct extent_buffer *leaf,
4856                                         struct btrfs_block_group_cache *group,
4857                                         struct btrfs_root *target_root)
4858 {
4859         struct btrfs_key key;
4860         struct inode *inode = NULL;
4861         struct btrfs_file_extent_item *fi;
4862         u64 num_bytes;
4863         u64 skip_objectid = 0;
4864         u32 nritems;
4865         u32 i;
4866
4867         nritems = btrfs_header_nritems(leaf);
4868         for (i = 0; i < nritems; i++) {
4869                 btrfs_item_key_to_cpu(leaf, &key, i);
4870                 if (key.objectid == skip_objectid ||
4871                     key.type != BTRFS_EXTENT_DATA_KEY)
4872                         continue;
4873                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4874                 if (btrfs_file_extent_type(leaf, fi) ==
4875                     BTRFS_FILE_EXTENT_INLINE)
4876                         continue;
4877                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4878                         continue;
4879                 if (!inode || inode->i_ino != key.objectid) {
4880                         iput(inode);
4881                         inode = btrfs_ilookup(target_root->fs_info->sb,
4882                                               key.objectid, target_root, 1);
4883                 }
4884                 if (!inode) {
4885                         skip_objectid = key.objectid;
4886                         continue;
4887                 }
4888                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4889
4890                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4891                             key.offset + num_bytes - 1, GFP_NOFS);
4892                 btrfs_drop_extent_cache(inode, key.offset,
4893                                         key.offset + num_bytes - 1, 1);
4894                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4895                               key.offset + num_bytes - 1, GFP_NOFS);
4896                 cond_resched();
4897         }
4898         iput(inode);
4899         return 0;
4900 }
4901
4902 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4903                                         struct btrfs_root *root,
4904                                         struct extent_buffer *leaf,
4905                                         struct btrfs_block_group_cache *group,
4906                                         struct inode *reloc_inode)
4907 {
4908         struct btrfs_key key;
4909         struct btrfs_key extent_key;
4910         struct btrfs_file_extent_item *fi;
4911         struct btrfs_leaf_ref *ref;
4912         struct disk_extent *new_extent;
4913         u64 bytenr;
4914         u64 num_bytes;
4915         u32 nritems;
4916         u32 i;
4917         int ext_index;
4918         int nr_extent;
4919         int ret;
4920
4921         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4922         BUG_ON(!new_extent);
4923
4924         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4925         BUG_ON(!ref);
4926
4927         ext_index = -1;
4928         nritems = btrfs_header_nritems(leaf);
4929         for (i = 0; i < nritems; i++) {
4930                 btrfs_item_key_to_cpu(leaf, &key, i);
4931                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4932                         continue;
4933                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4934                 if (btrfs_file_extent_type(leaf, fi) ==
4935                     BTRFS_FILE_EXTENT_INLINE)
4936                         continue;
4937                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4938                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4939                 if (bytenr == 0)
4940                         continue;
4941
4942                 ext_index++;
4943                 if (bytenr >= group->key.objectid + group->key.offset ||
4944                     bytenr + num_bytes <= group->key.objectid)
4945                         continue;
4946
4947                 extent_key.objectid = bytenr;
4948                 extent_key.offset = num_bytes;
4949                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4950                 nr_extent = 1;
4951                 ret = get_new_locations(reloc_inode, &extent_key,
4952                                         group->key.objectid, 1,
4953                                         &new_extent, &nr_extent);
4954                 if (ret > 0)
4955                         continue;
4956                 BUG_ON(ret < 0);
4957
4958                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4959                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4960                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4961                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4962
4963                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4964                                                 new_extent->disk_bytenr);
4965                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4966                                                 new_extent->disk_num_bytes);
4967                 btrfs_mark_buffer_dirty(leaf);
4968
4969                 ret = btrfs_inc_extent_ref(trans, root,
4970                                         new_extent->disk_bytenr,
4971                                         new_extent->disk_num_bytes,
4972                                         leaf->start,
4973                                         root->root_key.objectid,
4974                                         trans->transid, key.objectid);
4975                 BUG_ON(ret);
4976                 ret = btrfs_free_extent(trans, root,
4977                                         bytenr, num_bytes, leaf->start,
4978                                         btrfs_header_owner(leaf),
4979                                         btrfs_header_generation(leaf),
4980                                         key.objectid, 0);
4981                 BUG_ON(ret);
4982                 cond_resched();
4983         }
4984         kfree(new_extent);
4985         BUG_ON(ext_index + 1 != ref->nritems);
4986         btrfs_free_leaf_ref(root, ref);
4987         return 0;
4988 }
4989
4990 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4991                           struct btrfs_root *root)
4992 {
4993         struct btrfs_root *reloc_root;
4994         int ret;
4995
