fs/btrfs/extent-tree.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18 #include <linux/sched.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/writeback.h>
  21 #include <linux/blkdev.h>
  22 #include "compat.h"
  23 #include "hash.h"
  24 #include "crc32c.h"
  25 #include "ctree.h"
  26 #include "disk-io.h"
  27 #include "print-tree.h"
  28 #include "transaction.h"
  29 #include "volumes.h"
  30 #include "locking.h"
  31 #include "ref-cache.h"
  32 #include "compat.h"
  33
  34 #define PENDING_EXTENT_INSERT 0
  35 #define PENDING_EXTENT_DELETE 1
  36 #define PENDING_BACKREF_UPDATE 2
  37
  38 struct pending_extent_op {
  39         int type;
  40         u64 bytenr;
  41         u64 num_bytes;
  42         u64 parent;
  43         u64 orig_parent;
  44         u64 generation;
  45         u64 orig_generation;
  46         int level;
  47         struct list_head list;
  48         int del;
  49 };
  50
  51 static int finish_current_insert(struct btrfs_trans_handle *trans,
  52                                  struct btrfs_root *extent_root, int all);
  53 static int del_pending_extents(struct btrfs_trans_handle *trans,
  54                                struct btrfs_root *extent_root, int all);
  55 static int pin_down_bytes(struct btrfs_trans_handle *trans,
  56                           struct btrfs_root *root,
  57                           u64 bytenr, u64 num_bytes, int is_data);
  58 static int update_block_group(struct btrfs_trans_handle *trans,
  59                               struct btrfs_root *root,
  60                               u64 bytenr, u64 num_bytes, int alloc,
  61                               int mark_free);
  62
  63 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
  64 {
  65         return (cache->flags & bits) == bits;
  66 }
  67
  68 /*
  69  * this adds the block group to the fs_info rb tree for the block group
  70  * cache
  71  */
  72 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
  73                                 struct btrfs_block_group_cache *block_group)
  74 {
  75         struct rb_node **p;
  76         struct rb_node *parent = NULL;
  77         struct btrfs_block_group_cache *cache;
  78
  79         spin_lock(&info->block_group_cache_lock);
  80         p = &info->block_group_cache_tree.rb_node;
  81
  82         while (*p) {
  83                 parent = *p;
  84                 cache = rb_entry(parent, struct btrfs_block_group_cache,
  85                                  cache_node);
  86                 if (block_group->key.objectid < cache->key.objectid) {
  87                         p = &(*p)->rb_left;
  88                 } else if (block_group->key.objectid > cache->key.objectid) {
  89                         p = &(*p)->rb_right;
  90                 } else {
  91                         spin_unlock(&info->block_group_cache_lock);
  92                         return -EEXIST;
  93                 }
  94         }
  95
  96         rb_link_node(&block_group->cache_node, parent, p);
  97         rb_insert_color(&block_group->cache_node,
  98                         &info->block_group_cache_tree);
  99         spin_unlock(&info->block_group_cache_lock);
 100
 101         return 0;
 102 }
 103
 104 /*
 105  * This will return the block group at or after bytenr if contains is 0, else
 106  * it will return the block group that contains the bytenr
 107  */
 108 static struct btrfs_block_group_cache *
 109 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
 110                               int contains)
 111 {
 112         struct btrfs_block_group_cache *cache, *ret = NULL;
 113         struct rb_node *n;
 114         u64 end, start;
 115
 116         spin_lock(&info->block_group_cache_lock);
 117         n = info->block_group_cache_tree.rb_node;
 118
 119         while (n) {
 120                 cache = rb_entry(n, struct btrfs_block_group_cache,
 121                                  cache_node);
 122                 end = cache->key.objectid + cache->key.offset - 1;
 123                 start = cache->key.objectid;
 124
 125                 if (bytenr < start) {
 126                         if (!contains && (!ret || start < ret->key.objectid))
 127                                 ret = cache;
 128                         n = n->rb_left;
 129                 } else if (bytenr > start) {
 130                         if (contains && bytenr <= end) {
 131                                 ret = cache;
 132                                 break;
 133                         }
 134                         n = n->rb_right;
 135                 } else {
 136                         ret = cache;
 137                         break;
 138                 }
 139         }
 140         if (ret)
 141                 atomic_inc(&ret->count);
 142         spin_unlock(&info->block_group_cache_lock);
 143
 144         return ret;
 145 }
 146
 147 /*
 148  * this is only called by cache_block_group, since we could have freed extents
 149  * we need to check the pinned_extents for any extents that can't be used yet
 150  * since their free space will be released as soon as the transaction commits.
 151  */
 152 static int add_new_free_space(struct btrfs_block_group_cache *block_group,
 153                               struct btrfs_fs_info *info, u64 start, u64 end)
 154 {
 155         u64 extent_start, extent_end, size;
 156         int ret;
 157
 158         mutex_lock(&info->pinned_mutex);
 159         while (start < end) {
 160                 ret = find_first_extent_bit(&info->pinned_extents, start,
 161                                             &extent_start, &extent_end,
 162                                             EXTENT_DIRTY);
 163                 if (ret)
 164                         break;
 165
 166                 if (extent_start == start) {
 167                         start = extent_end + 1;
 168                 } else if (extent_start > start && extent_start < end) {
 169                         size = extent_start - start;
 170                         ret = btrfs_add_free_space(block_group, start,
 171                                                    size);
 172                         BUG_ON(ret);
 173                         start = extent_end + 1;
 174                 } else {
 175                         break;
 176                 }
 177         }
 178
 179         if (start < end) {
 180                 size = end - start;
 181                 ret = btrfs_add_free_space(block_group, start, size);
 182                 BUG_ON(ret);
 183         }
 184         mutex_unlock(&info->pinned_mutex);
 185
 186         return 0;
 187 }
 188
 189 static int remove_sb_from_cache(struct btrfs_root *root,
 190                                 struct btrfs_block_group_cache *cache)
 191 {
 192         u64 bytenr;
 193         u64 *logical;
 194         int stripe_len;
 195         int i, nr, ret;
 196
 197         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 198                 bytenr = btrfs_sb_offset(i);
 199                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
 200                                        cache->key.objectid, bytenr, 0,
 201                                        &logical, &nr, &stripe_len);
 202                 BUG_ON(ret);
 203                 while (nr--) {
 204                         btrfs_remove_free_space(cache, logical[nr],
 205                                                 stripe_len);
 206                 }
 207                 kfree(logical);
 208         }
 209         return 0;
 210 }
 211
 212 static int cache_block_group(struct btrfs_root *root,
 213                              struct btrfs_block_group_cache *block_group)
 214 {
 215         struct btrfs_path *path;
 216         int ret = 0;
 217         struct btrfs_key key;
 218         struct extent_buffer *leaf;
 219         int slot;
 220         u64 last;
 221
 222         if (!block_group)
 223                 return 0;
 224
 225         root = root->fs_info->extent_root;
 226
 227         if (block_group->cached)
 228                 return 0;
 229
 230         path = btrfs_alloc_path();
 231         if (!path)
 232                 return -ENOMEM;
 233
 234         path->reada = 2;
 235         /*
 236          * we get into deadlocks with paths held by callers of this function.
 237          * since the alloc_mutex is protecting things right now, just
 238          * skip the locking here
 239          */
 240         path->skip_locking = 1;
 241         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
 242         key.objectid = last;
 243         key.offset = 0;
 244         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 245         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 246         if (ret < 0)
 247                 goto err;
 248
 249         while (1) {
 250                 leaf = path->nodes[0];
 251                 slot = path->slots[0];
 252                 if (slot >= btrfs_header_nritems(leaf)) {
 253                         ret = btrfs_next_leaf(root, path);
 254                         if (ret < 0)
 255                                 goto err;
 256                         if (ret == 0)
 257                                 continue;
 258                         else
 259                                 break;
 260                 }
 261                 btrfs_item_key_to_cpu(leaf, &key, slot);
 262                 if (key.objectid < block_group->key.objectid)
 263                         goto next;
 264
 265                 if (key.objectid >= block_group->key.objectid +
 266                     block_group->key.offset)
 267                         break;
 268
 269                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
 270                         add_new_free_space(block_group, root->fs_info, last,
 271                                            key.objectid);
 272
 273                         last = key.objectid + key.offset;
 274                 }
 275 next:
 276                 path->slots[0]++;
 277         }
 278
 279         add_new_free_space(block_group, root->fs_info, last,
 280                            block_group->key.objectid +
 281                            block_group->key.offset);
 282
 283         remove_sb_from_cache(root, block_group);
 284         block_group->cached = 1;
 285         ret = 0;
 286 err:
 287         btrfs_free_path(path);
 288         return ret;
 289 }
 290
 291 /*
 292  * return the block group that starts at or after bytenr
 293  */
 294 static struct btrfs_block_group_cache *
 295 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
 296 {
 297         struct btrfs_block_group_cache *cache;
 298
 299         cache = block_group_cache_tree_search(info, bytenr, 0);
 300
 301         return cache;
 302 }
 303
 304 /*
 305  * return the block group that contains teh given bytenr
 306  */
 307 struct btrfs_block_group_cache *btrfs_lookup_block_group(
 308                                                  struct btrfs_fs_info *info,
 309                                                  u64 bytenr)
 310 {
 311         struct btrfs_block_group_cache *cache;
 312
 313         cache = block_group_cache_tree_search(info, bytenr, 1);
 314
 315         return cache;
 316 }
 317
 318 static inline void put_block_group(struct btrfs_block_group_cache *cache)
 319 {
 320         if (atomic_dec_and_test(&cache->count))
 321                 kfree(cache);
 322 }
 323
 324 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
 325                                                   u64 flags)
 326 {
 327         struct list_head *head = &info->space_info;
 328         struct btrfs_space_info *found;
 329         list_for_each_entry(found, head, list) {
 330                 if (found->flags == flags)
 331                         return found;
 332         }
 333         return NULL;
 334 }
 335
 336 static u64 div_factor(u64 num, int factor)
 337 {
 338         if (factor == 10)
 339                 return num;
 340         num *= factor;
 341         do_div(num, 10);
 342         return num;
 343 }
 344
 345 u64 btrfs_find_block_group(struct btrfs_root *root,
 346                            u64 search_start, u64 search_hint, int owner)
 347 {
 348         struct btrfs_block_group_cache *cache;
 349         u64 used;
 350         u64 last = max(search_hint, search_start);
 351         u64 group_start = 0;
 352         int full_search = 0;
 353         int factor = 9;
 354         int wrapped = 0;
 355 again:
 356         while (1) {
 357                 cache = btrfs_lookup_first_block_group(root->fs_info, last);
 358                 if (!cache)
 359                         break;
 360
 361                 spin_lock(&cache->lock);
 362                 last = cache->key.objectid + cache->key.offset;
 363                 used = btrfs_block_group_used(&cache->item);
 364
 365                 if ((full_search || !cache->ro) &&
 366                     block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) {
 367                         if (used + cache->pinned + cache->reserved <
 368                             div_factor(cache->key.offset, factor)) {
 369                                 group_start = cache->key.objectid;
 370                                 spin_unlock(&cache->lock);
 371                                 put_block_group(cache);
 372                                 goto found;
 373                         }
 374                 }
 375                 spin_unlock(&cache->lock);
 376                 put_block_group(cache);
 377                 cond_resched();
 378         }
 379         if (!wrapped) {
 380                 last = search_start;
 381                 wrapped = 1;
 382                 goto again;
 383         }
 384         if (!full_search && factor < 10) {
 385                 last = search_start;
 386                 full_search = 1;
 387                 factor = 10;
 388                 goto again;
 389         }
 390 found:
 391         return group_start;
 392 }
 393
 394 /* simple helper to search for an existing extent at a given offset */
 395 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
 396 {
 397         int ret;
 398         struct btrfs_key key;
 399         struct btrfs_path *path;
 400
 401         path = btrfs_alloc_path();
 402         BUG_ON(!path);
 403         key.objectid = start;
 404         key.offset = len;
 405         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 406         ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
 407                                 0, 0);
 408         btrfs_free_path(path);
 409         return ret;
 410 }
 411
 412 /*
 413  * Back reference rules.  Back refs have three main goals:
 414  *
 415  * 1) differentiate between all holders of references to an extent so that
 416  *    when a reference is dropped we can make sure it was a valid reference
 417  *    before freeing the extent.
 418  *
 419  * 2) Provide enough information to quickly find the holders of an extent
 420  *    if we notice a given block is corrupted or bad.
 421  *
 422  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
 423  *    maintenance.  This is actually the same as #2, but with a slightly
 424  *    different use case.
 425  *
 426  * File extents can be referenced by:
 427  *
 428  * - multiple snapshots, subvolumes, or different generations in one subvol
 429  * - different files inside a single subvolume
 430  * - different offsets inside a file (bookend extents in file.c)
 431  *
 432  * The extent ref structure has fields for:
 433  *
 434  * - Objectid of the subvolume root
 435  * - Generation number of the tree holding the reference
 436  * - objectid of the file holding the reference
 437  * - number of references holding by parent node (alway 1 for tree blocks)
 438  *
 439  * Btree leaf may hold multiple references to a file extent. In most cases,
 440  * these references are from same file and the corresponding offsets inside
 441  * the file are close together.
 442  *
 443  * When a file extent is allocated the fields are filled in:
 444  *     (root_key.objectid, trans->transid, inode objectid, 1)
 445  *
 446  * When a leaf is cow'd new references are added for every file extent found
 447  * in the leaf.  It looks similar to the create case, but trans->transid will
 448  * be different when the block is cow'd.
 449  *
 450  *     (root_key.objectid, trans->transid, inode objectid,
 451  *      number of references in the leaf)
 452  *
 453  * When a file extent is removed either during snapshot deletion or
 454  * file truncation, we find the corresponding back reference and check
 455  * the following fields:
 456  *
 457  *     (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
 458  *      inode objectid)
 459  *
 460  * Btree extents can be referenced by:
 461  *
 462  * - Different subvolumes
 463  * - Different generations of the same subvolume
 464  *
 465  * When a tree block is created, back references are inserted:
 466  *
 467  * (root->root_key.objectid, trans->transid, level, 1)
 468  *
 469  * When a tree block is cow'd, new back references are added for all the
 470  * blocks it points to. If the tree block isn't in reference counted root,
 471  * the old back references are removed. These new back references are of
 472  * the form (trans->transid will have increased since creation):
 473  *
 474  * (root->root_key.objectid, trans->transid, level, 1)
 475  *
 476  * When a backref is in deleting, the following fields are checked:
 477  *
 478  * if backref was for a tree root:
 479  *     (btrfs_header_owner(itself), btrfs_header_generation(itself), level)
 480  * else
 481  *     (btrfs_header_owner(parent), btrfs_header_generation(parent), level)
 482  *
 483  * Back Reference Key composing:
 484  *
 485  * The key objectid corresponds to the first byte in the extent, the key
 486  * type is set to BTRFS_EXTENT_REF_KEY, and the key offset is the first
 487  * byte of parent extent. If a extent is tree root, the key offset is set
 488  * to the key objectid.
 489  */
 490
 491 static noinline int lookup_extent_backref(struct btrfs_trans_handle *trans,
 492                                           struct btrfs_root *root,
 493                                           struct btrfs_path *path,
 494                                           u64 bytenr, u64 parent,
 495                                           u64 ref_root, u64 ref_generation,
 496                                           u64 owner_objectid, int del)
 497 {
 498         struct btrfs_key key;
 499         struct btrfs_extent_ref *ref;
 500         struct extent_buffer *leaf;
 501         u64 ref_objectid;
 502         int ret;
 503
 504         key.objectid = bytenr;
 505         key.type = BTRFS_EXTENT_REF_KEY;
 506         key.offset = parent;
 507
 508         ret = btrfs_search_slot(trans, root, &key, path, del ? -1 : 0, 1);
 509         if (ret < 0)
 510                 goto out;
 511         if (ret > 0) {
 512                 ret = -ENOENT;
 513                 goto out;
 514         }
 515
 516         leaf = path->nodes[0];
 517         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 518         ref_objectid = btrfs_ref_objectid(leaf, ref);
 519         if (btrfs_ref_root(leaf, ref) != ref_root ||
 520             btrfs_ref_generation(leaf, ref) != ref_generation ||
 521             (ref_objectid != owner_objectid &&
 522              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 523                 ret = -EIO;
 524                 WARN_ON(1);
 525                 goto out;
 526         }
 527         ret = 0;
 528 out:
 529         return ret;
 530 }
 531
 532 /*
 533  * updates all the backrefs that are pending on update_list for the
 534  * extent_root
 535  */
 536 static noinline int update_backrefs(struct btrfs_trans_handle *trans,
 537                                     struct btrfs_root *extent_root,
 538                                     struct btrfs_path *path,
 539                                     struct list_head *update_list)
 540 {
 541         struct btrfs_key key;
 542         struct btrfs_extent_ref *ref;
 543         struct btrfs_fs_info *info = extent_root->fs_info;
 544         struct pending_extent_op *op;
 545         struct extent_buffer *leaf;
 546         int ret = 0;
 547         struct list_head *cur = update_list->next;
 548         u64 ref_objectid;
 549         u64 ref_root = extent_root->root_key.objectid;
 550
 551         op = list_entry(cur, struct pending_extent_op, list);
 552
 553 search:
 554         key.objectid = op->bytenr;
 555         key.type = BTRFS_EXTENT_REF_KEY;
 556         key.offset = op->orig_parent;
 557
 558         ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
 559         BUG_ON(ret);
 560
 561         leaf = path->nodes[0];
 562
 563 loop:
 564         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 565
 566         ref_objectid = btrfs_ref_objectid(leaf, ref);
 567
 568         if (btrfs_ref_root(leaf, ref) != ref_root ||
 569             btrfs_ref_generation(leaf, ref) != op->orig_generation ||
 570             (ref_objectid != op->level &&
 571              ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
 572                 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
 573                        "root %llu, owner %u\n",
 574                        (unsigned long long)op->bytenr,
 575                        (unsigned long long)op->orig_parent,
 576                        (unsigned long long)ref_root, op->level);
 577                 btrfs_print_leaf(extent_root, leaf);
 578                 BUG();
 579         }
 580
 581         key.objectid = op->bytenr;
 582         key.offset = op->parent;
 583         key.type = BTRFS_EXTENT_REF_KEY;
 584         ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
 585         BUG_ON(ret);
 586         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 587         btrfs_set_ref_generation(leaf, ref, op->generation);
 588
 589         cur = cur->next;
 590
 591         list_del_init(&op->list);
 592         unlock_extent(&info->extent_ins, op->bytenr,
 593                       op->bytenr + op->num_bytes - 1, GFP_NOFS);
 594         kfree(op);
 595
 596         if (cur == update_list) {
 597                 btrfs_mark_buffer_dirty(path->nodes[0]);
 598                 btrfs_release_path(extent_root, path);
 599                 goto out;
 600         }
 601
 602         op = list_entry(cur, struct pending_extent_op, list);
 603
 604         path->slots[0]++;
 605         while (path->slots[0] < btrfs_header_nritems(leaf)) {
 606                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 607                 if (key.objectid == op->bytenr &&
 608                     key.type == BTRFS_EXTENT_REF_KEY)
 609                         goto loop;
 610                 path->slots[0]++;
 611         }
 612
 613         btrfs_mark_buffer_dirty(path->nodes[0]);
 614         btrfs_release_path(extent_root, path);
 615         goto search;
 616
 617 out:
 618         return 0;
 619 }
 620
 621 static noinline int insert_extents(struct btrfs_trans_handle *trans,
 622                                    struct btrfs_root *extent_root,
 623                                    struct btrfs_path *path,
 624                                    struct list_head *insert_list, int nr)
 625 {
 626         struct btrfs_key *keys;
 627         u32 *data_size;
 628         struct pending_extent_op *op;
 629         struct extent_buffer *leaf;
 630         struct list_head *cur = insert_list->next;
 631         struct btrfs_fs_info *info = extent_root->fs_info;
 632         u64 ref_root = extent_root->root_key.objectid;
 633         int i = 0, last = 0, ret;
 634         int total = nr * 2;
 635
 636         if (!nr)
 637                 return 0;
 638
 639         keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
 640         if (!keys)
 641                 return -ENOMEM;
 642
 643         data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
 644         if (!data_size) {
 645                 kfree(keys);
 646                 return -ENOMEM;
 647         }
 648
 649         list_for_each_entry(op, insert_list, list) {
 650                 keys[i].objectid = op->bytenr;
 651                 keys[i].offset = op->num_bytes;
 652                 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
 653                 data_size[i] = sizeof(struct btrfs_extent_item);
 654                 i++;
 655
 656                 keys[i].objectid = op->bytenr;
 657                 keys[i].offset = op->parent;
 658                 keys[i].type = BTRFS_EXTENT_REF_KEY;
 659                 data_size[i] = sizeof(struct btrfs_extent_ref);
 660                 i++;
 661         }
 662
 663         op = list_entry(cur, struct pending_extent_op, list);
 664         i = 0;
 665         while (i < total) {
 666                 int c;
 667                 ret = btrfs_insert_some_items(trans, extent_root, path,
 668                                               keys+i, data_size+i, total-i);
 669                 BUG_ON(ret < 0);
 670
 671                 if (last && ret > 1)
 672                         BUG();
 673
 674                 leaf = path->nodes[0];
 675                 for (c = 0; c < ret; c++) {
 676                         int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
 677
 678                         /*
 679                          * if the first item we inserted was a backref, then
 680                          * the EXTENT_ITEM will be the odd c's, else it will
 681                          * be the even c's
 682                          */
 683                         if ((ref_first && (c % 2)) ||
 684                             (!ref_first && !(c % 2))) {
 685                                 struct btrfs_extent_item *itm;
 686
 687                                 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
 688                                                      struct btrfs_extent_item);
 689                                 btrfs_set_extent_refs(path->nodes[0], itm, 1);
 690                                 op->del++;
 691                         } else {
 692                                 struct btrfs_extent_ref *ref;
 693
 694                                 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
 695                                                      struct btrfs_extent_ref);
 696                                 btrfs_set_ref_root(leaf, ref, ref_root);
 697                                 btrfs_set_ref_generation(leaf, ref,
 698                                                          op->generation);
 699                                 btrfs_set_ref_objectid(leaf, ref, op->level);
 700                                 btrfs_set_ref_num_refs(leaf, ref, 1);
 701                                 op->del++;
 702                         }
 703
 704                         /*
 705                          * using del to see when its ok to free up the
 706                          * pending_extent_op.  In the case where we insert the
 707                          * last item on the list in order to help do batching
 708                          * we need to not free the extent op until we actually
 709                          * insert the extent_item
 710                          */
 711                         if (op->del == 2) {
 712                                 unlock_extent(&info->extent_ins, op->bytenr,
 713                                               op->bytenr + op->num_bytes - 1,
 714                                               GFP_NOFS);
 715                                 cur = cur->next;
 716                                 list_del_init(&op->list);
 717                                 kfree(op);
 718                                 if (cur != insert_list)
 719                                         op = list_entry(cur,
 720                                                 struct pending_extent_op,
 721                                                 list);
 722                         }
 723                 }
 724                 btrfs_mark_buffer_dirty(leaf);
 725                 btrfs_release_path(extent_root, path);
 726
 727                 /*
 728                  * Ok backref's and items usually go right next to eachother,
 729                  * but if we could only insert 1 item that means that we
 730                  * inserted on the end of a leaf, and we have no idea what may
 731                  * be on the next leaf so we just play it safe.  In order to
 732                  * try and help this case we insert the last thing on our
 733                  * insert list so hopefully it will end up being the last
 734                  * thing on the leaf and everything else will be before it,
 735                  * which will let us insert a whole bunch of items at the same
 736                  * time.