4996         if (root->reloc_root) {
4997                 reloc_root = root->reloc_root;
4998                 root->reloc_root = NULL;
4999                 list_add(&reloc_root->dead_list,
5000                          &root->fs_info->dead_reloc_roots);
5001
5002                 btrfs_set_root_bytenr(&reloc_root->root_item,
5003                                       reloc_root->node->start);
5004                 btrfs_set_root_level(&root->root_item,
5005                                      btrfs_header_level(reloc_root->node));
5006                 memset(&reloc_root->root_item.drop_progress, 0,
5007                         sizeof(struct btrfs_disk_key));
5008                 reloc_root->root_item.drop_level = 0;
5009
5010                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
5011                                         &reloc_root->root_key,
5012                                         &reloc_root->root_item);
5013                 BUG_ON(ret);
5014         }
5015         return 0;
5016 }
5017
5018 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
5019 {
5020         struct btrfs_trans_handle *trans;
5021         struct btrfs_root *reloc_root;
5022         struct btrfs_root *prev_root = NULL;
5023         struct list_head dead_roots;
5024         int ret;
5025         unsigned long nr;
5026
5027         INIT_LIST_HEAD(&dead_roots);
5028         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
5029
5030         while (!list_empty(&dead_roots)) {
5031                 reloc_root = list_entry(dead_roots.prev,
5032                                         struct btrfs_root, dead_list);
5033                 list_del_init(&reloc_root->dead_list);
5034
5035                 BUG_ON(reloc_root->commit_root != NULL);
5036                 while (1) {
5037                         trans = btrfs_join_transaction(root, 1);
5038                         BUG_ON(!trans);
5039
5040                         mutex_lock(&root->fs_info->drop_mutex);
5041                         ret = btrfs_drop_snapshot(trans, reloc_root);
5042                         if (ret != -EAGAIN)
5043                                 break;
5044                         mutex_unlock(&root->fs_info->drop_mutex);
5045
5046                         nr = trans->blocks_used;
5047                         ret = btrfs_end_transaction(trans, root);
5048                         BUG_ON(ret);
5049                         btrfs_btree_balance_dirty(root, nr);
5050                 }
5051
5052                 free_extent_buffer(reloc_root->node);
5053
5054                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
5055                                      &reloc_root->root_key);
5056                 BUG_ON(ret);
5057                 mutex_unlock(&root->fs_info->drop_mutex);
5058
5059                 nr = trans->blocks_used;
5060                 ret = btrfs_end_transaction(trans, root);
5061                 BUG_ON(ret);
5062                 btrfs_btree_balance_dirty(root, nr);
5063
5064                 kfree(prev_root);
5065                 prev_root = reloc_root;
5066         }
5067         if (prev_root) {
5068                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5069                 kfree(prev_root);
5070         }
5071         return 0;
5072 }
5073
5074 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5075 {
5076         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5077         return 0;
5078 }
5079
5080 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5081 {
5082         struct btrfs_root *reloc_root;
5083         struct btrfs_trans_handle *trans;
5084         struct btrfs_key location;
5085         int found;
5086         int ret;
5087
5088         mutex_lock(&root->fs_info->tree_reloc_mutex);
5089         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5090         BUG_ON(ret);
5091         found = !list_empty(&root->fs_info->dead_reloc_roots);
5092         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5093
5094         if (found) {
5095                 trans = btrfs_start_transaction(root, 1);
5096                 BUG_ON(!trans);
5097                 ret = btrfs_commit_transaction(trans, root);
5098                 BUG_ON(ret);
5099         }
5100
5101         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5102         location.offset = (u64)-1;
5103         location.type = BTRFS_ROOT_ITEM_KEY;
5104
5105         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5106         BUG_ON(!reloc_root);
5107         btrfs_orphan_cleanup(reloc_root);
5108         return 0;
5109 }
5110
5111 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5112                                     struct btrfs_root *root)
5113 {
5114         struct btrfs_root *reloc_root;
5115         struct extent_buffer *eb;
5116         struct btrfs_root_item *root_item;
5117         struct btrfs_key root_key;
5118         int ret;
5119
5120         BUG_ON(!root->ref_cows);
5121         if (root->reloc_root)
5122                 return 0;
5123
5124         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5125         BUG_ON(!root_item);
5126
5127         ret = btrfs_copy_root(trans, root, root->commit_root,
5128                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5129         BUG_ON(ret);
5130
5131         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5132         root_key.offset = root->root_key.objectid;
5133         root_key.type = BTRFS_ROOT_ITEM_KEY;
5134
5135         memcpy(root_item, &root->root_item, sizeof(root_item));
5136         btrfs_set_root_refs(root_item, 0);
5137         btrfs_set_root_bytenr(root_item, eb->start);
5138         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5139         btrfs_set_root_generation(root_item, trans->transid);
5140
5141         btrfs_tree_unlock(eb);
5142         free_extent_buffer(eb);
5143
5144         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5145                                 &root_key, root_item);
5146         BUG_ON(ret);
5147         kfree(root_item);
5148
5149         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5150                                                  &root_key);
5151         BUG_ON(!reloc_root);
5152         reloc_root->last_trans = trans->transid;
5153         reloc_root->commit_root = NULL;
5154         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5155
5156         root->reloc_root = reloc_root;
5157         return 0;
5158 }
5159
5160 /*
5161  * Core function of space balance.