 737                  */
 738                 if (ret == 1 && !last && (i + ret < total)) {
 739                         /*
 740                          * last: where we will pick up the next time around
 741                          * i: our current key to insert, will be total - 1
 742                          * cur: the current op we are screwing with
 743                          * op: duh
 744                          */
 745                         last = i + ret;
 746                         i = total - 1;
 747                         cur = insert_list->prev;
 748                         op = list_entry(cur, struct pending_extent_op, list);
 749                 } else if (last) {
 750                         /*
 751                          * ok we successfully inserted the last item on the
 752                          * list, lets reset everything
 753                          *
 754                          * i: our current key to insert, so where we left off
 755                          *    last time
 756                          * last: done with this
 757                          * cur: the op we are messing with
 758                          * op: duh
 759                          * total: since we inserted the last key, we need to
 760                          *        decrement total so we dont overflow
 761                          */
 762                         i = last;
 763                         last = 0;
 764                         total--;
 765                         if (i < total) {
 766                                 cur = insert_list->next;
 767                                 op = list_entry(cur, struct pending_extent_op,
 768                                                 list);
 769                         }
 770                 } else {
 771                         i += ret;
 772                 }
 773
 774                 cond_resched();
 775         }
 776         ret = 0;
 777         kfree(keys);
 778         kfree(data_size);
 779         return ret;
 780 }
 781
 782 static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
 783                                           struct btrfs_root *root,
 784                                           struct btrfs_path *path,
 785                                           u64 bytenr, u64 parent,
 786                                           u64 ref_root, u64 ref_generation,
 787                                           u64 owner_objectid)
 788 {
 789         struct btrfs_key key;
 790         struct extent_buffer *leaf;
 791         struct btrfs_extent_ref *ref;
 792         u32 num_refs;
 793         int ret;
 794
 795         key.objectid = bytenr;
 796         key.type = BTRFS_EXTENT_REF_KEY;
 797         key.offset = parent;
 798
 799         ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*ref));
 800         if (ret == 0) {
 801                 leaf = path->nodes[0];
 802                 ref = btrfs_item_ptr(leaf, path->slots[0],
 803                                      struct btrfs_extent_ref);
 804                 btrfs_set_ref_root(leaf, ref, ref_root);
 805                 btrfs_set_ref_generation(leaf, ref, ref_generation);
 806                 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
 807                 btrfs_set_ref_num_refs(leaf, ref, 1);
 808         } else if (ret == -EEXIST) {
 809                 u64 existing_owner;
 810                 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
 811                 leaf = path->nodes[0];
 812                 ref = btrfs_item_ptr(leaf, path->slots[0],
 813                                      struct btrfs_extent_ref);
 814                 if (btrfs_ref_root(leaf, ref) != ref_root ||
 815                     btrfs_ref_generation(leaf, ref) != ref_generation) {
 816                         ret = -EIO;
 817                         WARN_ON(1);
 818                         goto out;
 819                 }
 820
 821                 num_refs = btrfs_ref_num_refs(leaf, ref);
 822                 BUG_ON(num_refs == 0);
 823                 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1);
 824
 825                 existing_owner = btrfs_ref_objectid(leaf, ref);
 826                 if (existing_owner != owner_objectid &&
 827                     existing_owner != BTRFS_MULTIPLE_OBJECTIDS) {
 828                         btrfs_set_ref_objectid(leaf, ref,
 829                                         BTRFS_MULTIPLE_OBJECTIDS);
 830                 }
 831                 ret = 0;
 832         } else {
 833                 goto out;
 834         }
 835         btrfs_mark_buffer_dirty(path->nodes[0]);
 836 out:
 837         btrfs_release_path(root, path);
 838         return ret;
 839 }
 840
 841 static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
 842                                           struct btrfs_root *root,
 843                                           struct btrfs_path *path)
 844 {
 845         struct extent_buffer *leaf;
 846         struct btrfs_extent_ref *ref;
 847         u32 num_refs;
 848         int ret = 0;
 849
 850         leaf = path->nodes[0];
 851         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
 852         num_refs = btrfs_ref_num_refs(leaf, ref);
 853         BUG_ON(num_refs == 0);
 854         num_refs -= 1;
 855         if (num_refs == 0) {
 856                 ret = btrfs_del_item(trans, root, path);
 857         } else {
 858                 btrfs_set_ref_num_refs(leaf, ref, num_refs);
 859                 btrfs_mark_buffer_dirty(leaf);
 860         }
 861         btrfs_release_path(root, path);
 862         return ret;
 863 }
 864
 865 #ifdef BIO_RW_DISCARD
 866 static void btrfs_issue_discard(struct block_device *bdev,
 867                                 u64 start, u64 len)
 868 {
 869         blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_KERNEL);
 870 }
 871 #endif
 872
 873 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
 874                                 u64 num_bytes)
 875 {
 876 #ifdef BIO_RW_DISCARD
 877         int ret;
 878         u64 map_length = num_bytes;
 879         struct btrfs_multi_bio *multi = NULL;
 880
 881         /* Tell the block device(s) that the sectors can be discarded */
 882         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
 883                               bytenr, &map_length, &multi, 0);
 884         if (!ret) {
 885                 struct btrfs_bio_stripe *stripe = multi->stripes;
 886                 int i;
 887
 888                 if (map_length > num_bytes)
 889                         map_length = num_bytes;
 890
 891                 for (i = 0; i < multi->num_stripes; i++, stripe++) {
 892                         btrfs_issue_discard(stripe->dev->bdev,
 893                                             stripe->physical,
 894                                             map_length);
 895                 }
 896                 kfree(multi);
 897         }
 898
 899         return ret;
 900 #else
 901         return 0;
 902 #endif
 903 }
 904
 905 static noinline int free_extents(struct btrfs_trans_handle *trans,
 906                                  struct btrfs_root *extent_root,
 907                                  struct list_head *del_list)
 908 {
 909         struct btrfs_fs_info *info = extent_root->fs_info;
 910         struct btrfs_path *path;
 911         struct btrfs_key key, found_key;
 912         struct extent_buffer *leaf;
 913         struct list_head *cur;
 914         struct pending_extent_op *op;
 915         struct btrfs_extent_item *ei;
 916         int ret, num_to_del, extent_slot = 0, found_extent = 0;
 917         u32 refs;
 918         u64 bytes_freed = 0;
 919
 920         path = btrfs_alloc_path();
 921         if (!path)
 922                 return -ENOMEM;
 923         path->reada = 1;
 924
 925 search:
 926         /* search for the backref for the current ref we want to delete */
 927         cur = del_list->next;
 928         op = list_entry(cur, struct pending_extent_op, list);
 929         ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
 930                                     op->orig_parent,
 931                                     extent_root->root_key.objectid,
 932                                     op->orig_generation, op->level, 1);
 933         if (ret) {
 934                 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
 935                        "root %llu gen %llu owner %u\n",
 936                        (unsigned long long)op->bytenr,
 937                        (unsigned long long)extent_root->root_key.objectid,
 938                        (unsigned long long)op->orig_generation, op->level);
 939                 btrfs_print_leaf(extent_root, path->nodes[0]);
 940                 WARN_ON(1);
 941                 goto out;
 942         }
 943
 944         extent_slot = path->slots[0];
 945         num_to_del = 1;
 946         found_extent = 0;
 947
 948         /*
 949          * if we aren't the first item on the leaf we can move back one and see
 950          * if our ref is right next to our extent item
 951          */
 952         if (likely(extent_slot)) {
 953                 extent_slot--;
 954                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 955                                       extent_slot);
 956                 if (found_key.objectid == op->bytenr &&
 957                     found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 958                     found_key.offset == op->num_bytes) {
 959                         num_to_del++;
 960                         found_extent = 1;
 961                 }
 962         }
 963
 964         /*
 965          * if we didn't find the extent we need to delete the backref and then
 966          * search for the extent item key so we can update its ref count
 967          */
 968         if (!found_extent) {
 969                 key.objectid = op->bytenr;
 970                 key.type = BTRFS_EXTENT_ITEM_KEY;
 971                 key.offset = op->num_bytes;
 972
 973                 ret = remove_extent_backref(trans, extent_root, path);
 974                 BUG_ON(ret);
 975                 btrfs_release_path(extent_root, path);
 976                 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
 977                 BUG_ON(ret);
 978                 extent_slot = path->slots[0];
 979         }
 980
 981         /* this is where we update the ref count for the extent */
 982         leaf = path->nodes[0];
 983         ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
 984         refs = btrfs_extent_refs(leaf, ei);
 985         BUG_ON(refs == 0);
 986         refs--;
 987         btrfs_set_extent_refs(leaf, ei, refs);
 988
 989         btrfs_mark_buffer_dirty(leaf);
 990
 991         /*
 992          * This extent needs deleting.  The reason cur_slot is extent_slot +
 993          * num_to_del is because extent_slot points to the slot where the extent
 994          * is, and if the backref was not right next to the extent we will be
 995          * deleting at least 1 item, and will want to start searching at the
 996          * slot directly next to extent_slot.  However if we did find the
 997          * backref next to the extent item them we will be deleting at least 2
 998          * items and will want to start searching directly after the ref slot
 999          */
1000         if (!refs) {
1001                 struct list_head *pos, *n, *end;
1002                 int cur_slot = extent_slot+num_to_del;
1003                 u64 super_used;
1004                 u64 root_used;
1005
1006                 path->slots[0] = extent_slot;
1007                 bytes_freed = op->num_bytes;
1008
1009                 mutex_lock(&info->pinned_mutex);
1010                 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1011                                      op->num_bytes, op->level >=
1012                                      BTRFS_FIRST_FREE_OBJECTID);
1013                 mutex_unlock(&info->pinned_mutex);
1014                 BUG_ON(ret < 0);
1015                 op->del = ret;
1016
1017                 /*
1018                  * we need to see if we can delete multiple things at once, so
1019                  * start looping through the list of extents we are wanting to
1020                  * delete and see if their extent/backref's are right next to
1021                  * eachother and the extents only have 1 ref
1022                  */
1023                 for (pos = cur->next; pos != del_list; pos = pos->next) {
1024                         struct pending_extent_op *tmp;
1025
1026                         tmp = list_entry(pos, struct pending_extent_op, list);
1027
1028                         /* we only want to delete extent+ref at this stage */
1029                         if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1030                                 break;
1031
1032                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1033                         if (found_key.objectid != tmp->bytenr ||
1034                             found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1035                             found_key.offset != tmp->num_bytes)
1036                                 break;
1037
1038                         /* check to make sure this extent only has one ref */
1039                         ei = btrfs_item_ptr(leaf, cur_slot,
1040                                             struct btrfs_extent_item);
1041                         if (btrfs_extent_refs(leaf, ei) != 1)
1042                                 break;
1043
1044                         btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1045                         if (found_key.objectid != tmp->bytenr ||
1046                             found_key.type != BTRFS_EXTENT_REF_KEY ||
1047                             found_key.offset != tmp->orig_parent)
1048                                 break;
1049
1050                         /*
1051                          * the ref is right next to the extent, we can set the
1052                          * ref count to 0 since we will delete them both now
1053                          */
1054                         btrfs_set_extent_refs(leaf, ei, 0);
1055
1056                         /* pin down the bytes for this extent */
1057                         mutex_lock(&info->pinned_mutex);
1058                         ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1059                                              tmp->num_bytes, tmp->level >=
1060                                              BTRFS_FIRST_FREE_OBJECTID);
1061                         mutex_unlock(&info->pinned_mutex);
1062                         BUG_ON(ret < 0);
1063
1064                         /*
1065                          * use the del field to tell if we need to go ahead and
1066                          * free up the extent when we delete the item or not.
1067                          */
1068                         tmp->del = ret;
1069                         bytes_freed += tmp->num_bytes;
1070
1071                         num_to_del += 2;
1072                         cur_slot += 2;
1073                 }
1074                 end = pos;
1075
1076                 /* update the free space counters */
1077                 spin_lock(&info->delalloc_lock);
1078                 super_used = btrfs_super_bytes_used(&info->super_copy);
1079                 btrfs_set_super_bytes_used(&info->super_copy,
1080                                            super_used - bytes_freed);
1081
1082                 root_used = btrfs_root_used(&extent_root->root_item);
1083                 btrfs_set_root_used(&extent_root->root_item,
1084                                     root_used - bytes_freed);
1085                 spin_unlock(&info->delalloc_lock);
1086
1087                 /* delete the items */
1088                 ret = btrfs_del_items(trans, extent_root, path,
1089                                       path->slots[0], num_to_del);
1090                 BUG_ON(ret);
1091
1092                 /*
1093                  * loop through the extents we deleted and do the cleanup work
1094                  * on them
1095                  */
1096                 for (pos = cur, n = pos->next; pos != end;
1097                      pos = n, n = pos->next) {
1098                         struct pending_extent_op *tmp;
1099                         tmp = list_entry(pos, struct pending_extent_op, list);
1100
1101                         /*
1102                          * remember tmp->del tells us wether or not we pinned
1103                          * down the extent
1104                          */
1105                         ret = update_block_group(trans, extent_root,
1106                                                  tmp->bytenr, tmp->num_bytes, 0,
1107                                                  tmp->del);
1108                         BUG_ON(ret);
1109
1110                         list_del_init(&tmp->list);
1111                         unlock_extent(&info->extent_ins, tmp->bytenr,
1112                                       tmp->bytenr + tmp->num_bytes - 1,
1113                                       GFP_NOFS);
1114                         kfree(tmp);
1115                 }
1116         } else if (refs && found_extent) {
1117                 /*
1118                  * the ref and extent were right next to eachother, but the
1119                  * extent still has a ref, so just free the backref and keep
1120                  * going
1121                  */
1122                 ret = remove_extent_backref(trans, extent_root, path);
1123                 BUG_ON(ret);
1124
1125                 list_del_init(&op->list);
1126                 unlock_extent(&info->extent_ins, op->bytenr,
1127                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1128                 kfree(op);
1129         } else {
1130                 /*
1131                  * the extent has multiple refs and the backref we were looking
1132                  * for was not right next to it, so just unlock and go next,
1133                  * we're good to go
1134                  */
1135                 list_del_init(&op->list);
1136                 unlock_extent(&info->extent_ins, op->bytenr,
1137                               op->bytenr + op->num_bytes - 1, GFP_NOFS);
1138                 kfree(op);
1139         }
1140
1141         btrfs_release_path(extent_root, path);
1142         if (!list_empty(del_list))
1143                 goto search;
1144
1145 out:
1146         btrfs_free_path(path);
1147         return ret;
1148 }
1149
1150 static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1151                                      struct btrfs_root *root, u64 bytenr,
1152                                      u64 orig_parent, u64 parent,
1153                                      u64 orig_root, u64 ref_root,
1154                                      u64 orig_generation, u64 ref_generation,
1155                                      u64 owner_objectid)
1156 {
1157         int ret;
1158         struct btrfs_root *extent_root = root->fs_info->extent_root;
1159         struct btrfs_path *path;
1160
1161         if (root == root->fs_info->extent_root) {
1162                 struct pending_extent_op *extent_op;
1163                 u64 num_bytes;
1164
1165                 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1166                 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1167                 mutex_lock(&root->fs_info->extent_ins_mutex);
1168                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1169                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1170                         u64 priv;
1171                         ret = get_state_private(&root->fs_info->extent_ins,
1172                                                 bytenr, &priv);
1173                         BUG_ON(ret);
1174                         extent_op = (struct pending_extent_op *)
1175                                                         (unsigned long)priv;
1176                         BUG_ON(extent_op->parent != orig_parent);
1177                         BUG_ON(extent_op->generation != orig_generation);
1178
1179                         extent_op->parent = parent;
1180                         extent_op->generation = ref_generation;
1181                 } else {
1182                         extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
1183                         BUG_ON(!extent_op);
1184
1185                         extent_op->type = PENDING_BACKREF_UPDATE;
1186                         extent_op->bytenr = bytenr;
1187                         extent_op->num_bytes = num_bytes;
1188                         extent_op->parent = parent;
1189                         extent_op->orig_parent = orig_parent;
1190                         extent_op->generation = ref_generation;
1191                         extent_op->orig_generation = orig_generation;
1192                         extent_op->level = (int)owner_objectid;
1193                         INIT_LIST_HEAD(&extent_op->list);
1194                         extent_op->del = 0;
1195
1196                         set_extent_bits(&root->fs_info->extent_ins,
1197                                         bytenr, bytenr + num_bytes - 1,
1198                                         EXTENT_WRITEBACK, GFP_NOFS);
1199                         set_state_private(&root->fs_info->extent_ins,
1200                                           bytenr, (unsigned long)extent_op);
1201                 }
1202                 mutex_unlock(&root->fs_info->extent_ins_mutex);
1203                 return 0;
1204         }
1205
1206         path = btrfs_alloc_path();
1207         if (!path)
1208                 return -ENOMEM;
1209         ret = lookup_extent_backref(trans, extent_root, path,
1210                                     bytenr, orig_parent, orig_root,
1211                                     orig_generation, owner_objectid, 1);
1212         if (ret)
1213                 goto out;
1214         ret = remove_extent_backref(trans, extent_root, path);
1215         if (ret)
1216                 goto out;
1217         ret = insert_extent_backref(trans, extent_root, path, bytenr,
1218                                     parent, ref_root, ref_generation,
1219                                     owner_objectid);
1220         BUG_ON(ret);
1221         finish_current_insert(trans, extent_root, 0);
1222         del_pending_extents(trans, extent_root, 0);
1223 out:
1224         btrfs_free_path(path);
1225         return ret;
1226 }
1227
1228 int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1229                             struct btrfs_root *root, u64 bytenr,
1230                             u64 orig_parent, u64 parent,
1231                             u64 ref_root, u64 ref_generation,
1232                             u64 owner_objectid)
1233 {
1234         int ret;
1235         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1236             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1237                 return 0;
1238         ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent,
1239                                         parent, ref_root, ref_root,
1240                                         ref_generation, ref_generation,
1241                                         owner_objectid);
1242         return ret;
1243 }
1244
1245 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1246                                   struct btrfs_root *root, u64 bytenr,
1247                                   u64 orig_parent, u64 parent,
1248                                   u64 orig_root, u64 ref_root,
1249                                   u64 orig_generation, u64 ref_generation,
1250                                   u64 owner_objectid)
1251 {
1252         struct btrfs_path *path;
1253         int ret;
1254         struct btrfs_key key;
1255         struct extent_buffer *l;
1256         struct btrfs_extent_item *item;
1257         u32 refs;
1258
1259         path = btrfs_alloc_path();
1260         if (!path)
1261                 return -ENOMEM;
1262
1263         path->reada = 1;
1264         key.objectid = bytenr;
1265         key.type = BTRFS_EXTENT_ITEM_KEY;
1266         key.offset = (u64)-1;
1267
1268         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1269                                 0, 1);
1270         if (ret < 0)
1271                 return ret;
1272         BUG_ON(ret == 0 || path->slots[0] == 0);
1273
1274         path->slots[0]--;
1275         l = path->nodes[0];
1276
1277         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1278         if (key.objectid != bytenr) {
1279                 btrfs_print_leaf(root->fs_info->extent_root, path->nodes[0]);
1280                 printk(KERN_ERR "btrfs wanted %llu found %llu\n",
1281                        (unsigned long long)bytenr,
1282                        (unsigned long long)key.objectid);
1283                 BUG();
1284         }
1285         BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1286
1287         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1288         refs = btrfs_extent_refs(l, item);
1289         btrfs_set_extent_refs(l, item, refs + 1);
1290         btrfs_mark_buffer_dirty(path->nodes[0]);
1291
1292         btrfs_release_path(root->fs_info->extent_root, path);
1293
1294         path->reada = 1;
1295         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1296                                     path, bytenr, parent,
1297                                     ref_root, ref_generation,
1298                                     owner_objectid);
1299         BUG_ON(ret);
1300         finish_current_insert(trans, root->fs_info->extent_root, 0);
1301         del_pending_extents(trans, root->fs_info->extent_root, 0);
1302
1303         btrfs_free_path(path);
1304         return 0;
1305 }
1306
1307 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1308                          struct btrfs_root *root,
1309                          u64 bytenr, u64 num_bytes, u64 parent,
1310                          u64 ref_root, u64 ref_generation,
1311                          u64 owner_objectid)
1312 {
1313         int ret;
1314         if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1315             owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1316                 return 0;
1317         ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent,
1318                                      0, ref_root, 0, ref_generation,
1319                                      owner_objectid);
1320         return ret;
1321 }
1322
1323 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1324                          struct btrfs_root *root)
1325 {
1326         finish_current_insert(trans, root->fs_info->extent_root, 1);
1327         del_pending_extents(trans, root->fs_info->extent_root, 1);
1328         return 0;
1329 }
1330
1331 int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans,
1332                             struct btrfs_root *root, u64 bytenr,
1333                             u64 num_bytes, u32 *refs)
1334 {
1335         struct btrfs_path *path;
1336         int ret;
1337         struct btrfs_key key;
1338         struct extent_buffer *l;
1339         struct btrfs_extent_item *item;
1340
1341         WARN_ON(num_bytes < root->sectorsize);
1342         path = btrfs_alloc_path();
1343         path->reada = 1;
1344         key.objectid = bytenr;
1345         key.offset = num_bytes;
1346         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1347         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1348                                 0, 0);
1349         if (ret < 0)
1350                 goto out;
1351         if (ret != 0) {
1352                 btrfs_print_leaf(root, path->nodes[0]);
1353                 printk(KERN_INFO "btrfs failed to find block number %llu\n",
1354                        (unsigned long long)bytenr);
1355                 BUG();
1356         }
1357         l = path->nodes[0];
1358         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1359         *refs = btrfs_extent_refs(l, item);
1360 out:
1361         btrfs_free_path(path);
1362         return 0;
1363 }
1364
1365 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
1366                           struct btrfs_root *root, u64 objectid, u64 bytenr)
1367 {
1368         struct btrfs_root *extent_root = root->fs_info->extent_root;
1369         struct btrfs_path *path;
1370         struct extent_buffer *leaf;
1371         struct btrfs_extent_ref *ref_item;
1372         struct btrfs_key key;
1373         struct btrfs_key found_key;
1374         u64 ref_root;
1375         u64 last_snapshot;
1376         u32 nritems;
1377         int ret;
1378
1379         key.objectid = bytenr;
1380         key.offset = (u64)-1;
1381         key.type = BTRFS_EXTENT_ITEM_KEY;
1382
1383         path = btrfs_alloc_path();
1384         ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1385         if (ret < 0)
1386                 goto out;
1387         BUG_ON(ret == 0);