5162  *
5163  * The idea is using reloc trees to relocate tree blocks in reference
5164  * counted roots. There is one reloc tree for each subvol, and all
5165  * reloc trees share same root key objectid. Reloc trees are snapshots
5166  * of the latest committed roots of subvols (root->commit_root).
5167  *
5168  * To relocate a tree block referenced by a subvol, there are two steps.
5169  * COW the block through subvol's reloc tree, then update block pointer
5170  * in the subvol to point to the new block. Since all reloc trees share
5171  * same root key objectid, doing special handing for tree blocks owned
5172  * by them is easy. Once a tree block has been COWed in one reloc tree,
5173  * we can use the resulting new block directly when the same block is
5174  * required to COW again through other reloc trees. By this way, relocated
5175  * tree blocks are shared between reloc trees, so they are also shared
5176  * between subvols.
5177  */
5178 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5179                                       struct btrfs_root *root,
5180                                       struct btrfs_path *path,
5181                                       struct btrfs_key *first_key,
5182                                       struct btrfs_ref_path *ref_path,
5183                                       struct btrfs_block_group_cache *group,
5184                                       struct inode *reloc_inode)
5185 {
5186         struct btrfs_root *reloc_root;
5187         struct extent_buffer *eb = NULL;
5188         struct btrfs_key *keys;
5189         u64 *nodes;
5190         int level;
5191         int shared_level;
5192         int lowest_level = 0;
5193         int ret;
5194
5195         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5196                 lowest_level = ref_path->owner_objectid;
5197
5198         if (!root->ref_cows) {
5199                 path->lowest_level = lowest_level;
5200                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5201                 BUG_ON(ret < 0);
5202                 path->lowest_level = 0;
5203                 btrfs_release_path(root, path);
5204                 return 0;
5205         }
5206
5207         mutex_lock(&root->fs_info->tree_reloc_mutex);
5208         ret = init_reloc_tree(trans, root);
5209         BUG_ON(ret);
5210         reloc_root = root->reloc_root;
5211
5212         shared_level = ref_path->shared_level;
5213         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5214
5215         keys = ref_path->node_keys;
5216         nodes = ref_path->new_nodes;
5217         memset(&keys[shared_level + 1], 0,
5218                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5219         memset(&nodes[shared_level + 1], 0,
5220                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5221
5222         if (nodes[lowest_level] == 0) {
5223                 path->lowest_level = lowest_level;
5224                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5225                                         0, 1);
5226                 BUG_ON(ret);
5227                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5228                         eb = path->nodes[level];
5229                         if (!eb || eb == reloc_root->node)
5230                                 break;
5231                         nodes[level] = eb->start;
5232                         if (level == 0)
5233                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5234                         else
5235                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5236                 }
5237                 if (nodes[0] &&
5238                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5239                         eb = path->nodes[0];
5240                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5241                                                       group, reloc_inode);
5242                         BUG_ON(ret);
5243                 }
5244                 btrfs_release_path(reloc_root, path);
5245         } else {
5246                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5247                                        lowest_level);
5248                 BUG_ON(ret);
5249         }
5250
5251         /*
5252          * replace tree blocks in the fs tree with tree blocks in
5253          * the reloc tree.