1388
1389         ret = -ENOENT;
1390         if (path->slots[0] == 0)
1391                 goto out;
1392
1393         path->slots[0]--;
1394         leaf = path->nodes[0];
1395         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1396
1397         if (found_key.objectid != bytenr ||
1398             found_key.type != BTRFS_EXTENT_ITEM_KEY)
1399                 goto out;
1400
1401         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1402         while (1) {
1403                 leaf = path->nodes[0];
1404                 nritems = btrfs_header_nritems(leaf);
1405                 if (path->slots[0] >= nritems) {
1406                         ret = btrfs_next_leaf(extent_root, path);
1407                         if (ret < 0)
1408                                 goto out;
1409                         if (ret == 0)
1410                                 continue;
1411                         break;
1412                 }
1413                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1414                 if (found_key.objectid != bytenr)
1415                         break;
1416
1417                 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
1418                         path->slots[0]++;
1419                         continue;
1420                 }
1421
1422                 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1423                                           struct btrfs_extent_ref);
1424                 ref_root = btrfs_ref_root(leaf, ref_item);
1425                 if ((ref_root != root->root_key.objectid &&
1426                      ref_root != BTRFS_TREE_LOG_OBJECTID) ||
1427                      objectid != btrfs_ref_objectid(leaf, ref_item)) {
1428                         ret = 1;
1429                         goto out;
1430                 }
1431                 if (btrfs_ref_generation(leaf, ref_item) <= last_snapshot) {
1432                         ret = 1;
1433                         goto out;
1434                 }
1435
1436                 path->slots[0]++;
1437         }
1438         ret = 0;
1439 out:
1440         btrfs_free_path(path);
1441         return ret;
1442 }
1443
1444 int btrfs_cache_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1445                     struct extent_buffer *buf, u32 nr_extents)
1446 {
1447         struct btrfs_key key;
1448         struct btrfs_file_extent_item *fi;
1449         u64 root_gen;
1450         u32 nritems;
1451         int i;
1452         int level;
1453         int ret = 0;
1454         int shared = 0;
1455
1456         if (!root->ref_cows)
1457                 return 0;
1458
1459         if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1460                 shared = 0;
1461                 root_gen = root->root_key.offset;
1462         } else {
1463                 shared = 1;
1464                 root_gen = trans->transid - 1;
1465         }
1466
1467         level = btrfs_header_level(buf);
1468         nritems = btrfs_header_nritems(buf);
1469
1470         if (level == 0) {
1471                 struct btrfs_leaf_ref *ref;
1472                 struct btrfs_extent_info *info;
1473
1474                 ref = btrfs_alloc_leaf_ref(root, nr_extents);
1475                 if (!ref) {
1476                         ret = -ENOMEM;
1477                         goto out;
1478                 }
1479
1480                 ref->root_gen = root_gen;
1481                 ref->bytenr = buf->start;
1482                 ref->owner = btrfs_header_owner(buf);
1483                 ref->generation = btrfs_header_generation(buf);
1484                 ref->nritems = nr_extents;
1485                 info = ref->extents;
1486
1487                 for (i = 0; nr_extents > 0 && i < nritems; i++) {
1488                         u64 disk_bytenr;
1489                         btrfs_item_key_to_cpu(buf, &key, i);
1490                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1491                                 continue;
1492                         fi = btrfs_item_ptr(buf, i,
1493                                             struct btrfs_file_extent_item);
1494                         if (btrfs_file_extent_type(buf, fi) ==
1495                             BTRFS_FILE_EXTENT_INLINE)
1496                                 continue;
1497                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1498                         if (disk_bytenr == 0)
1499                                 continue;
1500
1501                         info->bytenr = disk_bytenr;
1502                         info->num_bytes =
1503                                 btrfs_file_extent_disk_num_bytes(buf, fi);
1504                         info->objectid = key.objectid;
1505                         info->offset = key.offset;
1506                         info++;
1507                 }
1508
1509                 ret = btrfs_add_leaf_ref(root, ref, shared);
1510                 if (ret == -EEXIST && shared) {
1511                         struct btrfs_leaf_ref *old;
1512                         old = btrfs_lookup_leaf_ref(root, ref->bytenr);
1513                         BUG_ON(!old);
1514                         btrfs_remove_leaf_ref(root, old);
1515                         btrfs_free_leaf_ref(root, old);
1516                         ret = btrfs_add_leaf_ref(root, ref, shared);
1517                 }
1518                 WARN_ON(ret);
1519                 btrfs_free_leaf_ref(root, ref);
1520         }
1521 out:
1522         return ret;
1523 }
1524
1525 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1526                   struct extent_buffer *orig_buf, struct extent_buffer *buf,
1527                   u32 *nr_extents)
1528 {
1529         u64 bytenr;
1530         u64 ref_root;
1531         u64 orig_root;
1532         u64 ref_generation;
1533         u64 orig_generation;
1534         u32 nritems;
1535         u32 nr_file_extents = 0;
1536         struct btrfs_key key;
1537         struct btrfs_file_extent_item *fi;
1538         int i;
1539         int level;
1540         int ret = 0;
1541         int faili = 0;
1542         int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1543                             u64, u64, u64, u64, u64, u64, u64, u64);
1544
1545         ref_root = btrfs_header_owner(buf);
1546         ref_generation = btrfs_header_generation(buf);
1547         orig_root = btrfs_header_owner(orig_buf);
1548         orig_generation = btrfs_header_generation(orig_buf);
1549
1550         nritems = btrfs_header_nritems(buf);
1551         level = btrfs_header_level(buf);
1552
1553         if (root->ref_cows) {
1554                 process_func = __btrfs_inc_extent_ref;
1555         } else {
1556                 if (level == 0 &&
1557                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1558                         goto out;
1559                 if (level != 0 &&
1560                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1561                         goto out;
1562                 process_func = __btrfs_update_extent_ref;
1563         }
1564
1565         for (i = 0; i < nritems; i++) {
1566                 cond_resched();
1567                 if (level == 0) {
1568                         btrfs_item_key_to_cpu(buf, &key, i);
1569                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1570                                 continue;
1571                         fi = btrfs_item_ptr(buf, i,
1572                                             struct btrfs_file_extent_item);
1573                         if (btrfs_file_extent_type(buf, fi) ==
1574                             BTRFS_FILE_EXTENT_INLINE)
1575                                 continue;
1576                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1577                         if (bytenr == 0)
1578                                 continue;
1579
1580                         nr_file_extents++;
1581
1582                         ret = process_func(trans, root, bytenr,
1583                                            orig_buf->start, buf->start,
1584                                            orig_root, ref_root,
1585                                            orig_generation, ref_generation,
1586                                            key.objectid);
1587
1588                         if (ret) {
1589                                 faili = i;
1590                                 WARN_ON(1);
1591                                 goto fail;
1592                         }
1593                 } else {
1594                         bytenr = btrfs_node_blockptr(buf, i);
1595                         ret = process_func(trans, root, bytenr,
1596                                            orig_buf->start, buf->start,
1597                                            orig_root, ref_root,
1598                                            orig_generation, ref_generation,
1599                                            level - 1);
1600                         if (ret) {
1601                                 faili = i;
1602                                 WARN_ON(1);
1603                                 goto fail;
1604                         }
1605                 }
1606         }
1607 out:
1608         if (nr_extents) {
1609                 if (level == 0)
1610                         *nr_extents = nr_file_extents;
1611                 else
1612                         *nr_extents = nritems;
1613         }
1614         return 0;
1615 fail:
1616         WARN_ON(1);
1617         return ret;
1618 }
1619
1620 int btrfs_update_ref(struct btrfs_trans_handle *trans,
1621                      struct btrfs_root *root, struct extent_buffer *orig_buf,
1622                      struct extent_buffer *buf, int start_slot, int nr)
1623
1624 {
1625         u64 bytenr;
1626         u64 ref_root;
1627         u64 orig_root;
1628         u64 ref_generation;
1629         u64 orig_generation;
1630         struct btrfs_key key;
1631         struct btrfs_file_extent_item *fi;
1632         int i;
1633         int ret;
1634         int slot;
1635         int level;
1636
1637         BUG_ON(start_slot < 0);
1638         BUG_ON(start_slot + nr > btrfs_header_nritems(buf));
1639
1640         ref_root = btrfs_header_owner(buf);
1641         ref_generation = btrfs_header_generation(buf);
1642         orig_root = btrfs_header_owner(orig_buf);
1643         orig_generation = btrfs_header_generation(orig_buf);
1644         level = btrfs_header_level(buf);
1645
1646         if (!root->ref_cows) {
1647                 if (level == 0 &&
1648                     root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
1649                         return 0;
1650                 if (level != 0 &&
1651                     root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID)
1652                         return 0;
1653         }
1654
1655         for (i = 0, slot = start_slot; i < nr; i++, slot++) {
1656                 cond_resched();
1657                 if (level == 0) {
1658                         btrfs_item_key_to_cpu(buf, &key, slot);
1659                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1660                                 continue;
1661                         fi = btrfs_item_ptr(buf, slot,
1662                                             struct btrfs_file_extent_item);
1663                         if (btrfs_file_extent_type(buf, fi) ==
1664                             BTRFS_FILE_EXTENT_INLINE)
1665                                 continue;
1666                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1667                         if (bytenr == 0)
1668                                 continue;
1669                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1670                                             orig_buf->start, buf->start,
1671                                             orig_root, ref_root,
1672                                             orig_generation, ref_generation,
1673                                             key.objectid);
1674                         if (ret)
1675                                 goto fail;
1676                 } else {
1677                         bytenr = btrfs_node_blockptr(buf, slot);
1678                         ret = __btrfs_update_extent_ref(trans, root, bytenr,
1679                                             orig_buf->start, buf->start,
1680                                             orig_root, ref_root,
1681                                             orig_generation, ref_generation,
1682                                             level - 1);
1683                         if (ret)
1684                                 goto fail;
1685                 }
1686         }
1687         return 0;
1688 fail:
1689         WARN_ON(1);
1690         return -1;
1691 }
1692
1693 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1694                                  struct btrfs_root *root,
1695                                  struct btrfs_path *path,
1696                                  struct btrfs_block_group_cache *cache)
1697 {
1698         int ret;
1699         int pending_ret;
1700         struct btrfs_root *extent_root = root->fs_info->extent_root;
1701         unsigned long bi;
1702         struct extent_buffer *leaf;
1703
1704         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1705         if (ret < 0)
1706                 goto fail;
1707         BUG_ON(ret);
1708
1709         leaf = path->nodes[0];
1710         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1711         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1712         btrfs_mark_buffer_dirty(leaf);
1713         btrfs_release_path(extent_root, path);
1714 fail:
1715         finish_current_insert(trans, extent_root, 0);
1716         pending_ret = del_pending_extents(trans, extent_root, 0);
1717         if (ret)
1718                 return ret;
1719         if (pending_ret)
1720                 return pending_ret;
1721         return 0;
1722
1723 }
1724
1725 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1726                                    struct btrfs_root *root)
1727 {
1728         struct btrfs_block_group_cache *cache, *entry;
1729         struct rb_node *n;
1730         int err = 0;
1731         int werr = 0;
1732         struct btrfs_path *path;
1733         u64 last = 0;
1734
1735         path = btrfs_alloc_path();
1736         if (!path)
1737                 return -ENOMEM;
1738
1739         while (1) {
1740                 cache = NULL;
1741                 spin_lock(&root->fs_info->block_group_cache_lock);
1742                 for (n = rb_first(&root->fs_info->block_group_cache_tree);
1743                      n; n = rb_next(n)) {
1744                         entry = rb_entry(n, struct btrfs_block_group_cache,
1745                                          cache_node);
1746                         if (entry->dirty) {
1747                                 cache = entry;
1748                                 break;
1749                         }
1750                 }
1751                 spin_unlock(&root->fs_info->block_group_cache_lock);
1752
1753                 if (!cache)
1754                         break;
1755
1756                 cache->dirty = 0;
1757                 last += cache->key.offset;
1758
1759                 err = write_one_cache_group(trans, root,
1760                                             path, cache);
1761                 /*
1762                  * if we fail to write the cache group, we want
1763                  * to keep it marked dirty in hopes that a later
1764                  * write will work
1765                  */
1766                 if (err) {
1767                         werr = err;
1768                         continue;
1769                 }
1770         }
1771         btrfs_free_path(path);
1772         return werr;
1773 }
1774
1775 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
1776 {
1777         struct btrfs_block_group_cache *block_group;
1778         int readonly = 0;
1779
1780         block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
1781         if (!block_group || block_group->ro)
1782                 readonly = 1;
1783         if (block_group)
1784                 put_block_group(block_group);
1785         return readonly;
1786 }
1787
1788 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1789                              u64 total_bytes, u64 bytes_used,
1790                              struct btrfs_space_info **space_info)
1791 {
1792         struct btrfs_space_info *found;
1793
1794         found = __find_space_info(info, flags);
1795         if (found) {
1796                 spin_lock(&found->lock);
1797                 found->total_bytes += total_bytes;
1798                 found->bytes_used += bytes_used;
1799                 found->full = 0;
1800                 spin_unlock(&found->lock);
1801                 *space_info = found;
1802                 return 0;
1803         }
1804         found = kzalloc(sizeof(*found), GFP_NOFS);
1805         if (!found)
1806                 return -ENOMEM;
1807
1808         list_add(&found->list, &info->space_info);
1809         INIT_LIST_HEAD(&found->block_groups);
1810         init_rwsem(&found->groups_sem);
1811         spin_lock_init(&found->lock);
1812         found->flags = flags;
1813         found->total_bytes = total_bytes;
1814         found->bytes_used = bytes_used;
1815         found->bytes_pinned = 0;
1816         found->bytes_reserved = 0;
1817         found->bytes_readonly = 0;
1818         found->full = 0;
1819         found->force_alloc = 0;
1820         *space_info = found;
1821         return 0;
1822 }
1823
1824 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1825 {
1826         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1827                                    BTRFS_BLOCK_GROUP_RAID1 |
1828                                    BTRFS_BLOCK_GROUP_RAID10 |
1829                                    BTRFS_BLOCK_GROUP_DUP);
1830         if (extra_flags) {
1831                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1832                         fs_info->avail_data_alloc_bits |= extra_flags;
1833                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1834                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1835                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1836                         fs_info->avail_system_alloc_bits |= extra_flags;
1837         }
1838 }
1839
1840 static void set_block_group_readonly(struct btrfs_block_group_cache *cache)
1841 {
1842         spin_lock(&cache->space_info->lock);
1843         spin_lock(&cache->lock);
1844         if (!cache->ro) {
1845                 cache->space_info->bytes_readonly += cache->key.offset -
1846                                         btrfs_block_group_used(&cache->item);
1847                 cache->ro = 1;
1848         }
1849         spin_unlock(&cache->lock);
1850         spin_unlock(&cache->space_info->lock);
1851 }
1852
1853 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
1854 {
1855         u64 num_devices = root->fs_info->fs_devices->rw_devices;
1856
1857         if (num_devices == 1)
1858                 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0);
1859         if (num_devices < 4)
1860                 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
1861
1862         if ((flags & BTRFS_BLOCK_GROUP_DUP) &&
1863             (flags & (BTRFS_BLOCK_GROUP_RAID1 |
1864                       BTRFS_BLOCK_GROUP_RAID10))) {
1865                 flags &= ~BTRFS_BLOCK_GROUP_DUP;
1866         }
1867
1868         if ((flags & BTRFS_BLOCK_GROUP_RAID1) &&
1869             (flags & BTRFS_BLOCK_GROUP_RAID10)) {
1870                 flags &= ~BTRFS_BLOCK_GROUP_RAID1;
1871         }
1872
1873         if ((flags & BTRFS_BLOCK_GROUP_RAID0) &&
1874             ((flags & BTRFS_BLOCK_GROUP_RAID1) |
1875              (flags & BTRFS_BLOCK_GROUP_RAID10) |
1876              (flags & BTRFS_BLOCK_GROUP_DUP)))
1877                 flags &= ~BTRFS_BLOCK_GROUP_RAID0;
1878         return flags;
1879 }
1880
1881 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1882                           struct btrfs_root *extent_root, u64 alloc_bytes,
1883                           u64 flags, int force)
1884 {
1885         struct btrfs_space_info *space_info;
1886         u64 thresh;
1887         int ret = 0;
1888
1889         mutex_lock(&extent_root->fs_info->chunk_mutex);
1890
1891         flags = btrfs_reduce_alloc_profile(extent_root, flags);
1892
1893         space_info = __find_space_info(extent_root->fs_info, flags);
1894         if (!space_info) {
1895                 ret = update_space_info(extent_root->fs_info, flags,
1896                                         0, 0, &space_info);
1897                 BUG_ON(ret);
1898         }
1899         BUG_ON(!space_info);
1900
1901         spin_lock(&space_info->lock);
1902         if (space_info->force_alloc) {
1903                 force = 1;
1904                 space_info->force_alloc = 0;
1905         }
1906         if (space_info->full) {
1907                 spin_unlock(&space_info->lock);
1908                 goto out;
1909         }
1910
1911         thresh = space_info->total_bytes - space_info->bytes_readonly;
1912         thresh = div_factor(thresh, 6);
1913         if (!force &&
1914            (space_info->bytes_used + space_info->bytes_pinned +
1915             space_info->bytes_reserved + alloc_bytes) < thresh) {
1916                 spin_unlock(&space_info->lock);
1917                 goto out;
1918         }
1919         spin_unlock(&space_info->lock);
1920
1921         ret = btrfs_alloc_chunk(trans, extent_root, flags);
1922         if (ret)
1923                 space_info->full = 1;
1924 out:
1925         mutex_unlock(&extent_root->fs_info->chunk_mutex);
1926         return ret;
1927 }
1928
1929 static int update_block_group(struct btrfs_trans_handle *trans,
1930                               struct btrfs_root *root,
1931                               u64 bytenr, u64 num_bytes, int alloc,
1932                               int mark_free)
1933 {
1934         struct btrfs_block_group_cache *cache;
1935         struct btrfs_fs_info *info = root->fs_info;
1936         u64 total = num_bytes;
1937         u64 old_val;
1938         u64 byte_in_group;
1939
1940         while (total) {
1941                 cache = btrfs_lookup_block_group(info, bytenr);
1942                 if (!cache)
1943                         return -1;
1944                 byte_in_group = bytenr - cache->key.objectid;
1945                 WARN_ON(byte_in_group > cache->key.offset);
1946
1947                 spin_lock(&cache->space_info->lock);
1948                 spin_lock(&cache->lock);
1949                 cache->dirty = 1;
1950                 old_val = btrfs_block_group_used(&cache->item);
1951                 num_bytes = min(total, cache->key.offset - byte_in_group);
1952                 if (alloc) {
1953                         old_val += num_bytes;
1954                         cache->space_info->bytes_used += num_bytes;
1955                         if (cache->ro)
1956                                 cache->space_info->bytes_readonly -= num_bytes;
1957                         btrfs_set_block_group_used(&cache->item, old_val);
1958                         spin_unlock(&cache->lock);
1959                         spin_unlock(&cache->space_info->lock);
1960                 } else {
1961                         old_val -= num_bytes;
1962                         cache->space_info->bytes_used -= num_bytes;
1963                         if (cache->ro)
1964                                 cache->space_info->bytes_readonly += num_bytes;
1965                         btrfs_set_block_group_used(&cache->item, old_val);
1966                         spin_unlock(&cache->lock);
1967                         spin_unlock(&cache->space_info->lock);
1968                         if (mark_free) {
1969                                 int ret;
1970
1971                                 ret = btrfs_discard_extent(root, bytenr,
1972                                                            num_bytes);
1973                                 WARN_ON(ret);
1974
1975                                 ret = btrfs_add_free_space(cache, bytenr,
1976                                                            num_bytes);
1977                                 WARN_ON(ret);
1978                         }
1979                 }
1980                 put_block_group(cache);
1981                 total -= num_bytes;
1982                 bytenr += num_bytes;
1983         }
1984         return 0;
1985 }
1986
1987 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
1988 {
1989         struct btrfs_block_group_cache *cache;
1990         u64 bytenr;
1991
1992         cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
1993         if (!cache)
1994                 return 0;
1995
1996         bytenr = cache->key.objectid;
1997         put_block_group(cache);
1998
1999         return bytenr;
2000 }
2001
2002 int btrfs_update_pinned_extents(struct btrfs_root *root,
2003                                 u64 bytenr, u64 num, int pin)
2004 {
2005         u64 len;
2006         struct btrfs_block_group_cache *cache;
2007         struct btrfs_fs_info *fs_info = root->fs_info;
2008
2009         WARN_ON(!mutex_is_locked(&root->fs_info->pinned_mutex));
2010         if (pin) {
2011                 set_extent_dirty(&fs_info->pinned_extents,
2012                                 bytenr, bytenr + num - 1, GFP_NOFS);
2013         } else {
2014                 clear_extent_dirty(&fs_info->pinned_extents,
2015                                 bytenr, bytenr + num - 1, GFP_NOFS);
2016         }
2017         while (num > 0) {
2018                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2019                 BUG_ON(!cache);
2020                 len = min(num, cache->key.offset -
2021                           (bytenr - cache->key.objectid));
2022                 if (pin) {
2023                         spin_lock(&cache->space_info->lock);
2024                         spin_lock(&cache->lock);
2025                         cache->pinned += len;
2026                         cache->space_info->bytes_pinned += len;
2027                         spin_unlock(&cache->lock);
2028                         spin_unlock(&cache->space_info->lock);
2029                         fs_info->total_pinned += len;
2030                 } else {
2031                         spin_lock(&cache->space_info->lock);
2032                         spin_lock(&cache->lock);
2033                         cache->pinned -= len;
2034                         cache->space_info->bytes_pinned -= len;
2035                         spin_unlock(&cache->lock);
2036                         spin_unlock(&cache->space_info->lock);
2037                         fs_info->total_pinned -= len;
2038                         if (cache->cached)
2039                                 btrfs_add_free_space(cache, bytenr, len);
2040                 }
2041                 put_block_group(cache);
2042                 bytenr += len;
2043                 num -= len;
2044         }
2045         return 0;
2046 }
2047
2048 static int update_reserved_extents(struct btrfs_root *root,
2049                                    u64 bytenr, u64 num, int reserve)
2050 {
2051         u64 len;
2052         struct btrfs_block_group_cache *cache;
2053         struct btrfs_fs_info *fs_info = root->fs_info;
2054
2055         while (num > 0) {
2056                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2057                 BUG_ON(!cache);
2058                 len = min(num, cache->key.offset -
2059                           (bytenr - cache->key.objectid));
2060
2061                 spin_lock(&cache->space_info->lock);
2062                 spin_lock(&cache->lock);
2063                 if (reserve) {
2064                         cache->reserved += len;
2065                         cache->space_info->bytes_reserved += len;
2066                 } else {
2067                         cache->reserved -= len;
2068                         cache->space_info->bytes_reserved -= len;
2069                 }
2070                 spin_unlock(&cache->lock);
2071                 spin_unlock(&cache->space_info->lock);
2072                 put_block_group(cache);
2073                 bytenr += len;
2074                 num -= len;
2075         }
2076         return 0;
2077 }
2078
2079 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
2080 {
2081         u64 last = 0;
2082         u64 start;
2083         u64 end;
2084         struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
2085         int ret;
2086
2087         mutex_lock(&root->fs_info->pinned_mutex);
2088         while (1) {
2089                 ret = find_first_extent_bit(pinned_extents, last,
2090                                             &start, &end, EXTENT_DIRTY);
2091                 if (ret)
2092                         break;
2093                 set_extent_dirty(copy, start, end, GFP_NOFS);
2094                 last = end + 1;
2095         }
2096         mutex_unlock(&root->fs_info->pinned_mutex);