5254          */
5255         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5256         BUG_ON(ret < 0);
5257
5258         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5259                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5260                                         0, 0);
5261                 BUG_ON(ret);
5262                 extent_buffer_get(path->nodes[0]);
5263                 eb = path->nodes[0];
5264                 btrfs_release_path(reloc_root, path);
5265                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5266                 BUG_ON(ret);
5267                 free_extent_buffer(eb);
5268         }
5269
5270         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5271         path->lowest_level = 0;
5272         return 0;
5273 }
5274
5275 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5276                                         struct btrfs_root *root,
5277                                         struct btrfs_path *path,
5278                                         struct btrfs_key *first_key,
5279                                         struct btrfs_ref_path *ref_path)
5280 {
5281         int ret;
5282
5283         ret = relocate_one_path(trans, root, path, first_key,
5284                                 ref_path, NULL, NULL);
5285         BUG_ON(ret);
5286
5287         if (root == root->fs_info->extent_root)
5288                 btrfs_extent_post_op(trans, root);
5289
5290         return 0;
5291 }
5292
5293 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5294                                     struct btrfs_root *extent_root,
5295                                     struct btrfs_path *path,
5296                                     struct btrfs_key *extent_key)
5297 {
5298         int ret;
5299
5300         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5301         if (ret)
5302                 goto out;
5303         ret = btrfs_del_item(trans, extent_root, path);
5304 out:
5305         btrfs_release_path(extent_root, path);
5306         return ret;
5307 }
5308
5309 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5310                                                 struct btrfs_ref_path *ref_path)
5311 {
5312         struct btrfs_key root_key;
5313
5314         root_key.objectid = ref_path->root_objectid;
5315         root_key.type = BTRFS_ROOT_ITEM_KEY;
5316         if (is_cowonly_root(ref_path->root_objectid))
5317                 root_key.offset = 0;
5318         else
5319                 root_key.offset = (u64)-1;
5320
5321         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5322 }
5323
5324 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5325                                         struct btrfs_path *path,
5326                                         struct btrfs_key *extent_key,
5327                                         struct btrfs_block_group_cache *group,
5328                                         struct inode *reloc_inode, int pass)
5329 {
5330         struct btrfs_trans_handle *trans;
5331         struct btrfs_root *found_root;
5332         struct btrfs_ref_path *ref_path = NULL;
5333         struct disk_extent *new_extents = NULL;
5334         int nr_extents = 0;
5335         int loops;
5336         int ret;
5337         int level;
5338         struct btrfs_key first_key;
5339         u64 prev_block = 0;
5340
5341
5342         trans = btrfs_start_transaction(extent_root, 1);
5343         BUG_ON(!trans);
5344
5345         if (extent_key->objectid == 0) {
5346                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5347                 goto out;
5348         }
5349
5350         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5351         if (!ref_path) {
5352                 ret = -ENOMEM;
5353                 goto out;
5354         }
5355
5356         for (loops = 0; ; loops++) {
5357                 if (loops == 0) {
5358                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5359                                                    extent_key->objectid);
5360                 } else {
5361                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5362                 }
5363                 if (ret < 0)
5364                         goto out;
5365                 if (ret > 0)
5366                         break;
5367
5368                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5369                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5370                         continue;
5371
5372                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5373                 BUG_ON(!found_root);
5374                 /*
5375                  * for reference counted tree, only process reference paths
5376                  * rooted at the latest committed root.
5377                  */
5378                 if (found_root->ref_cows &&
5379                     ref_path->root_generation != found_root->root_key.offset)
5380                         continue;
5381
5382                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5383                         if (pass == 0) {
5384                                 /*
5385                                  * copy data extents to new locations
5386                                  */
5387                                 u64 group_start = group->key.objectid;
5388                                 ret = relocate_data_extent(reloc_inode,
5389                                                            extent_key,
5390                                                            group_start);
5391                                 if (ret < 0)
5392                                         goto out;
5393                                 break;
5394                         }
5395                         level = 0;
5396                 } else {
5397                         level = ref_path->owner_objectid;
5398                 }
5399
5400                 if (prev_block != ref_path->nodes[level]) {
5401                         struct extent_buffer *eb;
5402                         u64 block_start = ref_path->nodes[level];
5403                         u64 block_size = btrfs_level_size(found_root, level);
5404
5405                         eb = read_tree_block(found_root, block_start,
5406                                              block_size, 0);
5407                         btrfs_tree_lock(eb);
5408                         BUG_ON(level != btrfs_header_level(eb));
5409
5410                         if (level == 0)
5411                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5412                         else
5413                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5414
5415                         btrfs_tree_unlock(eb);
5416                         free_extent_buffer(eb);
5417                         prev_block = block_start;
5418                 }
5419
5420                 btrfs_record_root_in_trans(found_root);
5421                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5422                         /*
5423                          * try to update data extent references while
5424                          * keeping metadata shared between snapshots.
5425                          */
5426                         if (pass == 1) {
5427                                 ret = relocate_one_path(trans, found_root,
5428                                                 path, &first_key, ref_path,
5429                                                 group, reloc_inode);
5430                                 if (ret < 0)
5431                                         goto out;
5432                                 continue;
5433                         }
5434                         /*
5435                          * use fallback method to process the remaining
5436                          * references.
5437                          */
5438                         if (!new_extents) {
5439                                 u64 group_start = group->key.objectid;
5440                                 new_extents = kmalloc(sizeof(*new_extents),
5441                                                       GFP_NOFS);
5442                                 nr_extents = 1;
5443                                 ret = get_new_locations(reloc_inode,
5444                                                         extent_key,
5445                                                         group_start, 1,
5446                                                         &new_extents,
5447                                                         &nr_extents);
5448                                 if (ret)
5449                                         goto out;
5450                         }
5451                         ret = replace_one_extent(trans, found_root,
5452                                                 path, extent_key,
5453                                                 &first_key, ref_path,
5454                                                 new_extents, nr_extents);
5455                 } else {
5456                         ret = relocate_tree_block(trans, found_root, path,
5457                                                   &first_key, ref_path);
5458                 }
5459                 if (ret < 0)
5460                         goto out;
5461         }
5462         ret = 0;
5463 out:
5464         btrfs_end_transaction(trans, extent_root);
5465         kfree(new_extents);
5466         kfree(ref_path);
5467         return ret;
5468 }
5469
5470 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5471 {
5472         u64 num_devices;
5473         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5474                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5475
5476         num_devices = root->fs_info->fs_devices->rw_devices;
5477         if (num_devices == 1) {
5478                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5479                 stripped = flags & ~stripped;
5480
5481                 /* turn raid0 into single device chunks */
5482                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5483                         return stripped;
5484
5485                 /* turn mirroring into duplication */
5486                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5487                              BTRFS_BLOCK_GROUP_RAID10))
5488                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5489                 return flags;
5490         } else {
5491                 /* they already had raid on here, just return */
5492                 if (flags & stripped)
5493                         return flags;
5494
5495                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5496                 stripped = flags & ~stripped;
5497
5498                 /* switch duplicated blocks with raid1 */
5499                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5500                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5501
5502                 /* turn single device chunks into raid0 */
5503                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5504         }
5505         return flags;
5506 }
5507
5508 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5509                      struct btrfs_block_group_cache *shrink_block_group,
5510                      int force)
5511 {
5512         struct btrfs_trans_handle *trans;
5513         u64 new_alloc_flags;
5514         u64 calc;
5515
5516         spin_lock(&shrink_block_group->lock);
5517         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5518                 spin_unlock(&shrink_block_group->lock);
5519
5520                 trans = btrfs_start_transaction(root, 1);
5521                 spin_lock(&shrink_block_group->lock);
5522
5523                 new_alloc_flags = update_block_group_flags(root,
5524                                                    shrink_block_group->flags);
5525                 if (new_alloc_flags != shrink_block_group->flags) {
5526                         calc =
5527                              btrfs_block_group_used(&shrink_block_group->item);
5528                 } else {
5529                         calc = shrink_block_group->key.offset;
5530                 }
5531                 spin_unlock(&shrink_block_group->lock);
5532
5533                 do_chunk_alloc(trans, root->fs_info->extent_root,
5534                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5535
5536                 btrfs_end_transaction(trans, root);
5537         } else
5538                 spin_unlock(&shrink_block_group->lock);
5539         return 0;
5540 }
5541
5542 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5543                                  struct btrfs_root *root,
5544                                  u64 objectid, u64 size)
5545 {
5546         struct btrfs_path *path;
5547         struct btrfs_inode_item *item;
5548         struct extent_buffer *leaf;
5549         int ret;
5550
5551         path = btrfs_alloc_path();
5552         if (!path)
5553                 return -ENOMEM;
5554
5555         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5556         if (ret)
5557                 goto out;
5558
5559         leaf = path->nodes[0];
5560         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5561         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5562         btrfs_set_inode_generation(leaf, item, 1);
5563         btrfs_set_inode_size(leaf, item, size);
5564         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5565         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5566         btrfs_mark_buffer_dirty(leaf);
5567         btrfs_release_path(root, path);
5568 out:
5569         btrfs_free_path(path);
5570         return ret;
5571 }
5572
5573 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5574                                         struct btrfs_block_group_cache *group)
5575 {
5576         struct inode *inode = NULL;
5577         struct btrfs_trans_handle *trans;