2097         return 0;
2098 }
2099
2100 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2101                                struct btrfs_root *root,
2102                                struct extent_io_tree *unpin)
2103 {
2104         u64 start;
2105         u64 end;
2106         int ret;
2107
2108         mutex_lock(&root->fs_info->pinned_mutex);
2109         while (1) {
2110                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2111                                             EXTENT_DIRTY);
2112                 if (ret)
2113                         break;
2114
2115                 ret = btrfs_discard_extent(root, start, end + 1 - start);
2116
2117                 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2118                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2119
2120                 if (need_resched()) {
2121                         mutex_unlock(&root->fs_info->pinned_mutex);
2122                         cond_resched();
2123                         mutex_lock(&root->fs_info->pinned_mutex);
2124                 }
2125         }
2126         mutex_unlock(&root->fs_info->pinned_mutex);
2127         return ret;
2128 }
2129
2130 static int finish_current_insert(struct btrfs_trans_handle *trans,
2131                                  struct btrfs_root *extent_root, int all)
2132 {
2133         u64 start;
2134         u64 end;
2135         u64 priv;
2136         u64 search = 0;
2137         u64 skipped = 0;
2138         struct btrfs_fs_info *info = extent_root->fs_info;
2139         struct btrfs_path *path;
2140         struct pending_extent_op *extent_op, *tmp;
2141         struct list_head insert_list, update_list;
2142         int ret;
2143         int num_inserts = 0, max_inserts;
2144
2145         path = btrfs_alloc_path();
2146         INIT_LIST_HEAD(&insert_list);
2147         INIT_LIST_HEAD(&update_list);
2148
2149         max_inserts = extent_root->leafsize /
2150                 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2151                  sizeof(struct btrfs_extent_ref) +
2152                  sizeof(struct btrfs_extent_item));
2153 again:
2154         mutex_lock(&info->extent_ins_mutex);
2155         while (1) {
2156                 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2157                                             &end, EXTENT_WRITEBACK);
2158                 if (ret) {
2159                         if (skipped && all && !num_inserts &&
2160                             list_empty(&update_list)) {
2161                                 skipped = 0;
2162                                 search = 0;
2163                                 continue;
2164                         }
2165                         mutex_unlock(&info->extent_ins_mutex);
2166                         break;
2167                 }
2168
2169                 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2170                 if (!ret) {
2171                         skipped = 1;
2172                         search = end + 1;
2173                         if (need_resched()) {
2174                                 mutex_unlock(&info->extent_ins_mutex);
2175                                 cond_resched();
2176                                 mutex_lock(&info->extent_ins_mutex);
2177                         }
2178                         continue;
2179                 }
2180
2181                 ret = get_state_private(&info->extent_ins, start, &priv);
2182                 BUG_ON(ret);
2183                 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2184
2185                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2186                         num_inserts++;
2187                         list_add_tail(&extent_op->list, &insert_list);
2188                         search = end + 1;
2189                         if (num_inserts == max_inserts) {
2190                                 mutex_unlock(&info->extent_ins_mutex);
2191                                 break;
2192                         }
2193                 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2194                         list_add_tail(&extent_op->list, &update_list);
2195                         search = end + 1;
2196                 } else {
2197                         BUG();
2198                 }
2199         }
2200
2201         /*
2202          * process the update list, clear the writeback bit for it, and if
2203          * somebody marked this thing for deletion then just unlock it and be
2204          * done, the free_extents will handle it
2205          */
2206         mutex_lock(&info->extent_ins_mutex);
2207         list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2208                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2209                                   extent_op->bytenr + extent_op->num_bytes - 1,
2210                                   EXTENT_WRITEBACK, GFP_NOFS);
2211                 if (extent_op->del) {
2212                         list_del_init(&extent_op->list);
2213                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2214                                       extent_op->bytenr + extent_op->num_bytes
2215                                       - 1, GFP_NOFS);
2216                         kfree(extent_op);
2217                 }
2218         }
2219         mutex_unlock(&info->extent_ins_mutex);
2220
2221         /*
2222          * still have things left on the update list, go ahead an update
2223          * everything
2224          */
2225         if (!list_empty(&update_list)) {
2226                 ret = update_backrefs(trans, extent_root, path, &update_list);
2227                 BUG_ON(ret);
2228         }
2229
2230         /*
2231          * if no inserts need to be done, but we skipped some extents and we
2232          * need to make sure everything is cleaned then reset everything and
2233          * go back to the beginning
2234          */
2235         if (!num_inserts && all && skipped) {
2236                 search = 0;
2237                 skipped = 0;
2238                 INIT_LIST_HEAD(&update_list);
2239                 INIT_LIST_HEAD(&insert_list);
2240                 goto again;
2241         } else if (!num_inserts) {
2242                 goto out;
2243         }
2244
2245         /*
2246          * process the insert extents list.  Again if we are deleting this
2247          * extent, then just unlock it, pin down the bytes if need be, and be
2248          * done with it.  Saves us from having to actually insert the extent
2249          * into the tree and then subsequently come along and delete it
2250          */
2251         mutex_lock(&info->extent_ins_mutex);
2252         list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2253                 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2254                                   extent_op->bytenr + extent_op->num_bytes - 1,
2255                                   EXTENT_WRITEBACK, GFP_NOFS);
2256                 if (extent_op->del) {
2257                         u64 used;
2258                         list_del_init(&extent_op->list);
2259                         unlock_extent(&info->extent_ins, extent_op->bytenr,
2260                                       extent_op->bytenr + extent_op->num_bytes
2261                                       - 1, GFP_NOFS);
2262
2263                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2264                         ret = pin_down_bytes(trans, extent_root,
2265                                              extent_op->bytenr,
2266                                              extent_op->num_bytes, 0);
2267                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2268
2269                         spin_lock(&info->delalloc_lock);
2270                         used = btrfs_super_bytes_used(&info->super_copy);
2271                         btrfs_set_super_bytes_used(&info->super_copy,
2272                                         used - extent_op->num_bytes);
2273                         used = btrfs_root_used(&extent_root->root_item);
2274                         btrfs_set_root_used(&extent_root->root_item,
2275                                         used - extent_op->num_bytes);
2276                         spin_unlock(&info->delalloc_lock);
2277
2278                         ret = update_block_group(trans, extent_root,
2279                                                  extent_op->bytenr,
2280                                                  extent_op->num_bytes,
2281                                                  0, ret > 0);
2282                         BUG_ON(ret);
2283                         kfree(extent_op);
2284                         num_inserts--;
2285                 }
2286         }
2287         mutex_unlock(&info->extent_ins_mutex);
2288
2289         ret = insert_extents(trans, extent_root, path, &insert_list,
2290                              num_inserts);
2291         BUG_ON(ret);
2292
2293         /*
2294          * if we broke out of the loop in order to insert stuff because we hit
2295          * the maximum number of inserts at a time we can handle, then loop
2296          * back and pick up where we left off
2297          */
2298         if (num_inserts == max_inserts) {
2299                 INIT_LIST_HEAD(&insert_list);
2300                 INIT_LIST_HEAD(&update_list);
2301                 num_inserts = 0;
2302                 goto again;
2303         }
2304
2305         /*
2306          * again, if we need to make absolutely sure there are no more pending
2307          * extent operations left and we know that we skipped some, go back to
2308          * the beginning and do it all again
2309          */
2310         if (all && skipped) {
2311                 INIT_LIST_HEAD(&insert_list);
2312                 INIT_LIST_HEAD(&update_list);
2313                 search = 0;
2314                 skipped = 0;
2315                 num_inserts = 0;
2316                 goto again;
2317         }
2318 out:
2319         btrfs_free_path(path);
2320         return 0;
2321 }
2322
2323 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2324                           struct btrfs_root *root,
2325                           u64 bytenr, u64 num_bytes, int is_data)
2326 {
2327         int err = 0;
2328         struct extent_buffer *buf;
2329
2330         if (is_data)
2331                 goto pinit;
2332
2333         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2334         if (!buf)
2335                 goto pinit;
2336
2337         /* we can reuse a block if it hasn't been written
2338          * and it is from this transaction.  We can't
2339          * reuse anything from the tree log root because
2340          * it has tiny sub-transactions.
2341          */
2342         if (btrfs_buffer_uptodate(buf, 0) &&
2343             btrfs_try_tree_lock(buf)) {
2344                 u64 header_owner = btrfs_header_owner(buf);
2345                 u64 header_transid = btrfs_header_generation(buf);
2346                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2347                     header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2348                     header_transid == trans->transid &&
2349                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2350                         clean_tree_block(NULL, root, buf);
2351                         btrfs_tree_unlock(buf);
2352                         free_extent_buffer(buf);
2353                         return 1;
2354                 }
2355                 btrfs_tree_unlock(buf);
2356         }
2357         free_extent_buffer(buf);
2358 pinit:
2359         btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2360
2361         BUG_ON(err < 0);
2362         return 0;
2363 }
2364
2365 /*
2366  * remove an extent from the root, returns 0 on success
2367  */
2368 static int __free_extent(struct btrfs_trans_handle *trans,
2369                          struct btrfs_root *root,
2370                          u64 bytenr, u64 num_bytes, u64 parent,
2371                          u64 root_objectid, u64 ref_generation,
2372                          u64 owner_objectid, int pin, int mark_free)
2373 {
2374         struct btrfs_path *path;
2375         struct btrfs_key key;
2376         struct btrfs_fs_info *info = root->fs_info;
2377         struct btrfs_root *extent_root = info->extent_root;
2378         struct extent_buffer *leaf;
2379         int ret;
2380         int extent_slot = 0;
2381         int found_extent = 0;
2382         int num_to_del = 1;
2383         struct btrfs_extent_item *ei;
2384         u32 refs;
2385
2386         key.objectid = bytenr;
2387         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
2388         key.offset = num_bytes;
2389         path = btrfs_alloc_path();
2390         if (!path)
2391                 return -ENOMEM;
2392
2393         path->reada = 1;
2394         ret = lookup_extent_backref(trans, extent_root, path,
2395                                     bytenr, parent, root_objectid,
2396                                     ref_generation, owner_objectid, 1);
2397         if (ret == 0) {
2398                 struct btrfs_key found_key;
2399                 extent_slot = path->slots[0];
2400                 while (extent_slot > 0) {
2401                         extent_slot--;
2402                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2403                                               extent_slot);
2404                         if (found_key.objectid != bytenr)
2405                                 break;
2406                         if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
2407                             found_key.offset == num_bytes) {
2408                                 found_extent = 1;
2409                                 break;
2410                         }
2411                         if (path->slots[0] - extent_slot > 5)
2412                                 break;
2413                 }
2414                 if (!found_extent) {
2415                         ret = remove_extent_backref(trans, extent_root, path);
2416                         BUG_ON(ret);
2417                         btrfs_release_path(extent_root, path);
2418                         ret = btrfs_search_slot(trans, extent_root,
2419                                                 &key, path, -1, 1);
2420                         if (ret) {
2421                                 printk(KERN_ERR "umm, got %d back from search"
2422                                        ", was looking for %llu\n", ret,
2423                                        (unsigned long long)bytenr);
2424                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2425                         }
2426                         BUG_ON(ret);
2427                         extent_slot = path->slots[0];
2428                 }
2429         } else {
2430                 btrfs_print_leaf(extent_root, path->nodes[0]);
2431                 WARN_ON(1);
2432                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2433                        "root %llu gen %llu owner %llu\n",
2434                        (unsigned long long)bytenr,
2435                        (unsigned long long)root_objectid,
2436                        (unsigned long long)ref_generation,
2437                        (unsigned long long)owner_objectid);
2438         }
2439
2440         leaf = path->nodes[0];
2441         ei = btrfs_item_ptr(leaf, extent_slot,
2442                             struct btrfs_extent_item);
2443         refs = btrfs_extent_refs(leaf, ei);
2444         BUG_ON(refs == 0);
2445         refs -= 1;
2446         btrfs_set_extent_refs(leaf, ei, refs);
2447
2448         btrfs_mark_buffer_dirty(leaf);
2449
2450         if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
2451                 struct btrfs_extent_ref *ref;
2452                 ref = btrfs_item_ptr(leaf, path->slots[0],
2453                                      struct btrfs_extent_ref);
2454                 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1);
2455                 /* if the back ref and the extent are next to each other
2456                  * they get deleted below in one shot
2457                  */
2458                 path->slots[0] = extent_slot;
2459                 num_to_del = 2;
2460         } else if (found_extent) {
2461                 /* otherwise delete the extent back ref */
2462                 ret = remove_extent_backref(trans, extent_root, path);
2463                 BUG_ON(ret);
2464                 /* if refs are 0, we need to setup the path for deletion */
2465                 if (refs == 0) {
2466                         btrfs_release_path(extent_root, path);
2467                         ret = btrfs_search_slot(trans, extent_root, &key, path,
2468                                                 -1, 1);
2469                         BUG_ON(ret);
2470                 }
2471         }
2472
2473         if (refs == 0) {
2474                 u64 super_used;
2475                 u64 root_used;
2476
2477                 if (pin) {
2478                         mutex_lock(&root->fs_info->pinned_mutex);
2479                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2480                                 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID);
2481                         mutex_unlock(&root->fs_info->pinned_mutex);
2482                         if (ret > 0)
2483                                 mark_free = 1;
2484                         BUG_ON(ret < 0);
2485                 }
2486                 /* block accounting for super block */
2487                 spin_lock(&info->delalloc_lock);
2488                 super_used = btrfs_super_bytes_used(&info->super_copy);
2489                 btrfs_set_super_bytes_used(&info->super_copy,
2490                                            super_used - num_bytes);
2491
2492                 /* block accounting for root item */
2493                 root_used = btrfs_root_used(&root->root_item);
2494                 btrfs_set_root_used(&root->root_item,
2495                                            root_used - num_bytes);
2496                 spin_unlock(&info->delalloc_lock);
2497                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2498                                       num_to_del);
2499                 BUG_ON(ret);
2500                 btrfs_release_path(extent_root, path);
2501
2502                 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2503                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2504                         BUG_ON(ret);
2505                 }
2506
2507                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2508                                          mark_free);
2509                 BUG_ON(ret);
2510         }
2511         btrfs_free_path(path);
2512         finish_current_insert(trans, extent_root, 0);
2513         return ret;
2514 }
2515
2516 /*
2517  * find all the blocks marked as pending in the radix tree and remove
2518  * them from the extent map
2519  */
2520 static int del_pending_extents(struct btrfs_trans_handle *trans,
2521                                struct btrfs_root *extent_root, int all)
2522 {
2523         int ret;
2524         int err = 0;
2525         u64 start;
2526         u64 end;
2527         u64 priv;
2528         u64 search = 0;
2529         int nr = 0, skipped = 0;
2530         struct extent_io_tree *pending_del;
2531         struct extent_io_tree *extent_ins;
2532         struct pending_extent_op *extent_op;
2533         struct btrfs_fs_info *info = extent_root->fs_info;
2534         struct list_head delete_list;
2535
2536         INIT_LIST_HEAD(&delete_list);
2537         extent_ins = &extent_root->fs_info->extent_ins;
2538         pending_del = &extent_root->fs_info->pending_del;
2539
2540 again:
2541         mutex_lock(&info->extent_ins_mutex);
2542         while (1) {
2543                 ret = find_first_extent_bit(pending_del, search, &start, &end,
2544                                             EXTENT_WRITEBACK);
2545                 if (ret) {
2546                         if (all && skipped && !nr) {
2547                                 search = 0;
2548                                 skipped = 0;
2549                                 continue;
2550                         }
2551                         mutex_unlock(&info->extent_ins_mutex);
2552                         break;
2553                 }
2554
2555                 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2556                 if (!ret) {
2557                         search = end+1;
2558                         skipped = 1;
2559
2560                         if (need_resched()) {
2561                                 mutex_unlock(&info->extent_ins_mutex);
2562                                 cond_resched();
2563                                 mutex_lock(&info->extent_ins_mutex);
2564                         }
2565
2566                         continue;
2567                 }
2568                 BUG_ON(ret < 0);
2569
2570                 ret = get_state_private(pending_del, start, &priv);
2571                 BUG_ON(ret);
2572                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2573
2574                 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2575                                   GFP_NOFS);
2576                 if (!test_range_bit(extent_ins, start, end,
2577                                     EXTENT_WRITEBACK, 0)) {
2578                         list_add_tail(&extent_op->list, &delete_list);
2579                         nr++;
2580                 } else {
2581                         kfree(extent_op);
2582
2583                         ret = get_state_private(&info->extent_ins, start,
2584                                                 &priv);
2585                         BUG_ON(ret);
2586                         extent_op = (struct pending_extent_op *)
2587                                                 (unsigned long)priv;
2588
2589                         clear_extent_bits(&info->extent_ins, start, end,
2590                                           EXTENT_WRITEBACK, GFP_NOFS);
2591
2592                         if (extent_op->type == PENDING_BACKREF_UPDATE) {
2593                                 list_add_tail(&extent_op->list, &delete_list);
2594                                 search = end + 1;
2595                                 nr++;
2596                                 continue;
2597                         }
2598
2599                         mutex_lock(&extent_root->fs_info->pinned_mutex);
2600                         ret = pin_down_bytes(trans, extent_root, start,
2601                                              end + 1 - start, 0);
2602                         mutex_unlock(&extent_root->fs_info->pinned_mutex);
2603
2604                         ret = update_block_group(trans, extent_root, start,
2605                                                 end + 1 - start, 0, ret > 0);
2606
2607                         unlock_extent(extent_ins, start, end, GFP_NOFS);
2608                         BUG_ON(ret);
2609                         kfree(extent_op);
2610                 }
2611                 if (ret)
2612                         err = ret;
2613
2614                 search = end + 1;
2615
2616                 if (need_resched()) {
2617                         mutex_unlock(&info->extent_ins_mutex);
2618                         cond_resched();
2619                         mutex_lock(&info->extent_ins_mutex);
2620                 }
2621         }
2622
2623         if (nr) {
2624                 ret = free_extents(trans, extent_root, &delete_list);
2625                 BUG_ON(ret);
2626         }
2627
2628         if (all && skipped) {
2629                 INIT_LIST_HEAD(&delete_list);
2630                 search = 0;
2631                 nr = 0;
2632                 goto again;
2633         }
2634
2635         return err;
2636 }
2637
2638 /*
2639  * remove an extent from the root, returns 0 on success
2640  */
2641 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2642                                struct btrfs_root *root,
2643                                u64 bytenr, u64 num_bytes, u64 parent,
2644                                u64 root_objectid, u64 ref_generation,
2645                                u64 owner_objectid, int pin)
2646 {
2647         struct btrfs_root *extent_root = root->fs_info->extent_root;
2648         int pending_ret;
2649         int ret;
2650
2651         WARN_ON(num_bytes < root->sectorsize);
2652         if (root == extent_root) {
2653                 struct pending_extent_op *extent_op = NULL;
2654
2655                 mutex_lock(&root->fs_info->extent_ins_mutex);
2656                 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
2657                                 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
2658                         u64 priv;
2659                         ret = get_state_private(&root->fs_info->extent_ins,
2660                                                 bytenr, &priv);
2661                         BUG_ON(ret);
2662                         extent_op = (struct pending_extent_op *)
2663                                                 (unsigned long)priv;
2664
2665                         extent_op->del = 1;
2666                         if (extent_op->type == PENDING_EXTENT_INSERT) {
2667                                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2668                                 return 0;
2669                         }
2670                 }
2671
2672                 if (extent_op) {
2673                         ref_generation = extent_op->orig_generation;
2674                         parent = extent_op->orig_parent;
2675                 }
2676
2677                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2678                 BUG_ON(!extent_op);
2679
2680                 extent_op->type = PENDING_EXTENT_DELETE;
2681                 extent_op->bytenr = bytenr;
2682                 extent_op->num_bytes = num_bytes;
2683                 extent_op->parent = parent;
2684                 extent_op->orig_parent = parent;
2685                 extent_op->generation = ref_generation;
2686                 extent_op->orig_generation = ref_generation;
2687                 extent_op->level = (int)owner_objectid;
2688                 INIT_LIST_HEAD(&extent_op->list);
2689                 extent_op->del = 0;
2690
2691                 set_extent_bits(&root->fs_info->pending_del,
2692                                 bytenr, bytenr + num_bytes - 1,
2693                                 EXTENT_WRITEBACK, GFP_NOFS);
2694                 set_state_private(&root->fs_info->pending_del,
2695                                   bytenr, (unsigned long)extent_op);
2696                 mutex_unlock(&root->fs_info->extent_ins_mutex);
2697                 return 0;
2698         }
2699         /* if metadata always pin */
2700         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2701                 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
2702                         struct btrfs_block_group_cache *cache;
2703
2704                         /* btrfs_free_reserved_extent */
2705                         cache = btrfs_lookup_block_group(root->fs_info, bytenr);
2706                         BUG_ON(!cache);
2707                         btrfs_add_free_space(cache, bytenr, num_bytes);
2708                         put_block_group(cache);
2709                         update_reserved_extents(root, bytenr, num_bytes, 0);
2710                         return 0;
2711                 }
2712                 pin = 1;
2713         }
2714
2715         /* if data pin when any transaction has committed this */
2716         if (ref_generation != trans->transid)
2717                 pin = 1;
2718
2719         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2720                             root_objectid, ref_generation,
2721                             owner_objectid, pin, pin == 0);
2722
2723         finish_current_insert(trans, root->fs_info->extent_root, 0);
2724         pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0);
2725         return ret ? ret : pending_ret;
2726 }
2727
2728 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2729                       struct btrfs_root *root,
2730                       u64 bytenr, u64 num_bytes, u64 parent,
2731                       u64 root_objectid, u64 ref_generation,
2732                       u64 owner_objectid, int pin)
2733 {
2734         int ret;
2735
2736         ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent,
2737                                   root_objectid, ref_generation,
2738                                   owner_objectid, pin);
2739         return ret;
2740 }
2741
2742 static u64 stripe_align(struct btrfs_root *root, u64 val)
2743 {
2744         u64 mask = ((u64)root->stripesize - 1);
2745         u64 ret = (val + mask) & ~mask;
2746         return ret;
2747 }
2748
2749 /*
2750  * walks the btree of allocated extents and find a hole of a given size.