5578         struct btrfs_root *root;
5579         struct btrfs_key root_key;
5580         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5581         int err = 0;
5582
5583         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5584         root_key.type = BTRFS_ROOT_ITEM_KEY;
5585         root_key.offset = (u64)-1;
5586         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5587         if (IS_ERR(root))
5588                 return ERR_CAST(root);
5589
5590         trans = btrfs_start_transaction(root, 1);
5591         BUG_ON(!trans);
5592
5593         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5594         if (err)
5595                 goto out;
5596
5597         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5598         BUG_ON(err);
5599
5600         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5601                                        group->key.offset, 0, group->key.offset,
5602                                        0, 0, 0);
5603         BUG_ON(err);
5604
5605         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5606         if (inode->i_state & I_NEW) {
5607                 BTRFS_I(inode)->root = root;
5608                 BTRFS_I(inode)->location.objectid = objectid;
5609                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5610                 BTRFS_I(inode)->location.offset = 0;
5611                 btrfs_read_locked_inode(inode);
5612                 unlock_new_inode(inode);
5613                 BUG_ON(is_bad_inode(inode));
5614         } else {
5615                 BUG_ON(1);
5616         }
5617         BTRFS_I(inode)->index_cnt = group->key.objectid;
5618
5619         err = btrfs_orphan_add(trans, inode);
5620 out:
5621         btrfs_end_transaction(trans, root);
5622         if (err) {
5623                 if (inode)
5624                         iput(inode);
5625                 inode = ERR_PTR(err);
5626         }
5627         return inode;
5628 }
5629
5630 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5631 {
5632
5633         struct btrfs_ordered_sum *sums;
5634         struct btrfs_sector_sum *sector_sum;
5635         struct btrfs_ordered_extent *ordered;
5636         struct btrfs_root *root = BTRFS_I(inode)->root;
5637         struct list_head list;
5638         size_t offset;
5639         int ret;
5640         u64 disk_bytenr;
5641
5642         INIT_LIST_HEAD(&list);
5643
5644         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5645         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5646
5647         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5648         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5649                                        disk_bytenr + len - 1, &list);
5650
5651         while (!list_empty(&list)) {
5652                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5653                 list_del_init(&sums->list);
5654
5655                 sector_sum = sums->sums;
5656                 sums->bytenr = ordered->start;
5657
5658                 offset = 0;
5659                 while (offset < sums->len) {
5660                         sector_sum->bytenr += ordered->start - disk_bytenr;
5661                         sector_sum++;
5662                         offset += root->sectorsize;
5663                 }
5664
5665                 btrfs_add_ordered_sum(inode, ordered, sums);
5666         }
5667         btrfs_put_ordered_extent(ordered);
5668         return 0;
5669 }
5670
5671 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5672 {
5673         struct btrfs_trans_handle *trans;
5674         struct btrfs_path *path;
5675         struct btrfs_fs_info *info = root->fs_info;
5676         struct extent_buffer *leaf;
5677         struct inode *reloc_inode;
5678         struct btrfs_block_group_cache *block_group;
5679         struct btrfs_key key;
5680         u64 skipped;
5681         u64 cur_byte;
5682         u64 total_found;
5683         u32 nritems;
5684         int ret;
5685         int progress;
5686         int pass = 0;
5687
5688         root = root->fs_info->extent_root;
5689
5690         block_group = btrfs_lookup_block_group(info, group_start);
5691         BUG_ON(!block_group);
5692
5693         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5694                (unsigned long long)block_group->key.objectid,
5695                (unsigned long long)block_group->flags);
5696
5697         path = btrfs_alloc_path();
5698         BUG_ON(!path);
5699
5700         reloc_inode = create_reloc_inode(info, block_group);
5701         BUG_ON(IS_ERR(reloc_inode));
5702
5703         __alloc_chunk_for_shrink(root, block_group, 1);
5704         set_block_group_readonly(block_group);
5705
5706         btrfs_start_delalloc_inodes(info->tree_root);
5707         btrfs_wait_ordered_extents(info->tree_root, 0);
5708 again:
5709         skipped = 0;
5710         total_found = 0;
5711         progress = 0;
5712         key.objectid = block_group->key.objectid;
5713         key.offset = 0;
5714         key.type = 0;
5715         cur_byte = key.objectid;
5716
5717         trans = btrfs_start_transaction(info->tree_root, 1);
5718         btrfs_commit_transaction(trans, info->tree_root);
5719
5720         mutex_lock(&root->fs_info->cleaner_mutex);
5721         btrfs_clean_old_snapshots(info->tree_root);
5722         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5723         mutex_unlock(&root->fs_info->cleaner_mutex);
5724
5725         while (1) {
5726                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5727                 if (ret < 0)
5728                         goto out;
5729 next:
5730                 leaf = path->nodes[0];
5731                 nritems = btrfs_header_nritems(leaf);
5732                 if (path->slots[0] >= nritems) {
5733                         ret = btrfs_next_leaf(root, path);
5734                         if (ret < 0)
5735                                 goto out;
5736                         if (ret == 1) {
5737                                 ret = 0;
5738                                 break;
5739                         }
5740                         leaf = path->nodes[0];