2751  * The key ins is changed to record the hole:
2752  * ins->objectid == block start
2753  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2754  * ins->offset == number of blocks
2755  * Any available blocks before search_start are skipped.
2756  */
2757 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
2758                                      struct btrfs_root *orig_root,
2759                                      u64 num_bytes, u64 empty_size,
2760                                      u64 search_start, u64 search_end,
2761                                      u64 hint_byte, struct btrfs_key *ins,
2762                                      u64 exclude_start, u64 exclude_nr,
2763                                      int data)
2764 {
2765         int ret = 0;
2766         struct btrfs_root *root = orig_root->fs_info->extent_root;
2767         u64 total_needed = num_bytes;
2768         u64 *last_ptr = NULL;
2769         u64 last_wanted = 0;
2770         struct btrfs_block_group_cache *block_group = NULL;
2771         int chunk_alloc_done = 0;
2772         int empty_cluster = 2 * 1024 * 1024;
2773         int allowed_chunk_alloc = 0;
2774         struct list_head *head = NULL, *cur = NULL;
2775         int loop = 0;
2776         int extra_loop = 0;
2777         struct btrfs_space_info *space_info;
2778
2779         WARN_ON(num_bytes < root->sectorsize);
2780         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2781         ins->objectid = 0;
2782         ins->offset = 0;
2783
2784         if (orig_root->ref_cows || empty_size)
2785                 allowed_chunk_alloc = 1;
2786
2787         if (data & BTRFS_BLOCK_GROUP_METADATA) {
2788                 last_ptr = &root->fs_info->last_alloc;
2789                 empty_cluster = 64 * 1024;
2790         }
2791
2792         if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD))
2793                 last_ptr = &root->fs_info->last_data_alloc;
2794
2795         if (last_ptr) {
2796                 if (*last_ptr) {
2797                         hint_byte = *last_ptr;
2798                         last_wanted = *last_ptr;
2799                 } else
2800                         empty_size += empty_cluster;
2801         } else {
2802                 empty_cluster = 0;
2803         }
2804         search_start = max(search_start, first_logical_byte(root, 0));
2805         search_start = max(search_start, hint_byte);
2806
2807         if (last_wanted && search_start != last_wanted) {
2808                 last_wanted = 0;
2809                 empty_size += empty_cluster;
2810         }
2811
2812         total_needed += empty_size;
2813         block_group = btrfs_lookup_block_group(root->fs_info, search_start);
2814         if (!block_group)
2815                 block_group = btrfs_lookup_first_block_group(root->fs_info,
2816                                                              search_start);
2817         space_info = __find_space_info(root->fs_info, data);
2818
2819         down_read(&space_info->groups_sem);
2820         while (1) {
2821                 struct btrfs_free_space *free_space;
2822                 /*
2823                  * the only way this happens if our hint points to a block
2824                  * group thats not of the proper type, while looping this
2825                  * should never happen
2826                  */
2827                 if (empty_size)
2828                         extra_loop = 1;
2829
2830                 if (!block_group)
2831                         goto new_group_no_lock;
2832
2833                 if (unlikely(!block_group->cached)) {
2834                         mutex_lock(&block_group->cache_mutex);
2835                         ret = cache_block_group(root, block_group);
2836                         mutex_unlock(&block_group->cache_mutex);
2837                         if (ret)
2838                                 break;
2839                 }
2840
2841                 mutex_lock(&block_group->alloc_mutex);
2842                 if (unlikely(!block_group_bits(block_group, data)))
2843                         goto new_group;
2844
2845                 if (unlikely(block_group->ro))
2846                         goto new_group;
2847
2848                 free_space = btrfs_find_free_space(block_group, search_start,
2849                                                    total_needed);
2850                 if (free_space) {
2851                         u64 start = block_group->key.objectid;
2852                         u64 end = block_group->key.objectid +
2853                                 block_group->key.offset;
2854
2855                         search_start = stripe_align(root, free_space->offset);
2856
2857                         /* move on to the next group */
2858                         if (search_start + num_bytes >= search_end)
2859                                 goto new_group;
2860
2861                         /* move on to the next group */
2862                         if (search_start + num_bytes > end)
2863                                 goto new_group;
2864
2865                         if (last_wanted && search_start != last_wanted) {
2866                                 total_needed += empty_cluster;
2867                                 empty_size += empty_cluster;
2868                                 last_wanted = 0;
2869                                 /*
2870                                  * if search_start is still in this block group
2871                                  * then we just re-search this block group
2872                                  */
2873                                 if (search_start >= start &&
2874                                     search_start < end) {
2875                                         mutex_unlock(&block_group->alloc_mutex);
2876                                         continue;
2877                                 }
2878
2879                                 /* else we go to the next block group */
2880                                 goto new_group;
2881                         }
2882
2883                         if (exclude_nr > 0 &&
2884                             (search_start + num_bytes > exclude_start &&
2885                              search_start < exclude_start + exclude_nr)) {
2886                                 search_start = exclude_start + exclude_nr;
2887                                 /*
2888                                  * if search_start is still in this block group
2889                                  * then we just re-search this block group
2890                                  */
2891                                 if (search_start >= start &&
2892                                     search_start < end) {
2893                                         mutex_unlock(&block_group->alloc_mutex);
2894                                         last_wanted = 0;
2895                                         continue;
2896                                 }
2897
2898                                 /* else we go to the next block group */
2899                                 goto new_group;
2900                         }
2901
2902                         ins->objectid = search_start;
2903                         ins->offset = num_bytes;
2904
2905                         btrfs_remove_free_space_lock(block_group, search_start,
2906                                                      num_bytes);
2907                         /* we are all good, lets return */
2908                         mutex_unlock(&block_group->alloc_mutex);
2909                         break;
2910                 }
2911 new_group:
2912                 mutex_unlock(&block_group->alloc_mutex);
2913                 put_block_group(block_group);
2914                 block_group = NULL;
2915 new_group_no_lock:
2916                 /* don't try to compare new allocations against the
2917                  * last allocation any more
2918                  */
2919                 last_wanted = 0;
2920
2921                 /*
2922                  * Here's how this works.
2923                  * loop == 0: we were searching a block group via a hint
2924                  *              and didn't find anything, so we start at
2925                  *              the head of the block groups and keep searching
2926                  * loop == 1: we're searching through all of the block groups
2927                  *              if we hit the head again we have searched
2928                  *              all of the block groups for this space and we
2929                  *              need to try and allocate, if we cant error out.
2930                  * loop == 2: we allocated more space and are looping through
2931                  *              all of the block groups again.
2932                  */
2933                 if (loop == 0) {
2934                         head = &space_info->block_groups;
2935                         cur = head->next;
2936                         loop++;
2937                 } else if (loop == 1 && cur == head) {
2938                         int keep_going;
2939
2940                         /* at this point we give up on the empty_size
2941                          * allocations and just try to allocate the min
2942                          * space.
2943                          *
2944                          * The extra_loop field was set if an empty_size
2945                          * allocation was attempted above, and if this
2946                          * is try we need to try the loop again without
2947                          * the additional empty_size.
2948                          */
2949                         total_needed -= empty_size;
2950                         empty_size = 0;
2951                         keep_going = extra_loop;
2952                         loop++;
2953
2954                         if (allowed_chunk_alloc && !chunk_alloc_done) {
2955                                 up_read(&space_info->groups_sem);
2956                                 ret = do_chunk_alloc(trans, root, num_bytes +
2957                                                      2 * 1024 * 1024, data, 1);
2958                                 down_read(&space_info->groups_sem);
2959                                 if (ret < 0)
2960                                         goto loop_check;
2961                                 head = &space_info->block_groups;
2962                                 /*
2963                                  * we've allocated a new chunk, keep
2964                                  * trying
2965                                  */
2966                                 keep_going = 1;
2967                                 chunk_alloc_done = 1;
2968                         } else if (!allowed_chunk_alloc) {
2969                                 space_info->force_alloc = 1;
2970                         }
2971 loop_check:
2972                         if (keep_going) {
2973                                 cur = head->next;
2974                                 extra_loop = 0;
2975                         } else {
2976                                 break;
2977                         }
2978                 } else if (cur == head) {
2979                         break;
2980                 }
2981
2982                 block_group = list_entry(cur, struct btrfs_block_group_cache,
2983                                          list);
2984                 atomic_inc(&block_group->count);
2985
2986                 search_start = block_group->key.objectid;
2987                 cur = cur->next;
2988         }
2989
2990         /* we found what we needed */
2991         if (ins->objectid) {
2992                 if (!(data & BTRFS_BLOCK_GROUP_DATA))
2993                         trans->block_group = block_group->key.objectid;
2994
2995                 if (last_ptr)
2996                         *last_ptr = ins->objectid + ins->offset;
2997                 ret = 0;
2998         } else if (!ret) {
2999                 printk(KERN_ERR "btrfs searching for %llu bytes, "
3000                        "num_bytes %llu, loop %d, allowed_alloc %d\n",
3001                        (unsigned long long)total_needed,
3002                        (unsigned long long)num_bytes,
3003                        loop, allowed_chunk_alloc);
3004                 ret = -ENOSPC;
3005         }
3006         if (block_group)
3007                 put_block_group(block_group);
3008
3009         up_read(&space_info->groups_sem);
3010         return ret;
3011 }
3012
3013 static void dump_space_info(struct btrfs_space_info *info, u64 bytes)
3014 {
3015         struct btrfs_block_group_cache *cache;
3016
3017         printk(KERN_INFO "space_info has %llu free, is %sfull\n",
3018                (unsigned long long)(info->total_bytes - info->bytes_used -
3019                                     info->bytes_pinned - info->bytes_reserved),
3020                (info->full) ? "" : "not ");
3021
3022         down_read(&info->groups_sem);
3023         list_for_each_entry(cache, &info->block_groups, list) {
3024                 spin_lock(&cache->lock);
3025                 printk(KERN_INFO "block group %llu has %llu bytes, %llu used "
3026                        "%llu pinned %llu reserved\n",
3027                        (unsigned long long)cache->key.objectid,
3028                        (unsigned long long)cache->key.offset,
3029                        (unsigned long long)btrfs_block_group_used(&cache->item),
3030                        (unsigned long long)cache->pinned,
3031                        (unsigned long long)cache->reserved);
3032                 btrfs_dump_free_space(cache, bytes);
3033                 spin_unlock(&cache->lock);
3034         }
3035         up_read(&info->groups_sem);
3036 }
3037
3038 static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3039                                   struct btrfs_root *root,
3040                                   u64 num_bytes, u64 min_alloc_size,
3041                                   u64 empty_size, u64 hint_byte,
3042                                   u64 search_end, struct btrfs_key *ins,
3043                                   u64 data)
3044 {
3045         int ret;
3046         u64 search_start = 0;
3047         u64 alloc_profile;
3048         struct btrfs_fs_info *info = root->fs_info;
3049
3050         if (data) {
3051                 alloc_profile = info->avail_data_alloc_bits &
3052                         info->data_alloc_profile;
3053                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
3054         } else if (root == root->fs_info->chunk_root) {
3055                 alloc_profile = info->avail_system_alloc_bits &
3056                         info->system_alloc_profile;
3057                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
3058         } else {
3059                 alloc_profile = info->avail_metadata_alloc_bits &
3060                         info->metadata_alloc_profile;
3061                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
3062         }
3063 again:
3064         data = btrfs_reduce_alloc_profile(root, data);
3065         /*
3066          * the only place that sets empty_size is btrfs_realloc_node, which
3067          * is not called recursively on allocations
3068          */
3069         if (empty_size || root->ref_cows) {
3070                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
3071                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3072                                      2 * 1024 * 1024,
3073                                      BTRFS_BLOCK_GROUP_METADATA |
3074                                      (info->metadata_alloc_profile &
3075                                       info->avail_metadata_alloc_bits), 0);
3076                 }
3077                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3078                                      num_bytes + 2 * 1024 * 1024, data, 0);
3079         }
3080
3081         WARN_ON(num_bytes < root->sectorsize);
3082         ret = find_free_extent(trans, root, num_bytes, empty_size,
3083                                search_start, search_end, hint_byte, ins,
3084                                trans->alloc_exclude_start,
3085                                trans->alloc_exclude_nr, data);
3086
3087         if (ret == -ENOSPC && num_bytes > min_alloc_size) {
3088                 num_bytes = num_bytes >> 1;
3089                 num_bytes = num_bytes & ~(root->sectorsize - 1);
3090                 num_bytes = max(num_bytes, min_alloc_size);
3091                 do_chunk_alloc(trans, root->fs_info->extent_root,
3092                                num_bytes, data, 1);
3093                 goto again;
3094         }
3095         if (ret) {
3096                 struct btrfs_space_info *sinfo;
3097
3098                 sinfo = __find_space_info(root->fs_info, data);
3099                 printk(KERN_ERR "btrfs allocation failed flags %llu, "
3100                        "wanted %llu\n", (unsigned long long)data,
3101                        (unsigned long long)num_bytes);
3102                 dump_space_info(sinfo, num_bytes);
3103                 BUG();
3104         }
3105
3106         return ret;
3107 }
3108
3109 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len)
3110 {
3111         struct btrfs_block_group_cache *cache;
3112         int ret = 0;
3113
3114         cache = btrfs_lookup_block_group(root->fs_info, start);
3115         if (!cache) {
3116                 printk(KERN_ERR "Unable to find block group for %llu\n",
3117                        (unsigned long long)start);
3118                 return -ENOSPC;
3119         }
3120
3121         ret = btrfs_discard_extent(root, start, len);
3122
3123         btrfs_add_free_space(cache, start, len);
3124         put_block_group(cache);
3125         update_reserved_extents(root, start, len, 0);
3126
3127         return ret;
3128 }
3129
3130 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3131                                   struct btrfs_root *root,
3132                                   u64 num_bytes, u64 min_alloc_size,
3133                                   u64 empty_size, u64 hint_byte,
3134                                   u64 search_end, struct btrfs_key *ins,
3135                                   u64 data)
3136 {
3137         int ret;
3138         ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
3139                                      empty_size, hint_byte, search_end, ins,
3140                                      data);
3141         update_reserved_extents(root, ins->objectid, ins->offset, 1);
3142         return ret;
3143 }
3144
3145 static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3146                                          struct btrfs_root *root, u64 parent,
3147                                          u64 root_objectid, u64 ref_generation,
3148                                          u64 owner, struct btrfs_key *ins)
3149 {
3150         int ret;
3151         int pending_ret;
3152         u64 super_used;
3153         u64 root_used;
3154         u64 num_bytes = ins->offset;
3155         u32 sizes[2];
3156         struct btrfs_fs_info *info = root->fs_info;
3157         struct btrfs_root *extent_root = info->extent_root;
3158         struct btrfs_extent_item *extent_item;
3159         struct btrfs_extent_ref *ref;
3160         struct btrfs_path *path;
3161         struct btrfs_key keys[2];
3162
3163         if (parent == 0)
3164                 parent = ins->objectid;
3165
3166         /* block accounting for super block */
3167         spin_lock(&info->delalloc_lock);
3168         super_used = btrfs_super_bytes_used(&info->super_copy);
3169         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
3170
3171         /* block accounting for root item */
3172         root_used = btrfs_root_used(&root->root_item);
3173         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3174         spin_unlock(&info->delalloc_lock);
3175
3176         if (root == extent_root) {
3177                 struct pending_extent_op *extent_op;
3178
3179                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3180                 BUG_ON(!extent_op);
3181
3182                 extent_op->type = PENDING_EXTENT_INSERT;
3183                 extent_op->bytenr = ins->objectid;
3184                 extent_op->num_bytes = ins->offset;
3185                 extent_op->parent = parent;
3186                 extent_op->orig_parent = 0;
3187                 extent_op->generation = ref_generation;
3188                 extent_op->orig_generation = 0;
3189                 extent_op->level = (int)owner;
3190                 INIT_LIST_HEAD(&extent_op->list);
3191                 extent_op->del = 0;
3192
3193                 mutex_lock(&root->fs_info->extent_ins_mutex);
3194                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3195                                 ins->objectid + ins->offset - 1,
3196                                 EXTENT_WRITEBACK, GFP_NOFS);
3197                 set_state_private(&root->fs_info->extent_ins,
3198                                   ins->objectid, (unsigned long)extent_op);
3199                 mutex_unlock(&root->fs_info->extent_ins_mutex);
3200                 goto update_block;
3201         }
3202
3203         memcpy(&keys[0], ins, sizeof(*ins));
3204         keys[1].objectid = ins->objectid;
3205         keys[1].type = BTRFS_EXTENT_REF_KEY;
3206         keys[1].offset = parent;
3207         sizes[0] = sizeof(*extent_item);
3208         sizes[1] = sizeof(*ref);
3209
3210         path = btrfs_alloc_path();
3211         BUG_ON(!path);
3212
3213         ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3214                                        sizes, 2);
3215         BUG_ON(ret);
3216
3217         extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3218                                      struct btrfs_extent_item);
3219         btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
3220         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3221                              struct btrfs_extent_ref);
3222
3223         btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3224         btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3225         btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3226         btrfs_set_ref_num_refs(path->nodes[0], ref, 1);
3227
3228         btrfs_mark_buffer_dirty(path->nodes[0]);
3229
3230         trans->alloc_exclude_start = 0;
3231         trans->alloc_exclude_nr = 0;
3232         btrfs_free_path(path);
3233         finish_current_insert(trans, extent_root, 0);
3234         pending_ret = del_pending_extents(trans, extent_root, 0);
3235
3236         if (ret)
3237                 goto out;
3238         if (pending_ret) {
3239                 ret = pending_ret;
3240                 goto out;
3241         }
3242
3243 update_block:
3244         ret = update_block_group(trans, root, ins->objectid,
3245                                  ins->offset, 1, 0);
3246         if (ret) {
3247                 printk(KERN_ERR "btrfs update block group failed for %llu "
3248                        "%llu\n", (unsigned long long)ins->objectid,
3249                        (unsigned long long)ins->offset);
3250                 BUG();
3251         }
3252 out:
3253         return ret;
3254 }
3255
3256 int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3257                                 struct btrfs_root *root, u64 parent,
3258                                 u64 root_objectid, u64 ref_generation,
3259                                 u64 owner, struct btrfs_key *ins)
3260 {
3261         int ret;
3262
3263         if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3264                 return 0;
3265         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3266                                             ref_generation, owner, ins);
3267         update_reserved_extents(root, ins->objectid, ins->offset, 0);
3268         return ret;
3269 }
3270
3271 /*
3272  * this is used by the tree logging recovery code.  It records that
3273  * an extent has been allocated and makes sure to clear the free
3274  * space cache bits as well
3275  */
3276 int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3277                                 struct btrfs_root *root, u64 parent,
3278                                 u64 root_objectid, u64 ref_generation,
3279                                 u64 owner, struct btrfs_key *ins)
3280 {
3281         int ret;
3282         struct btrfs_block_group_cache *block_group;
3283
3284         block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
3285         mutex_lock(&block_group->cache_mutex);
3286         cache_block_group(root, block_group);
3287         mutex_unlock(&block_group->cache_mutex);
3288
3289         ret = btrfs_remove_free_space(block_group, ins->objectid,
3290                                       ins->offset);
3291         BUG_ON(ret);
3292         put_block_group(block_group);
3293         ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3294                                             ref_generation, owner, ins);
3295         return ret;
3296 }
3297
3298 /*
3299  * finds a free extent and does all the dirty work required for allocation
3300  * returns the key for the extent through ins, and a tree buffer for
3301  * the first block of the extent through buf.