5741                         nritems = btrfs_header_nritems(leaf);
5742                 }
5743
5744                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5745
5746                 if (key.objectid >= block_group->key.objectid +
5747                     block_group->key.offset)
5748                         break;
5749
5750                 if (progress && need_resched()) {
5751                         btrfs_release_path(root, path);
5752                         cond_resched();
5753                         progress = 0;
5754                         continue;
5755                 }
5756                 progress = 1;
5757
5758                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5759                     key.objectid + key.offset <= cur_byte) {
5760                         path->slots[0]++;
5761                         goto next;
5762                 }
5763
5764                 total_found++;
5765                 cur_byte = key.objectid + key.offset;
5766                 btrfs_release_path(root, path);
5767
5768                 __alloc_chunk_for_shrink(root, block_group, 0);
5769                 ret = relocate_one_extent(root, path, &key, block_group,
5770                                           reloc_inode, pass);
5771                 BUG_ON(ret < 0);
5772                 if (ret > 0)
5773                         skipped++;
5774
5775                 key.objectid = cur_byte;
5776                 key.type = 0;
5777                 key.offset = 0;
5778         }
5779
5780         btrfs_release_path(root, path);
5781
5782         if (pass == 0) {
5783                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5784                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5785         }
5786
5787         if (total_found > 0) {
5788                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5789                        (unsigned long long)total_found, pass);
5790                 pass++;
5791                 if (total_found == skipped && pass > 2) {
5792                         iput(reloc_inode);
5793                         reloc_inode = create_reloc_inode(info, block_group);
5794                         pass = 0;
5795                 }
5796                 goto again;
5797         }
5798
5799         /* delete reloc_inode */
5800         iput(reloc_inode);
5801
5802         /* unpin extents in this range */
5803         trans = btrfs_start_transaction(info->tree_root, 1);
5804         btrfs_commit_transaction(trans, info->tree_root);
5805
5806         spin_lock(&block_group->lock);
5807         WARN_ON(block_group->pinned > 0);
5808         WARN_ON(block_group->reserved > 0);
5809         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5810         spin_unlock(&block_group->lock);
5811         put_block_group(block_group);
5812         ret = 0;
5813 out:
5814         btrfs_free_path(path);
5815         return ret;
5816 }
5817
5818 static int find_first_block_group(struct btrfs_root *root,
5819                 struct btrfs_path *path, struct btrfs_key *key)
5820 {
5821         int ret = 0;
5822         struct btrfs_key found_key;
5823         struct extent_buffer *leaf;
5824         int slot;
5825
5826         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5827         if (ret < 0)
5828                 goto out;
5829
5830         while (1) {
5831                 slot = path->slots[0];
5832                 leaf = path->nodes[0];
5833                 if (slot >= btrfs_header_nritems(leaf)) {
5834                         ret = btrfs_next_leaf(root, path);
5835                         if (ret == 0)
5836                                 continue;
5837                         if (ret < 0)
5838                                 goto out;
5839                         break;
5840                 }
5841                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5842
5843                 if (found_key.objectid >= key->objectid &&
5844                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5845                         ret = 0;
5846                         goto out;
5847                 }
5848                 path->slots[0]++;
5849         }
5850         ret = -ENOENT;
5851 out:
5852         return ret;
5853 }
5854
5855 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5856 {
5857         struct btrfs_block_group_cache *block_group;
5858         struct rb_node *n;
5859
5860         spin_lock(&info->block_group_cache_lock);
5861         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5862                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5863                                        cache_node);
5864                 rb_erase(&block_group->cache_node,
5865                          &info->block_group_cache_tree);
5866                 spin_unlock(&info->block_group_cache_lock);
5867
5868                 btrfs_remove_free_space_cache(block_group);
5869                 down_write(&block_group->space_info->groups_sem);
5870                 list_del(&block_group->list);
5871                 up_write(&block_group->space_info->groups_sem);
5872
5873                 WARN_ON(atomic_read(&block_group->count) != 1);
5874                 kfree(block_group);
5875
5876                 spin_lock(&info->block_group_cache_lock);
5877         }
5878         spin_unlock(&info->block_group_cache_lock);
5879         return 0;
5880 }
5881
5882 int btrfs_read_block_groups(struct btrfs_root *root)
5883 {
5884         struct btrfs_path *path;
5885         int ret;
5886         struct btrfs_block_group_cache *cache;
5887         struct btrfs_fs_info *info = root->fs_info;
5888         struct btrfs_space_info *space_info;
5889         struct btrfs_key key;
5890         struct btrfs_key found_key;
5891         struct extent_buffer *leaf;
5892
5893         root = info->extent_root;
5894         key.objectid = 0;
5895         key.offset = 0;
5896         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5897         path = btrfs_alloc_path();
5898         if (!path)
5899                 return -ENOMEM;
5900
5901         while (1) {
5902                 ret = find_first_block_group(root, path, &key);