3302  *
3303  * returns 0 if everything worked, non-zero otherwise.
3304  */
3305 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3306                        struct btrfs_root *root,
3307                        u64 num_bytes, u64 parent, u64 min_alloc_size,
3308                        u64 root_objectid, u64 ref_generation,
3309                        u64 owner_objectid, u64 empty_size, u64 hint_byte,
3310                        u64 search_end, struct btrfs_key *ins, u64 data)
3311 {
3312         int ret;
3313
3314         ret = __btrfs_reserve_extent(trans, root, num_bytes,
3315                                      min_alloc_size, empty_size, hint_byte,
3316                                      search_end, ins, data);
3317         BUG_ON(ret);
3318         if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3319                 ret = __btrfs_alloc_reserved_extent(trans, root, parent,
3320                                         root_objectid, ref_generation,
3321                                         owner_objectid, ins);
3322                 BUG_ON(ret);
3323
3324         } else {
3325                 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3326         }
3327         return ret;
3328 }
3329
3330 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
3331                                             struct btrfs_root *root,
3332                                             u64 bytenr, u32 blocksize)
3333 {
3334         struct extent_buffer *buf;
3335
3336         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
3337         if (!buf)
3338                 return ERR_PTR(-ENOMEM);
3339         btrfs_set_header_generation(buf, trans->transid);
3340         btrfs_tree_lock(buf);
3341         clean_tree_block(trans, root, buf);
3342         btrfs_set_buffer_uptodate(buf);
3343         if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3344                 set_extent_dirty(&root->dirty_log_pages, buf->start,
3345                          buf->start + buf->len - 1, GFP_NOFS);
3346         } else {
3347                 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
3348                          buf->start + buf->len - 1, GFP_NOFS);
3349         }
3350         trans->blocks_used++;
3351         return buf;
3352 }
3353
3354 /*
3355  * helper function to allocate a block for a given tree
3356  * returns the tree buffer or NULL.
3357  */
3358 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
3359                                              struct btrfs_root *root,
3360                                              u32 blocksize, u64 parent,
3361                                              u64 root_objectid,
3362                                              u64 ref_generation,
3363                                              int level,
3364                                              u64 hint,
3365                                              u64 empty_size)
3366 {
3367         struct btrfs_key ins;
3368         int ret;
3369         struct extent_buffer *buf;
3370
3371         ret = btrfs_alloc_extent(trans, root, blocksize, parent, blocksize,
3372                                  root_objectid, ref_generation, level,
3373                                  empty_size, hint, (u64)-1, &ins, 0);
3374         if (ret) {
3375                 BUG_ON(ret > 0);
3376                 return ERR_PTR(ret);
3377         }
3378
3379         buf = btrfs_init_new_buffer(trans, root, ins.objectid, blocksize);
3380         return buf;
3381 }
3382
3383 int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3384                         struct btrfs_root *root, struct extent_buffer *leaf)
3385 {
3386         u64 leaf_owner;
3387         u64 leaf_generation;
3388         struct btrfs_key key;
3389         struct btrfs_file_extent_item *fi;
3390         int i;
3391         int nritems;
3392         int ret;
3393
3394         BUG_ON(!btrfs_is_leaf(leaf));
3395         nritems = btrfs_header_nritems(leaf);
3396         leaf_owner = btrfs_header_owner(leaf);
3397         leaf_generation = btrfs_header_generation(leaf);
3398
3399         for (i = 0; i < nritems; i++) {
3400                 u64 disk_bytenr;
3401                 cond_resched();
3402
3403                 btrfs_item_key_to_cpu(leaf, &key, i);
3404                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
3405                         continue;
3406                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3407                 if (btrfs_file_extent_type(leaf, fi) ==
3408                     BTRFS_FILE_EXTENT_INLINE)
3409                         continue;
3410                 /*
3411                  * FIXME make sure to insert a trans record that
3412                  * repeats the snapshot del on crash
3413                  */
3414                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
3415                 if (disk_bytenr == 0)
3416                         continue;
3417
3418                 ret = __btrfs_free_extent(trans, root, disk_bytenr,
3419                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
3420                                 leaf->start, leaf_owner, leaf_generation,
3421                                 key.objectid, 0);
3422                 BUG_ON(ret);
3423
3424                 atomic_inc(&root->fs_info->throttle_gen);
3425                 wake_up(&root->fs_info->transaction_throttle);
3426                 cond_resched();
3427         }
3428         return 0;
3429 }
3430
3431 static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3432                                         struct btrfs_root *root,
3433                                         struct btrfs_leaf_ref *ref)
3434 {
3435         int i;
3436         int ret;
3437         struct btrfs_extent_info *info = ref->extents;
3438
3439         for (i = 0; i < ref->nritems; i++) {
3440                 ret = __btrfs_free_extent(trans, root, info->bytenr,
3441                                           info->num_bytes, ref->bytenr,
3442                                           ref->owner, ref->generation,
3443                                           info->objectid, 0);
3444
3445                 atomic_inc(&root->fs_info->throttle_gen);
3446                 wake_up(&root->fs_info->transaction_throttle);
3447                 cond_resched();
3448
3449                 BUG_ON(ret);
3450                 info++;
3451         }
3452
3453         return 0;
3454 }
3455
3456 static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start,
3457                                      u64 len, u32 *refs)
3458 {
3459         int ret;
3460
3461         ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs);
3462         BUG_ON(ret);
3463
3464 #if 0 /* some debugging code in case we see problems here */
3465         /* if the refs count is one, it won't get increased again.  But
3466          * if the ref count is > 1, someone may be decreasing it at
3467          * the same time we are.
3468          */
3469         if (*refs != 1) {
3470                 struct extent_buffer *eb = NULL;
3471                 eb = btrfs_find_create_tree_block(root, start, len);
3472                 if (eb)
3473                         btrfs_tree_lock(eb);
3474
3475                 mutex_lock(&root->fs_info->alloc_mutex);
3476                 ret = lookup_extent_ref(NULL, root, start, len, refs);
3477                 BUG_ON(ret);
3478                 mutex_unlock(&root->fs_info->alloc_mutex);
3479
3480                 if (eb) {
3481                         btrfs_tree_unlock(eb);
3482                         free_extent_buffer(eb);
3483                 }
3484                 if (*refs == 1) {
3485                         printk(KERN_ERR "btrfs block %llu went down to one "
3486                                "during drop_snap\n", (unsigned long long)start);
3487                 }
3488
3489         }
3490 #endif
3491
3492         cond_resched();
3493         return ret;
3494 }
3495
3496 /*
3497  * helper function for drop_snapshot, this walks down the tree dropping ref
3498  * counts as it goes.
3499  */
3500 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
3501                                    struct btrfs_root *root,
3502                                    struct btrfs_path *path, int *level)
3503 {
3504         u64 root_owner;
3505         u64 root_gen;
3506         u64 bytenr;
3507         u64 ptr_gen;
3508         struct extent_buffer *next;
3509         struct extent_buffer *cur;
3510         struct extent_buffer *parent;
3511         struct btrfs_leaf_ref *ref;
3512         u32 blocksize;
3513         int ret;
3514         u32 refs;
3515
3516         WARN_ON(*level < 0);
3517         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3518         ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start,
3519                                 path->nodes[*level]->len, &refs);
3520         BUG_ON(ret);
3521         if (refs > 1)
3522                 goto out;
3523
3524         /*
3525          * walk down to the last node level and free all the leaves
3526          */
3527         while (*level >= 0) {
3528                 WARN_ON(*level < 0);
3529                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3530                 cur = path->nodes[*level];
3531
3532                 if (btrfs_header_level(cur) != *level)
3533                         WARN_ON(1);
3534
3535                 if (path->slots[*level] >=
3536                     btrfs_header_nritems(cur))
3537                         break;
3538                 if (*level == 0) {
3539                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3540                         BUG_ON(ret);
3541                         break;
3542                 }
3543                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3544                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3545                 blocksize = btrfs_level_size(root, *level - 1);
3546
3547                 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs);
3548                 BUG_ON(ret);
3549                 if (refs != 1) {
3550                         parent = path->nodes[*level];
3551                         root_owner = btrfs_header_owner(parent);
3552                         root_gen = btrfs_header_generation(parent);
3553                         path->slots[*level]++;
3554
3555                         ret = __btrfs_free_extent(trans, root, bytenr,
3556                                                 blocksize, parent->start,
3557                                                 root_owner, root_gen,
3558                                                 *level - 1, 1);
3559                         BUG_ON(ret);
3560
3561                         atomic_inc(&root->fs_info->throttle_gen);
3562                         wake_up(&root->fs_info->transaction_throttle);
3563                         cond_resched();
3564
3565                         continue;
3566                 }
3567                 /*
3568                  * at this point, we have a single ref, and since the
3569                  * only place referencing this extent is a dead root
3570                  * the reference count should never go higher.
3571                  * So, we don't need to check it again
3572                  */
3573                 if (*level == 1) {
3574                         ref = btrfs_lookup_leaf_ref(root, bytenr);
3575                         if (ref && ref->generation != ptr_gen) {
3576                                 btrfs_free_leaf_ref(root, ref);
3577                                 ref = NULL;
3578                         }
3579                         if (ref) {
3580                                 ret = cache_drop_leaf_ref(trans, root, ref);
3581                                 BUG_ON(ret);
3582                                 btrfs_remove_leaf_ref(root, ref);
3583                                 btrfs_free_leaf_ref(root, ref);
3584                                 *level = 0;
3585                                 break;
3586                         }
3587                 }
3588                 next = btrfs_find_tree_block(root, bytenr, blocksize);
3589                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
3590                         free_extent_buffer(next);
3591
3592                         next = read_tree_block(root, bytenr, blocksize,
3593                                                ptr_gen);
3594                         cond_resched();
3595 #if 0
3596                         /*
3597                          * this is a debugging check and can go away
3598                          * the ref should never go all the way down to 1
3599                          * at this point
3600                          */
3601                         ret = lookup_extent_ref(NULL, root, bytenr, blocksize,
3602                                                 &refs);
3603                         BUG_ON(ret);
3604                         WARN_ON(refs != 1);
3605 #endif
3606                 }
3607                 WARN_ON(*level <= 0);
3608                 if (path->nodes[*level-1])
3609                         free_extent_buffer(path->nodes[*level-1]);
3610                 path->nodes[*level-1] = next;
3611                 *level = btrfs_header_level(next);
3612                 path->slots[*level] = 0;
3613                 cond_resched();
3614         }
3615 out:
3616         WARN_ON(*level < 0);
3617         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3618
3619         if (path->nodes[*level] == root->node) {
3620                 parent = path->nodes[*level];
3621                 bytenr = path->nodes[*level]->start;
3622         } else {
3623                 parent = path->nodes[*level + 1];
3624                 bytenr = btrfs_node_blockptr(parent, path->slots[*level + 1]);
3625         }
3626
3627         blocksize = btrfs_level_size(root, *level);
3628         root_owner = btrfs_header_owner(parent);
3629         root_gen = btrfs_header_generation(parent);
3630
3631         ret = __btrfs_free_extent(trans, root, bytenr, blocksize,
3632                                   parent->start, root_owner, root_gen,
3633                                   *level, 1);
3634         free_extent_buffer(path->nodes[*level]);
3635         path->nodes[*level] = NULL;
3636         *level += 1;
3637         BUG_ON(ret);
3638
3639         cond_resched();
3640         return 0;
3641 }
3642
3643 /*
3644  * helper function for drop_subtree, this function is similar to
3645  * walk_down_tree. The main difference is that it checks reference
3646  * counts while tree blocks are locked.
3647  */
3648 static noinline int walk_down_subtree(struct btrfs_trans_handle *trans,
3649                                       struct btrfs_root *root,
3650                                       struct btrfs_path *path, int *level)
3651 {
3652         struct extent_buffer *next;
3653         struct extent_buffer *cur;
3654         struct extent_buffer *parent;
3655         u64 bytenr;
3656         u64 ptr_gen;
3657         u32 blocksize;
3658         u32 refs;
3659         int ret;
3660
3661         cur = path->nodes[*level];
3662         ret = btrfs_lookup_extent_ref(trans, root, cur->start, cur->len,
3663                                       &refs);
3664         BUG_ON(ret);
3665         if (refs > 1)
3666                 goto out;
3667
3668         while (*level >= 0) {
3669                 cur = path->nodes[*level];
3670                 if (*level == 0) {
3671                         ret = btrfs_drop_leaf_ref(trans, root, cur);
3672                         BUG_ON(ret);
3673                         clean_tree_block(trans, root, cur);
3674                         break;
3675                 }
3676                 if (path->slots[*level] >= btrfs_header_nritems(cur)) {
3677                         clean_tree_block(trans, root, cur);
3678                         break;
3679                 }
3680
3681                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
3682                 blocksize = btrfs_level_size(root, *level - 1);
3683                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
3684
3685                 next = read_tree_block(root, bytenr, blocksize, ptr_gen);
3686                 btrfs_tree_lock(next);
3687
3688                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
3689                                               &refs);
3690                 BUG_ON(ret);
3691                 if (refs > 1) {
3692                         parent = path->nodes[*level];
3693                         ret = btrfs_free_extent(trans, root, bytenr,
3694                                         blocksize, parent->start,
3695                                         btrfs_header_owner(parent),
3696                                         btrfs_header_generation(parent),
3697                                         *level - 1, 1);
3698                         BUG_ON(ret);
3699                         path->slots[*level]++;
3700                         btrfs_tree_unlock(next);
3701                         free_extent_buffer(next);
3702                         continue;
3703                 }
3704
3705                 *level = btrfs_header_level(next);
3706                 path->nodes[*level] = next;
3707                 path->slots[*level] = 0;
3708                 path->locks[*level] = 1;
3709                 cond_resched();
3710         }
3711 out:
3712         parent = path->nodes[*level + 1];
3713         bytenr = path->nodes[*level]->start;
3714         blocksize = path->nodes[*level]->len;
3715
3716         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
3717                         parent->start, btrfs_header_owner(parent),
3718                         btrfs_header_generation(parent), *level, 1);
3719         BUG_ON(ret);
3720
3721         if (path->locks[*level]) {
3722                 btrfs_tree_unlock(path->nodes[*level]);
3723                 path->locks[*level] = 0;
3724         }
3725         free_extent_buffer(path->nodes[*level]);
3726         path->nodes[*level] = NULL;
3727         *level += 1;
3728         cond_resched();
3729         return 0;
3730 }
3731
3732 /*
3733  * helper for dropping snapshots.  This walks back up the tree in the path
3734  * to find the first node higher up where we haven't yet gone through
3735  * all the slots
3736  */
3737 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
3738                                  struct btrfs_root *root,
3739                                  struct btrfs_path *path,
3740                                  int *level, int max_level)
3741 {
3742         u64 root_owner;
3743         u64 root_gen;
3744         struct btrfs_root_item *root_item = &root->root_item;
3745         int i;
3746         int slot;
3747         int ret;
3748
3749         for (i = *level; i < max_level && path->nodes[i]; i++) {
3750                 slot = path->slots[i];
3751                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3752                         struct extent_buffer *node;
3753                         struct btrfs_disk_key disk_key;
3754                         node = path->nodes[i];
3755                         path->slots[i]++;
3756                         *level = i;
3757                         WARN_ON(*level == 0);
3758                         btrfs_node_key(node, &disk_key, path->slots[i]);
3759                         memcpy(&root_item->drop_progress,
3760                                &disk_key, sizeof(disk_key));
3761                         root_item->drop_level = i;
3762                         return 0;
3763                 } else {
3764                         struct extent_buffer *parent;
3765                         if (path->nodes[*level] == root->node)
3766                                 parent = path->nodes[*level];
3767                         else
3768                                 parent = path->nodes[*level + 1];
3769
3770                         root_owner = btrfs_header_owner(parent);
3771                         root_gen = btrfs_header_generation(parent);
3772
3773                         clean_tree_block(trans, root, path->nodes[*level]);
3774                         ret = btrfs_free_extent(trans, root,
3775                                                 path->nodes[*level]->start,
3776                                                 path->nodes[*level]->len,
3777                                                 parent->start, root_owner,
3778                                                 root_gen, *level, 1);
3779                         BUG_ON(ret);
3780                         if (path->locks[*level]) {
3781                                 btrfs_tree_unlock(path->nodes[*level]);
3782                                 path->locks[*level] = 0;
3783                         }
3784                         free_extent_buffer(path->nodes[*level]);
3785                         path->nodes[*level] = NULL;
3786                         *level = i + 1;
3787                 }
3788         }
3789         return 1;
3790 }
3791
3792 /*
3793  * drop the reference count on the tree rooted at 'snap'.  This traverses
3794  * the tree freeing any blocks that have a ref count of zero after being
3795  * decremented.
3796  */
3797 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
3798                         *root)
3799 {
3800         int ret = 0;
3801         int wret;
3802         int level;
3803         struct btrfs_path *path;
3804         int i;
3805         int orig_level;
3806         struct btrfs_root_item *root_item = &root->root_item;
3807
3808         WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
3809         path = btrfs_alloc_path();
3810         BUG_ON(!path);
3811
3812         level = btrfs_header_level(root->node);
3813         orig_level = level;
3814         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3815                 path->nodes[level] = root->node;
3816                 extent_buffer_get(root->node);
3817                 path->slots[level] = 0;
3818         } else {
3819                 struct btrfs_key key;
3820                 struct btrfs_disk_key found_key;
3821                 struct extent_buffer *node;
3822
3823                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3824                 level = root_item->drop_level;
3825                 path->lowest_level = level;
3826                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3827                 if (wret < 0) {
3828                         ret = wret;
3829                         goto out;
3830                 }
3831                 node = path->nodes[level];
3832                 btrfs_node_key(node, &found_key, path->slots[level]);
3833                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3834                                sizeof(found_key)));
3835                 /*
3836                  * unlock our path, this is safe because only this
3837                  * function is allowed to delete this snapshot
3838                  */
3839                 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3840                         if (path->nodes[i] && path->locks[i]) {
3841                                 path->locks[i] = 0;
3842                                 btrfs_tree_unlock(path->nodes[i]);
3843                         }
3844                 }
3845         }
3846         while (1) {
3847                 wret = walk_down_tree(trans, root, path, &level);
3848                 if (wret > 0)
3849                         break;
3850                 if (wret < 0)
3851                         ret = wret;
3852
3853                 wret = walk_up_tree(trans, root, path, &level,
3854                                     BTRFS_MAX_LEVEL);
3855                 if (wret > 0)
3856                         break;
3857                 if (wret < 0)
3858                         ret = wret;
3859                 if (trans->transaction->in_commit) {
3860                         ret = -EAGAIN;
3861                         break;
3862                 }
3863                 atomic_inc(&root->fs_info->throttle_gen);
3864                 wake_up(&root->fs_info->transaction_throttle);
3865         }
3866         for (i = 0; i <= orig_level; i++) {
3867                 if (path->nodes[i]) {
3868                         free_extent_buffer(path->nodes[i]);
3869                         path->nodes[i] = NULL;
3870                 }
3871         }
3872 out:
3873         btrfs_free_path(path);
3874         return ret;
3875 }
3876
3877 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
3878                         struct btrfs_root *root,
3879                         struct extent_buffer *node,
3880                         struct extent_buffer *parent)
3881 {
3882         struct btrfs_path *path;
3883         int level;
3884         int parent_level;
3885         int ret = 0;
3886         int wret;
3887
3888         path = btrfs_alloc_path();
3889         BUG_ON(!path);
3890
3891         BUG_ON(!btrfs_tree_locked(parent));
3892         parent_level = btrfs_header_level(parent);
3893         extent_buffer_get(parent);
3894         path->nodes[parent_level] = parent;
3895         path->slots[parent_level] = btrfs_header_nritems(parent);
3896
3897         BUG_ON(!btrfs_tree_locked(node));
3898         level = btrfs_header_level(node);
3899         extent_buffer_get(node);
3900         path->nodes[level] = node;
3901         path->slots[level] = 0;
3902
3903         while (1) {
3904                 wret = walk_down_subtree(trans, root, path, &level);
3905                 if (wret < 0)
3906                         ret = wret;
3907                 if (wret != 0)
3908                         break;
3909
3910                 wret = walk_up_tree(trans, root, path, &level, parent_level);
3911                 if (wret < 0)
3912                         ret = wret;
3913                 if (wret != 0)
3914                         break;
3915         }
3916
3917         btrfs_free_path(path);
3918         return ret;
3919 }
3920
3921 static unsigned long calc_ra(unsigned long start, unsigned long last,
3922                              unsigned long nr)
3923 {
3924         return min(last, start + nr - 1);
3925 }
3926
3927 static noinline int relocate_inode_pages(struct inode *inode, u64 start,
3928                                          u64 len)
3929 {
3930         u64 page_start;
3931         u64 page_end;
3932         unsigned long first_index;
3933         unsigned long last_index;
3934         unsigned long i;
3935         struct page *page;
3936         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3937         struct file_ra_state *ra;
3938         struct btrfs_ordered_extent *ordered;
3939         unsigned int total_read = 0;
3940         unsigned int total_dirty = 0;
3941         int ret = 0;
3942
3943         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3944
3945         mutex_lock(&inode->i_mutex);
3946         first_index = start >> PAGE_CACHE_SHIFT;
3947         last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
3948
3949         /* make sure the dirty trick played by the caller work */
3950         ret = invalidate_inode_pages2_range(inode->i_mapping,
3951                                             first_index, last_index);
3952         if (ret)
3953                 goto out_unlock;
3954
3955         file_ra_state_init(ra, inode->i_mapping);
3956
3957         for (i = first_index ; i <= last_index; i++) {
3958                 if (total_read % ra->ra_pages == 0) {
3959                         btrfs_force_ra(inode->i_mapping, ra, NULL, i,
3960                                        calc_ra(i, last_index, ra->ra_pages));
3961                 }
3962                 total_read++;
3963 again:
3964                 if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
3965                         BUG_ON(1);
3966                 page = grab_cache_page(inode->i_mapping, i);
3967                 if (!page) {
3968                         ret = -ENOMEM;
3969                         goto out_unlock;
3970                 }
3971                 if (!PageUptodate(page)) {
3972                         btrfs_readpage(NULL, page);
3973                         lock_page(page);
3974                         if (!PageUptodate(page)) {
3975                                 unlock_page(page);
3976                                 page_cache_release(page);
3977                                 ret = -EIO;
3978                                 goto out_unlock;
3979                         }
3980                 }
3981                 wait_on_page_writeback(page);
3982
3983                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
3984                 page_end = page_start + PAGE_CACHE_SIZE - 1;
3985                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3986
3987                 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3988                 if (ordered) {
3989                         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3990                         unlock_page(page);
3991                         page_cache_release(page);
3992                         btrfs_start_ordered_extent(inode, ordered, 1);
3993                         btrfs_put_ordered_extent(ordered);
3994                         goto again;
3995                 }
3996                 set_page_extent_mapped(page);
3997
3998                 if (i == first_index)
3999                         set_extent_bits(io_tree, page_start, page_end,
4000                                         EXTENT_BOUNDARY, GFP_NOFS);
4001                 btrfs_set_extent_delalloc(inode, page_start, page_end);
4002
4003                 set_page_dirty(page);
4004                 total_dirty++;
4005
4006                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4007                 unlock_page(page);
4008                 page_cache_release(page);
4009         }
4010
4011 out_unlock:
4012         kfree(ra);
4013         mutex_unlock(&inode->i_mutex);
4014         balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
4015         return ret;
4016 }
4017
4018 static noinline int relocate_data_extent(struct inode *reloc_inode,
4019                                          struct btrfs_key *extent_key,
4020                                          u64 offset)
4021 {
4022         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4023         struct extent_map_tree *em_tree = &BTRFS_I(reloc_inode)->extent_tree;
4024         struct extent_map *em;
4025         u64 start = extent_key->objectid - offset;
4026         u64 end = start + extent_key->offset - 1;
4027
4028         em = alloc_extent_map(GFP_NOFS);
4029         BUG_ON(!em || IS_ERR(em));
4030
4031         em->start = start;
4032         em->len = extent_key->offset;
4033         em->block_len = extent_key->offset;
4034         em->block_start = extent_key->objectid;
4035         em->bdev = root->fs_info->fs_devices->latest_bdev;
4036         set_bit(EXTENT_FLAG_PINNED, &em->flags);
4037
4038         /* setup extent map to cheat btrfs_readpage */
4039         lock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4040         while (1) {
4041                 int ret;
4042                 spin_lock(&em_tree->lock);
4043                 ret = add_extent_mapping(em_tree, em);
4044                 spin_unlock(&em_tree->lock);
4045                 if (ret != -EEXIST) {
4046                         free_extent_map(em);
4047                         break;
4048                 }
4049                 btrfs_drop_extent_cache(reloc_inode, start, end, 0);
4050         }
4051         unlock_extent(&BTRFS_I(reloc_inode)->io_tree, start, end, GFP_NOFS);
4052
4053         return relocate_inode_pages(reloc_inode, start, extent_key->offset);
4054 }
4055
4056 struct btrfs_ref_path {
4057         u64 extent_start;
4058         u64 nodes[BTRFS_MAX_LEVEL];
4059         u64 root_objectid;
4060         u64 root_generation;
4061         u64 owner_objectid;
4062         u32 num_refs;
4063         int lowest_level;
4064         int current_level;
4065         int shared_level;
4066
4067         struct btrfs_key node_keys[BTRFS_MAX_LEVEL];
4068         u64 new_nodes[BTRFS_MAX_LEVEL];
4069 };
4070
4071 struct disk_extent {
4072         u64 ram_bytes;
4073         u64 disk_bytenr;
4074         u64 disk_num_bytes;
4075         u64 offset;
4076         u64 num_bytes;
4077         u8 compression;
4078         u8 encryption;
4079         u16 other_encoding;
4080 };
4081
4082 static int is_cowonly_root(u64 root_objectid)
4083 {
4084         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
4085             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
4086             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
4087             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
4088             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
4089             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
4090                 return 1;
4091         return 0;
4092 }
4093
4094 static noinline int __next_ref_path(struct btrfs_trans_handle *trans,
4095                                     struct btrfs_root *extent_root,
4096                                     struct btrfs_ref_path *ref_path,
4097                                     int first_time)
4098 {
4099         struct extent_buffer *leaf;
4100         struct btrfs_path *path;
4101         struct btrfs_extent_ref *ref;
4102         struct btrfs_key key;
4103         struct btrfs_key found_key;
4104         u64 bytenr;
4105         u32 nritems;
4106         int level;
4107         int ret = 1;
4108
4109         path = btrfs_alloc_path();
4110         if (!path)
4111                 return -ENOMEM;
4112
4113         if (first_time) {
4114                 ref_path->lowest_level = -1;
4115                 ref_path->current_level = -1;
4116                 ref_path->shared_level = -1;
4117                 goto walk_up;
4118         }
4119 walk_down:
4120         level = ref_path->current_level - 1;
4121         while (level >= -1) {
4122                 u64 parent;
4123                 if (level < ref_path->lowest_level)
4124                         break;
4125
4126                 if (level >= 0)
4127                         bytenr = ref_path->nodes[level];
4128                 else
4129                         bytenr = ref_path->extent_start;
4130                 BUG_ON(bytenr == 0);
4131
4132                 parent = ref_path->nodes[level + 1];
4133                 ref_path->nodes[level + 1] = 0;
4134                 ref_path->current_level = level;
4135                 BUG_ON(parent == 0);
4136
4137                 key.objectid = bytenr;
4138                 key.offset = parent + 1;
4139                 key.type = BTRFS_EXTENT_REF_KEY;
4140
4141                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4142                 if (ret < 0)
4143                         goto out;
4144                 BUG_ON(ret == 0);
4145
4146                 leaf = path->nodes[0];
4147                 nritems = btrfs_header_nritems(leaf);
4148                 if (path->slots[0] >= nritems) {
4149                         ret = btrfs_next_leaf(extent_root, path);
4150                         if (ret < 0)
4151                                 goto out;
4152                         if (ret > 0)
4153                                 goto next;
4154                         leaf = path->nodes[0];
4155                 }
4156
4157                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4158                 if (found_key.objectid == bytenr &&
4159                     found_key.type == BTRFS_EXTENT_REF_KEY) {
4160                         if (level < ref_path->shared_level)
4161                                 ref_path->shared_level = level;
4162                         goto found;
4163                 }
4164 next:
4165                 level--;
4166                 btrfs_release_path(extent_root, path);
4167                 cond_resched();
4168         }
4169         /* reached lowest level */
4170         ret = 1;
4171         goto out;
4172 walk_up:
4173         level = ref_path->current_level;
4174         while (level < BTRFS_MAX_LEVEL - 1) {
4175                 u64 ref_objectid;
4176
4177                 if (level >= 0)
4178                         bytenr = ref_path->nodes[level];
4179                 else
4180                         bytenr = ref_path->extent_start;
4181
4182                 BUG_ON(bytenr == 0);
4183
4184                 key.objectid = bytenr;
4185                 key.offset = 0;
4186                 key.type = BTRFS_EXTENT_REF_KEY;
4187
4188                 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
4189                 if (ret < 0)
4190                         goto out;
4191
4192                 leaf = path->nodes[0];
4193                 nritems = btrfs_header_nritems(leaf);
4194                 if (path->slots[0] >= nritems) {
4195                         ret = btrfs_next_leaf(extent_root, path);
4196                         if (ret < 0)
4197                                 goto out;
4198                         if (ret > 0) {
4199                                 /* the extent was freed by someone */
4200                                 if (ref_path->lowest_level == level)
4201                                         goto out;
4202                                 btrfs_release_path(extent_root, path);
4203                                 goto walk_down;
4204                         }
4205                         leaf = path->nodes[0];
4206                 }
4207
4208                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4209                 if (found_key.objectid != bytenr ||
4210                                 found_key.type != BTRFS_EXTENT_REF_KEY) {
4211                         /* the extent was freed by someone */
4212                         if (ref_path->lowest_level == level) {
4213                                 ret = 1;
4214                                 goto out;
4215                         }
4216                         btrfs_release_path(extent_root, path);
4217                         goto walk_down;
4218                 }
4219 found:
4220                 ref = btrfs_item_ptr(leaf, path->slots[0],
4221                                 struct btrfs_extent_ref);
4222                 ref_objectid = btrfs_ref_objectid(leaf, ref);
4223                 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID) {
4224                         if (first_time) {
4225                                 level = (int)ref_objectid;
4226                                 BUG_ON(level >= BTRFS_MAX_LEVEL);
4227                                 ref_path->lowest_level = level;
4228                                 ref_path->current_level = level;
4229                                 ref_path->nodes[level] = bytenr;
4230                         } else {
4231                                 WARN_ON(ref_objectid != level);
4232                         }
4233                 } else {
4234                         WARN_ON(level != -1);
4235                 }
4236                 first_time = 0;
4237
4238                 if (ref_path->lowest_level == level) {
4239                         ref_path->owner_objectid = ref_objectid;
4240                         ref_path->num_refs = btrfs_ref_num_refs(leaf, ref);
4241                 }
4242
4243                 /*
4244                  * the block is tree root or the block isn't in reference
4245                  * counted tree.
4246                  */
4247                 if (found_key.objectid == found_key.offset ||
4248                     is_cowonly_root(btrfs_ref_root(leaf, ref))) {
4249                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4250                         ref_path->root_generation =
4251                                 btrfs_ref_generation(leaf, ref);
4252                         if (level < 0) {
4253                                 /* special reference from the tree log */
4254                                 ref_path->nodes[0] = found_key.offset;
4255                                 ref_path->current_level = 0;
4256                         }
4257                         ret = 0;
4258                         goto out;
4259                 }
4260
4261                 level++;
4262                 BUG_ON(ref_path->nodes[level] != 0);
4263                 ref_path->nodes[level] = found_key.offset;
4264                 ref_path->current_level = level;
4265
4266                 /*
4267                  * the reference was created in the running transaction,
4268                  * no need to continue walking up.
4269                  */
4270                 if (btrfs_ref_generation(leaf, ref) == trans->transid) {
4271                         ref_path->root_objectid = btrfs_ref_root(leaf, ref);
4272                         ref_path->root_generation =
4273                                 btrfs_ref_generation(leaf, ref);
4274                         ret = 0;
4275                         goto out;
4276                 }
4277
4278                 btrfs_release_path(extent_root, path);
4279                 cond_resched();
4280         }
4281         /* reached max tree level, but no tree root found. */
4282         BUG();
4283 out:
4284         btrfs_free_path(path);
4285         return ret;
4286 }
4287
4288 static int btrfs_first_ref_path(struct btrfs_trans_handle *trans,
4289                                 struct btrfs_root *extent_root,
4290                                 struct btrfs_ref_path *ref_path,
4291                                 u64 extent_start)
4292 {
4293         memset(ref_path, 0, sizeof(*ref_path));
4294         ref_path->extent_start = extent_start;
4295
4296         return __next_ref_path(trans, extent_root, ref_path, 1);
4297 }
4298
4299 static int btrfs_next_ref_path(struct btrfs_trans_handle *trans,
4300                                struct btrfs_root *extent_root,
4301                                struct btrfs_ref_path *ref_path)
4302 {
4303         return __next_ref_path(trans, extent_root, ref_path, 0);
4304 }
4305
4306 static noinline int get_new_locations(struct inode *reloc_inode,
4307                                       struct btrfs_key *extent_key,
4308                                       u64 offset, int no_fragment,
4309                                       struct disk_extent **extents,
4310                                       int *nr_extents)
4311 {
4312         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
4313         struct btrfs_path *path;
4314         struct btrfs_file_extent_item *fi;
4315         struct extent_buffer *leaf;
4316         struct disk_extent *exts = *extents;
4317         struct btrfs_key found_key;
4318         u64 cur_pos;
4319         u64 last_byte;
4320         u32 nritems;
4321         int nr = 0;
4322         int max = *nr_extents;
4323         int ret;
4324
4325         WARN_ON(!no_fragment && *extents);
4326         if (!exts) {
4327                 max = 1;
4328                 exts = kmalloc(sizeof(*exts) * max, GFP_NOFS);
4329                 if (!exts)
4330                         return -ENOMEM;
4331         }
4332
4333         path = btrfs_alloc_path();
4334         BUG_ON(!path);
4335
4336         cur_pos = extent_key->objectid - offset;
4337         last_byte = extent_key->objectid + extent_key->offset;
4338         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
4339                                        cur_pos, 0);
4340         if (ret < 0)
4341                 goto out;
4342         if (ret > 0) {
4343                 ret = -ENOENT;
4344                 goto out;
4345         }
4346
4347         while (1) {
4348                 leaf = path->nodes[0];
4349                 nritems = btrfs_header_nritems(leaf);
4350                 if (path->slots[0] >= nritems) {
4351                         ret = btrfs_next_leaf(root, path);
4352                         if (ret < 0)
4353                                 goto out;
4354                         if (ret > 0)
4355                                 break;
4356                         leaf = path->nodes[0];
4357                 }
4358
4359                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4360                 if (found_key.offset != cur_pos ||
4361                     found_key.type != BTRFS_EXTENT_DATA_KEY ||
4362                     found_key.objectid != reloc_inode->i_ino)
4363                         break;
4364
4365                 fi = btrfs_item_ptr(leaf, path->slots[0],
4366                                     struct btrfs_file_extent_item);
4367                 if (btrfs_file_extent_type(leaf, fi) !=
4368                     BTRFS_FILE_EXTENT_REG ||
4369                     btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4370                         break;
4371
4372                 if (nr == max) {
4373                         struct disk_extent *old = exts;
4374                         max *= 2;
4375                         exts = kzalloc(sizeof(*exts) * max, GFP_NOFS);
4376                         memcpy(exts, old, sizeof(*exts) * nr);
4377                         if (old != *extents)
4378                                 kfree(old);
4379                 }
4380
4381                 exts[nr].disk_bytenr =
4382                         btrfs_file_extent_disk_bytenr(leaf, fi);
4383                 exts[nr].disk_num_bytes =
4384                         btrfs_file_extent_disk_num_bytes(leaf, fi);
4385                 exts[nr].offset = btrfs_file_extent_offset(leaf, fi);
4386                 exts[nr].num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4387                 exts[nr].ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
4388                 exts[nr].compression = btrfs_file_extent_compression(leaf, fi);
4389                 exts[nr].encryption = btrfs_file_extent_encryption(leaf, fi);
4390                 exts[nr].other_encoding = btrfs_file_extent_other_encoding(leaf,
4391                                                                            fi);
4392                 BUG_ON(exts[nr].offset > 0);
4393                 BUG_ON(exts[nr].compression || exts[nr].encryption);
4394                 BUG_ON(exts[nr].num_bytes != exts[nr].disk_num_bytes);
4395
4396                 cur_pos += exts[nr].num_bytes;
4397                 nr++;
4398
4399                 if (cur_pos + offset >= last_byte)
4400                         break;
4401
4402                 if (no_fragment) {
4403                         ret = 1;
4404                         goto out;
4405                 }
4406                 path->slots[0]++;
4407         }
4408
4409         BUG_ON(cur_pos + offset > last_byte);
4410         if (cur_pos + offset < last_byte) {
4411                 ret = -ENOENT;
4412                 goto out;
4413         }
4414         ret = 0;
4415 out:
4416         btrfs_free_path(path);
4417         if (ret) {
4418                 if (exts != *extents)
4419                         kfree(exts);
4420         } else {
4421                 *extents = exts;
4422                 *nr_extents = nr;
4423         }
4424         return ret;
4425 }
4426
4427 static noinline int replace_one_extent(struct btrfs_trans_handle *trans,
4428                                         struct btrfs_root *root,
4429                                         struct btrfs_path *path,
4430                                         struct btrfs_key *extent_key,
4431                                         struct btrfs_key *leaf_key,
4432                                         struct btrfs_ref_path *ref_path,
4433                                         struct disk_extent *new_extents,
4434                                         int nr_extents)
4435 {
4436         struct extent_buffer *leaf;
4437         struct btrfs_file_extent_item *fi;
4438         struct inode *inode = NULL;
4439         struct btrfs_key key;
4440         u64 lock_start = 0;
4441         u64 lock_end = 0;
4442         u64 num_bytes;
4443         u64 ext_offset;
4444         u64 first_pos;
4445         u32 nritems;
4446         int nr_scaned = 0;
4447         int extent_locked = 0;
4448         int extent_type;
4449         int ret;
4450
4451         memcpy(&key, leaf_key, sizeof(key));
4452         first_pos = INT_LIMIT(loff_t) - extent_key->offset;
4453         if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4454                 if (key.objectid < ref_path->owner_objectid ||
4455                     (key.objectid == ref_path->owner_objectid &&
4456                      key.type < BTRFS_EXTENT_DATA_KEY)) {
4457                         key.objectid = ref_path->owner_objectid;
4458                         key.type = BTRFS_EXTENT_DATA_KEY;
4459                         key.offset = 0;
4460                 }
4461         }
4462
4463         while (1) {
4464                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4465                 if (ret < 0)
4466                         goto out;
4467
4468                 leaf = path->nodes[0];
4469                 nritems = btrfs_header_nritems(leaf);
4470 next:
4471                 if (extent_locked && ret > 0) {
4472                         /*
4473                          * the file extent item was modified by someone
4474                          * before the extent got locked.
4475                          */
4476                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4477                                       lock_end, GFP_NOFS);
4478                         extent_locked = 0;
4479                 }
4480
4481                 if (path->slots[0] >= nritems) {
4482                         if (++nr_scaned > 2)
4483                                 break;
4484
4485                         BUG_ON(extent_locked);
4486                         ret = btrfs_next_leaf(root, path);
4487                         if (ret < 0)
4488                                 goto out;
4489                         if (ret > 0)
4490                                 break;
4491                         leaf = path->nodes[0];
4492                         nritems = btrfs_header_nritems(leaf);
4493                 }
4494
4495                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4496
4497                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS) {
4498                         if ((key.objectid > ref_path->owner_objectid) ||
4499                             (key.objectid == ref_path->owner_objectid &&
4500                              key.type > BTRFS_EXTENT_DATA_KEY) ||
4501                             (key.offset >= first_pos + extent_key->offset))
4502                                 break;
4503                 }
4504
4505                 if (inode && key.objectid != inode->i_ino) {
4506                         BUG_ON(extent_locked);
4507                         btrfs_release_path(root, path);
4508                         mutex_unlock(&inode->i_mutex);
4509                         iput(inode);
4510                         inode = NULL;
4511                         continue;
4512                 }
4513
4514                 if (key.type != BTRFS_EXTENT_DATA_KEY) {
4515                         path->slots[0]++;
4516                         ret = 1;
4517                         goto next;
4518                 }
4519                 fi = btrfs_item_ptr(leaf, path->slots[0],
4520                                     struct btrfs_file_extent_item);
4521                 extent_type = btrfs_file_extent_type(leaf, fi);
4522                 if ((extent_type != BTRFS_FILE_EXTENT_REG &&
4523                      extent_type != BTRFS_FILE_EXTENT_PREALLOC) ||
4524                     (btrfs_file_extent_disk_bytenr(leaf, fi) !=
4525                      extent_key->objectid)) {
4526                         path->slots[0]++;
4527                         ret = 1;
4528                         goto next;
4529                 }
4530
4531                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4532                 ext_offset = btrfs_file_extent_offset(leaf, fi);
4533
4534                 if (first_pos > key.offset - ext_offset)
4535                         first_pos = key.offset - ext_offset;
4536
4537                 if (!extent_locked) {
4538                         lock_start = key.offset;
4539                         lock_end = lock_start + num_bytes - 1;
4540                 } else {
4541                         if (lock_start > key.offset ||
4542                             lock_end + 1 < key.offset + num_bytes) {
4543                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4544                                               lock_start, lock_end, GFP_NOFS);
4545                                 extent_locked = 0;
4546                         }
4547                 }
4548
4549                 if (!inode) {
4550                         btrfs_release_path(root, path);
4551
4552                         inode = btrfs_iget_locked(root->fs_info->sb,
4553                                                   key.objectid, root);
4554                         if (inode->i_state & I_NEW) {
4555                                 BTRFS_I(inode)->root = root;
4556                                 BTRFS_I(inode)->location.objectid =
4557                                         key.objectid;
4558                                 BTRFS_I(inode)->location.type =
4559                                         BTRFS_INODE_ITEM_KEY;
4560                                 BTRFS_I(inode)->location.offset = 0;
4561                                 btrfs_read_locked_inode(inode);
4562                                 unlock_new_inode(inode);
4563                         }
4564                         /*
4565                          * some code call btrfs_commit_transaction while
4566                          * holding the i_mutex, so we can't use mutex_lock
4567                          * here.