5903                 if (ret > 0) {
5904                         ret = 0;
5905                         goto error;
5906                 }
5907                 if (ret != 0)
5908                         goto error;
5909
5910                 leaf = path->nodes[0];
5911                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5912                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5913                 if (!cache) {
5914                         ret = -ENOMEM;
5915                         break;
5916                 }
5917
5918                 atomic_set(&cache->count, 1);
5919                 spin_lock_init(&cache->lock);
5920                 mutex_init(&cache->alloc_mutex);
5921                 mutex_init(&cache->cache_mutex);
5922                 INIT_LIST_HEAD(&cache->list);
5923                 read_extent_buffer(leaf, &cache->item,
5924                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5925                                    sizeof(cache->item));
5926                 memcpy(&cache->key, &found_key, sizeof(found_key));
5927
5928                 key.objectid = found_key.objectid + found_key.offset;
5929                 btrfs_release_path(root, path);
5930                 cache->flags = btrfs_block_group_flags(&cache->item);
5931
5932                 ret = update_space_info(info, cache->flags, found_key.offset,
5933                                         btrfs_block_group_used(&cache->item),
5934                                         &space_info);
5935                 BUG_ON(ret);
5936                 cache->space_info = space_info;
5937                 down_write(&space_info->groups_sem);
5938                 list_add_tail(&cache->list, &space_info->block_groups);
5939                 up_write(&space_info->groups_sem);
5940
5941                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5942                 BUG_ON(ret);
5943
5944                 set_avail_alloc_bits(root->fs_info, cache->flags);
5945                 if (btrfs_chunk_readonly(root, cache->key.objectid))
5946                         set_block_group_readonly(cache);
5947         }
5948         ret = 0;
5949 error:
5950         btrfs_free_path(path);
5951         return ret;
5952 }
5953
5954 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5955                            struct btrfs_root *root, u64 bytes_used,
5956                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5957                            u64 size)
5958 {
5959         int ret;
5960         struct btrfs_root *extent_root;
5961         struct btrfs_block_group_cache *cache;
5962
5963         extent_root = root->fs_info->extent_root;
5964
5965         root->fs_info->last_trans_new_blockgroup = trans->transid;
5966
5967         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5968         if (!cache)
5969                 return -ENOMEM;
5970
5971         cache->key.objectid = chunk_offset;
5972         cache->key.offset = size;
5973         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5974         atomic_set(&cache->count, 1);
5975         spin_lock_init(&cache->lock);
5976         mutex_init(&cache->alloc_mutex);
5977         mutex_init(&cache->cache_mutex);
5978         INIT_LIST_HEAD(&cache->list);
5979
5980         btrfs_set_block_group_used(&cache->item, bytes_used);
5981         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5982         cache->flags = type;
5983         btrfs_set_block_group_flags(&cache->item, type);
5984
5985         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5986                                 &cache->space_info);
5987         BUG_ON(ret);
5988         down_write(&cache->space_info->groups_sem);
5989         list_add_tail(&cache->list, &cache->space_info->block_groups);
5990         up_write(&cache->space_info->groups_sem);
5991
5992         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5993         BUG_ON(ret);
5994
5995         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5996                                 sizeof(cache->item));
5997         BUG_ON(ret);
5998
5999         finish_current_insert(trans, extent_root, 0);
6000         ret = del_pending_extents(trans, extent_root, 0);
6001         BUG_ON(ret);
6002         set_avail_alloc_bits(extent_root->fs_info, type);
6003
6004         return 0;
6005 }
6006
6007 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
6008                              struct btrfs_root *root, u64 group_start)
6009 {
6010         struct btrfs_path *path;
6011         struct btrfs_block_group_cache *block_group;
6012         struct btrfs_key key;
6013         int ret;
6014
6015         root = root->fs_info->extent_root;
6016
6017         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
6018         BUG_ON(!block_group);
6019         BUG_ON(!block_group->ro);
6020
6021         memcpy(&key, &block_group->key, sizeof(key));
6022
6023         path = btrfs_alloc_path();
6024         BUG_ON(!path);
6025
6026         spin_lock(&root->fs_info->block_group_cache_lock);
6027         rb_erase(&block_group->cache_node,
6028                  &root->fs_info->block_group_cache_tree);
6029         spin_unlock(&root->fs_info->block_group_cache_lock);
6030         btrfs_remove_free_space_cache(block_group);
6031         down_write(&block_group->space_info->groups_sem);
6032         list_del(&block_group->list);
6033         up_write(&block_group->space_info->groups_sem);
6034
6035         spin_lock(&block_group->space_info->lock);
6036         block_group->space_info->total_bytes -= block_group->key.offset;
6037         block_group->space_info->bytes_readonly -= block_group->key.offset;
6038         spin_unlock(&block_group->space_info->lock);
6039         block_group->space_info->full = 0;
6040
6041         put_block_group(block_group);
6042         put_block_group(block_group);
6043
6044         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6045         if (ret > 0)
6046                 ret = -EIO;
6047         if (ret < 0)
6048                 goto out;
6049
6050         ret = btrfs_del_item(trans, root, path);
6051 out:
6052         btrfs_free_path(path);
6053         return ret;
6054 }