4568                          */
4569                         if (is_bad_inode(inode) ||
4570                             !mutex_trylock(&inode->i_mutex)) {
4571                                 iput(inode);
4572                                 inode = NULL;
4573                                 key.offset = (u64)-1;
4574                                 goto skip;
4575                         }
4576                 }
4577
4578                 if (!extent_locked) {
4579                         struct btrfs_ordered_extent *ordered;
4580
4581                         btrfs_release_path(root, path);
4582
4583                         lock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4584                                     lock_end, GFP_NOFS);
4585                         ordered = btrfs_lookup_first_ordered_extent(inode,
4586                                                                     lock_end);
4587                         if (ordered &&
4588                             ordered->file_offset <= lock_end &&
4589                             ordered->file_offset + ordered->len > lock_start) {
4590                                 unlock_extent(&BTRFS_I(inode)->io_tree,
4591                                               lock_start, lock_end, GFP_NOFS);
4592                                 btrfs_start_ordered_extent(inode, ordered, 1);
4593                                 btrfs_put_ordered_extent(ordered);
4594                                 key.offset += num_bytes;
4595                                 goto skip;
4596                         }
4597                         if (ordered)
4598                                 btrfs_put_ordered_extent(ordered);
4599
4600                         extent_locked = 1;
4601                         continue;
4602                 }
4603
4604                 if (nr_extents == 1) {
4605                         /* update extent pointer in place */
4606                         btrfs_set_file_extent_disk_bytenr(leaf, fi,
4607                                                 new_extents[0].disk_bytenr);
4608                         btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4609                                                 new_extents[0].disk_num_bytes);
4610                         btrfs_mark_buffer_dirty(leaf);
4611
4612                         btrfs_drop_extent_cache(inode, key.offset,
4613                                                 key.offset + num_bytes - 1, 0);
4614
4615                         ret = btrfs_inc_extent_ref(trans, root,
4616                                                 new_extents[0].disk_bytenr,
4617                                                 new_extents[0].disk_num_bytes,
4618                                                 leaf->start,
4619                                                 root->root_key.objectid,
4620                                                 trans->transid,
4621                                                 key.objectid);
4622                         BUG_ON(ret);
4623
4624                         ret = btrfs_free_extent(trans, root,
4625                                                 extent_key->objectid,
4626                                                 extent_key->offset,
4627                                                 leaf->start,
4628                                                 btrfs_header_owner(leaf),
4629                                                 btrfs_header_generation(leaf),
4630                                                 key.objectid, 0);
4631                         BUG_ON(ret);
4632
4633                         btrfs_release_path(root, path);
4634                         key.offset += num_bytes;
4635                 } else {
4636                         BUG_ON(1);
4637 #if 0
4638                         u64 alloc_hint;
4639                         u64 extent_len;
4640                         int i;
4641                         /*
4642                          * drop old extent pointer at first, then insert the
4643                          * new pointers one bye one
4644                          */
4645                         btrfs_release_path(root, path);
4646                         ret = btrfs_drop_extents(trans, root, inode, key.offset,
4647                                                  key.offset + num_bytes,
4648                                                  key.offset, &alloc_hint);
4649                         BUG_ON(ret);
4650
4651                         for (i = 0; i < nr_extents; i++) {
4652                                 if (ext_offset >= new_extents[i].num_bytes) {
4653                                         ext_offset -= new_extents[i].num_bytes;
4654                                         continue;
4655                                 }
4656                                 extent_len = min(new_extents[i].num_bytes -
4657                                                  ext_offset, num_bytes);
4658
4659                                 ret = btrfs_insert_empty_item(trans, root,
4660                                                               path, &key,
4661                                                               sizeof(*fi));
4662                                 BUG_ON(ret);
4663
4664                                 leaf = path->nodes[0];
4665                                 fi = btrfs_item_ptr(leaf, path->slots[0],
4666                                                 struct btrfs_file_extent_item);
4667                                 btrfs_set_file_extent_generation(leaf, fi,
4668                                                         trans->transid);
4669                                 btrfs_set_file_extent_type(leaf, fi,
4670                                                         BTRFS_FILE_EXTENT_REG);
4671                                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4672                                                 new_extents[i].disk_bytenr);
4673                                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4674                                                 new_extents[i].disk_num_bytes);
4675                                 btrfs_set_file_extent_ram_bytes(leaf, fi,
4676                                                 new_extents[i].ram_bytes);
4677
4678                                 btrfs_set_file_extent_compression(leaf, fi,
4679                                                 new_extents[i].compression);
4680                                 btrfs_set_file_extent_encryption(leaf, fi,
4681                                                 new_extents[i].encryption);
4682                                 btrfs_set_file_extent_other_encoding(leaf, fi,
4683                                                 new_extents[i].other_encoding);
4684
4685                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4686                                                         extent_len);
4687                                 ext_offset += new_extents[i].offset;
4688                                 btrfs_set_file_extent_offset(leaf, fi,
4689                                                         ext_offset);
4690                                 btrfs_mark_buffer_dirty(leaf);
4691
4692                                 btrfs_drop_extent_cache(inode, key.offset,
4693                                                 key.offset + extent_len - 1, 0);
4694
4695                                 ret = btrfs_inc_extent_ref(trans, root,
4696                                                 new_extents[i].disk_bytenr,
4697                                                 new_extents[i].disk_num_bytes,
4698                                                 leaf->start,
4699                                                 root->root_key.objectid,
4700                                                 trans->transid, key.objectid);
4701                                 BUG_ON(ret);
4702                                 btrfs_release_path(root, path);
4703
4704                                 inode_add_bytes(inode, extent_len);
4705
4706                                 ext_offset = 0;
4707                                 num_bytes -= extent_len;
4708                                 key.offset += extent_len;
4709
4710                                 if (num_bytes == 0)
4711                                         break;
4712                         }
4713                         BUG_ON(i >= nr_extents);
4714 #endif
4715                 }
4716
4717                 if (extent_locked) {
4718                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4719                                       lock_end, GFP_NOFS);
4720                         extent_locked = 0;
4721                 }
4722 skip:
4723                 if (ref_path->owner_objectid != BTRFS_MULTIPLE_OBJECTIDS &&
4724                     key.offset >= first_pos + extent_key->offset)
4725                         break;
4726
4727                 cond_resched();
4728         }
4729         ret = 0;
4730 out:
4731         btrfs_release_path(root, path);
4732         if (inode) {
4733                 mutex_unlock(&inode->i_mutex);
4734                 if (extent_locked) {
4735                         unlock_extent(&BTRFS_I(inode)->io_tree, lock_start,
4736                                       lock_end, GFP_NOFS);
4737                 }
4738                 iput(inode);
4739         }
4740         return ret;
4741 }
4742
4743 int btrfs_reloc_tree_cache_ref(struct btrfs_trans_handle *trans,
4744                                struct btrfs_root *root,
4745                                struct extent_buffer *buf, u64 orig_start)
4746 {
4747         int level;
4748         int ret;
4749
4750         BUG_ON(btrfs_header_generation(buf) != trans->transid);
4751         BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
4752
4753         level = btrfs_header_level(buf);
4754         if (level == 0) {
4755                 struct btrfs_leaf_ref *ref;
4756                 struct btrfs_leaf_ref *orig_ref;
4757
4758                 orig_ref = btrfs_lookup_leaf_ref(root, orig_start);
4759                 if (!orig_ref)
4760                         return -ENOENT;
4761
4762                 ref = btrfs_alloc_leaf_ref(root, orig_ref->nritems);
4763                 if (!ref) {
4764                         btrfs_free_leaf_ref(root, orig_ref);
4765                         return -ENOMEM;
4766                 }
4767
4768                 ref->nritems = orig_ref->nritems;
4769                 memcpy(ref->extents, orig_ref->extents,
4770                         sizeof(ref->extents[0]) * ref->nritems);
4771
4772                 btrfs_free_leaf_ref(root, orig_ref);
4773
4774                 ref->root_gen = trans->transid;
4775                 ref->bytenr = buf->start;
4776                 ref->owner = btrfs_header_owner(buf);
4777                 ref->generation = btrfs_header_generation(buf);
4778                 ret = btrfs_add_leaf_ref(root, ref, 0);
4779                 WARN_ON(ret);
4780                 btrfs_free_leaf_ref(root, ref);
4781         }
4782         return 0;
4783 }
4784
4785 static noinline int invalidate_extent_cache(struct btrfs_root *root,
4786                                         struct extent_buffer *leaf,
4787                                         struct btrfs_block_group_cache *group,
4788                                         struct btrfs_root *target_root)
4789 {
4790         struct btrfs_key key;
4791         struct inode *inode = NULL;
4792         struct btrfs_file_extent_item *fi;
4793         u64 num_bytes;
4794         u64 skip_objectid = 0;
4795         u32 nritems;
4796         u32 i;
4797
4798         nritems = btrfs_header_nritems(leaf);
4799         for (i = 0; i < nritems; i++) {
4800                 btrfs_item_key_to_cpu(leaf, &key, i);
4801                 if (key.objectid == skip_objectid ||
4802                     key.type != BTRFS_EXTENT_DATA_KEY)
4803                         continue;
4804                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4805                 if (btrfs_file_extent_type(leaf, fi) ==
4806                     BTRFS_FILE_EXTENT_INLINE)
4807                         continue;
4808                 if (btrfs_file_extent_disk_bytenr(leaf, fi) == 0)
4809                         continue;
4810                 if (!inode || inode->i_ino != key.objectid) {
4811                         iput(inode);
4812                         inode = btrfs_ilookup(target_root->fs_info->sb,
4813                                               key.objectid, target_root, 1);
4814                 }
4815                 if (!inode) {
4816                         skip_objectid = key.objectid;
4817                         continue;
4818                 }
4819                 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
4820
4821                 lock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4822                             key.offset + num_bytes - 1, GFP_NOFS);
4823                 btrfs_drop_extent_cache(inode, key.offset,
4824                                         key.offset + num_bytes - 1, 1);
4825                 unlock_extent(&BTRFS_I(inode)->io_tree, key.offset,
4826                               key.offset + num_bytes - 1, GFP_NOFS);
4827                 cond_resched();
4828         }
4829         iput(inode);
4830         return 0;
4831 }
4832
4833 static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
4834                                         struct btrfs_root *root,
4835                                         struct extent_buffer *leaf,
4836                                         struct btrfs_block_group_cache *group,
4837                                         struct inode *reloc_inode)
4838 {
4839         struct btrfs_key key;
4840         struct btrfs_key extent_key;
4841         struct btrfs_file_extent_item *fi;
4842         struct btrfs_leaf_ref *ref;
4843         struct disk_extent *new_extent;
4844         u64 bytenr;
4845         u64 num_bytes;
4846         u32 nritems;
4847         u32 i;
4848         int ext_index;
4849         int nr_extent;
4850         int ret;
4851
4852         new_extent = kmalloc(sizeof(*new_extent), GFP_NOFS);
4853         BUG_ON(!new_extent);
4854
4855         ref = btrfs_lookup_leaf_ref(root, leaf->start);
4856         BUG_ON(!ref);
4857
4858         ext_index = -1;
4859         nritems = btrfs_header_nritems(leaf);
4860         for (i = 0; i < nritems; i++) {
4861                 btrfs_item_key_to_cpu(leaf, &key, i);
4862                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
4863                         continue;
4864                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
4865                 if (btrfs_file_extent_type(leaf, fi) ==
4866                     BTRFS_FILE_EXTENT_INLINE)
4867                         continue;
4868                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4869                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4870                 if (bytenr == 0)
4871                         continue;
4872
4873                 ext_index++;
4874                 if (bytenr >= group->key.objectid + group->key.offset ||
4875                     bytenr + num_bytes <= group->key.objectid)
4876                         continue;
4877
4878                 extent_key.objectid = bytenr;
4879                 extent_key.offset = num_bytes;
4880                 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
4881                 nr_extent = 1;
4882                 ret = get_new_locations(reloc_inode, &extent_key,
4883                                         group->key.objectid, 1,
4884                                         &new_extent, &nr_extent);
4885                 if (ret > 0)
4886                         continue;
4887                 BUG_ON(ret < 0);
4888
4889                 BUG_ON(ref->extents[ext_index].bytenr != bytenr);
4890                 BUG_ON(ref->extents[ext_index].num_bytes != num_bytes);
4891                 ref->extents[ext_index].bytenr = new_extent->disk_bytenr;
4892                 ref->extents[ext_index].num_bytes = new_extent->disk_num_bytes;
4893
4894                 btrfs_set_file_extent_disk_bytenr(leaf, fi,
4895                                                 new_extent->disk_bytenr);
4896                 btrfs_set_file_extent_disk_num_bytes(leaf, fi,
4897                                                 new_extent->disk_num_bytes);
4898                 btrfs_mark_buffer_dirty(leaf);
4899
4900                 ret = btrfs_inc_extent_ref(trans, root,
4901                                         new_extent->disk_bytenr,
4902                                         new_extent->disk_num_bytes,
4903                                         leaf->start,
4904                                         root->root_key.objectid,
4905                                         trans->transid, key.objectid);
4906                 BUG_ON(ret);
4907                 ret = btrfs_free_extent(trans, root,
4908                                         bytenr, num_bytes, leaf->start,
4909                                         btrfs_header_owner(leaf),
4910                                         btrfs_header_generation(leaf),
4911                                         key.objectid, 0);
4912                 BUG_ON(ret);
4913                 cond_resched();
4914         }
4915         kfree(new_extent);
4916         BUG_ON(ext_index + 1 != ref->nritems);
4917         btrfs_free_leaf_ref(root, ref);
4918         return 0;
4919 }
4920
4921 int btrfs_free_reloc_root(struct btrfs_trans_handle *trans,
4922                           struct btrfs_root *root)
4923 {
4924         struct btrfs_root *reloc_root;
4925         int ret;
4926
4927         if (root->reloc_root) {
4928                 reloc_root = root->reloc_root;
4929                 root->reloc_root = NULL;
4930                 list_add(&reloc_root->dead_list,
4931                          &root->fs_info->dead_reloc_roots);
4932
4933                 btrfs_set_root_bytenr(&reloc_root->root_item,
4934                                       reloc_root->node->start);
4935                 btrfs_set_root_level(&root->root_item,
4936                                      btrfs_header_level(reloc_root->node));
4937                 memset(&reloc_root->root_item.drop_progress, 0,
4938                         sizeof(struct btrfs_disk_key));
4939                 reloc_root->root_item.drop_level = 0;
4940
4941                 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4942                                         &reloc_root->root_key,
4943                                         &reloc_root->root_item);
4944                 BUG_ON(ret);
4945         }
4946         return 0;
4947 }
4948
4949 int btrfs_drop_dead_reloc_roots(struct btrfs_root *root)
4950 {
4951         struct btrfs_trans_handle *trans;
4952         struct btrfs_root *reloc_root;
4953         struct btrfs_root *prev_root = NULL;
4954         struct list_head dead_roots;
4955         int ret;
4956         unsigned long nr;
4957
4958         INIT_LIST_HEAD(&dead_roots);
4959         list_splice_init(&root->fs_info->dead_reloc_roots, &dead_roots);
4960
4961         while (!list_empty(&dead_roots)) {
4962                 reloc_root = list_entry(dead_roots.prev,
4963                                         struct btrfs_root, dead_list);
4964                 list_del_init(&reloc_root->dead_list);
4965
4966                 BUG_ON(reloc_root->commit_root != NULL);
4967                 while (1) {
4968                         trans = btrfs_join_transaction(root, 1);
4969                         BUG_ON(!trans);
4970
4971                         mutex_lock(&root->fs_info->drop_mutex);
4972                         ret = btrfs_drop_snapshot(trans, reloc_root);
4973                         if (ret != -EAGAIN)
4974                                 break;
4975                         mutex_unlock(&root->fs_info->drop_mutex);
4976
4977                         nr = trans->blocks_used;
4978                         ret = btrfs_end_transaction(trans, root);
4979                         BUG_ON(ret);
4980                         btrfs_btree_balance_dirty(root, nr);
4981                 }
4982
4983                 free_extent_buffer(reloc_root->node);
4984
4985                 ret = btrfs_del_root(trans, root->fs_info->tree_root,
4986                                      &reloc_root->root_key);
4987                 BUG_ON(ret);
4988                 mutex_unlock(&root->fs_info->drop_mutex);
4989
4990                 nr = trans->blocks_used;
4991                 ret = btrfs_end_transaction(trans, root);
4992                 BUG_ON(ret);
4993                 btrfs_btree_balance_dirty(root, nr);
4994
4995                 kfree(prev_root);
4996                 prev_root = reloc_root;
4997         }
4998         if (prev_root) {
4999                 btrfs_remove_leaf_refs(prev_root, (u64)-1, 0);
5000                 kfree(prev_root);
5001         }
5002         return 0;
5003 }
5004
5005 int btrfs_add_dead_reloc_root(struct btrfs_root *root)
5006 {
5007         list_add(&root->dead_list, &root->fs_info->dead_reloc_roots);
5008         return 0;
5009 }
5010
5011 int btrfs_cleanup_reloc_trees(struct btrfs_root *root)
5012 {
5013         struct btrfs_root *reloc_root;
5014         struct btrfs_trans_handle *trans;
5015         struct btrfs_key location;
5016         int found;
5017         int ret;
5018
5019         mutex_lock(&root->fs_info->tree_reloc_mutex);
5020         ret = btrfs_find_dead_roots(root, BTRFS_TREE_RELOC_OBJECTID, NULL);
5021         BUG_ON(ret);
5022         found = !list_empty(&root->fs_info->dead_reloc_roots);
5023         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5024
5025         if (found) {
5026                 trans = btrfs_start_transaction(root, 1);
5027                 BUG_ON(!trans);
5028                 ret = btrfs_commit_transaction(trans, root);
5029                 BUG_ON(ret);
5030         }
5031
5032         location.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5033         location.offset = (u64)-1;
5034         location.type = BTRFS_ROOT_ITEM_KEY;
5035
5036         reloc_root = btrfs_read_fs_root_no_name(root->fs_info, &location);
5037         BUG_ON(!reloc_root);
5038         btrfs_orphan_cleanup(reloc_root);
5039         return 0;
5040 }
5041
5042 static noinline int init_reloc_tree(struct btrfs_trans_handle *trans,
5043                                     struct btrfs_root *root)
5044 {
5045         struct btrfs_root *reloc_root;
5046         struct extent_buffer *eb;
5047         struct btrfs_root_item *root_item;
5048         struct btrfs_key root_key;
5049         int ret;
5050
5051         BUG_ON(!root->ref_cows);
5052         if (root->reloc_root)
5053                 return 0;
5054
5055         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
5056         BUG_ON(!root_item);
5057
5058         ret = btrfs_copy_root(trans, root, root->commit_root,
5059                               &eb, BTRFS_TREE_RELOC_OBJECTID);
5060         BUG_ON(ret);
5061
5062         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5063         root_key.offset = root->root_key.objectid;
5064         root_key.type = BTRFS_ROOT_ITEM_KEY;
5065
5066         memcpy(root_item, &root->root_item, sizeof(root_item));
5067         btrfs_set_root_refs(root_item, 0);
5068         btrfs_set_root_bytenr(root_item, eb->start);
5069         btrfs_set_root_level(root_item, btrfs_header_level(eb));
5070         btrfs_set_root_generation(root_item, trans->transid);
5071
5072         btrfs_tree_unlock(eb);
5073         free_extent_buffer(eb);
5074
5075         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
5076                                 &root_key, root_item);
5077         BUG_ON(ret);
5078         kfree(root_item);
5079
5080         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
5081                                                  &root_key);
5082         BUG_ON(!reloc_root);
5083         reloc_root->last_trans = trans->transid;
5084         reloc_root->commit_root = NULL;
5085         reloc_root->ref_tree = &root->fs_info->reloc_ref_tree;
5086
5087         root->reloc_root = reloc_root;
5088         return 0;
5089 }
5090
5091 /*
5092  * Core function of space balance.
5093  *
5094  * The idea is using reloc trees to relocate tree blocks in reference
5095  * counted roots. There is one reloc tree for each subvol, and all
5096  * reloc trees share same root key objectid. Reloc trees are snapshots
5097  * of the latest committed roots of subvols (root->commit_root).
5098  *
5099  * To relocate a tree block referenced by a subvol, there are two steps.
5100  * COW the block through subvol's reloc tree, then update block pointer
5101  * in the subvol to point to the new block. Since all reloc trees share
5102  * same root key objectid, doing special handing for tree blocks owned
5103  * by them is easy. Once a tree block has been COWed in one reloc tree,
5104  * we can use the resulting new block directly when the same block is
5105  * required to COW again through other reloc trees. By this way, relocated
5106  * tree blocks are shared between reloc trees, so they are also shared
5107  * between subvols.
5108  */
5109 static noinline int relocate_one_path(struct btrfs_trans_handle *trans,
5110                                       struct btrfs_root *root,
5111                                       struct btrfs_path *path,
5112                                       struct btrfs_key *first_key,
5113                                       struct btrfs_ref_path *ref_path,
5114                                       struct btrfs_block_group_cache *group,
5115                                       struct inode *reloc_inode)
5116 {
5117         struct btrfs_root *reloc_root;
5118         struct extent_buffer *eb = NULL;
5119         struct btrfs_key *keys;
5120         u64 *nodes;
5121         int level;
5122         int shared_level;
5123         int lowest_level = 0;
5124         int ret;
5125
5126         if (ref_path->owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
5127                 lowest_level = ref_path->owner_objectid;
5128
5129         if (!root->ref_cows) {
5130                 path->lowest_level = lowest_level;
5131                 ret = btrfs_search_slot(trans, root, first_key, path, 0, 1);
5132                 BUG_ON(ret < 0);
5133                 path->lowest_level = 0;
5134                 btrfs_release_path(root, path);
5135                 return 0;
5136         }
5137
5138         mutex_lock(&root->fs_info->tree_reloc_mutex);
5139         ret = init_reloc_tree(trans, root);
5140         BUG_ON(ret);
5141         reloc_root = root->reloc_root;
5142
5143         shared_level = ref_path->shared_level;
5144         ref_path->shared_level = BTRFS_MAX_LEVEL - 1;
5145
5146         keys = ref_path->node_keys;
5147         nodes = ref_path->new_nodes;
5148         memset(&keys[shared_level + 1], 0,
5149                sizeof(*keys) * (BTRFS_MAX_LEVEL - shared_level - 1));
5150         memset(&nodes[shared_level + 1], 0,
5151                sizeof(*nodes) * (BTRFS_MAX_LEVEL - shared_level - 1));
5152
5153         if (nodes[lowest_level] == 0) {
5154                 path->lowest_level = lowest_level;
5155                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5156                                         0, 1);
5157                 BUG_ON(ret);
5158                 for (level = lowest_level; level < BTRFS_MAX_LEVEL; level++) {
5159                         eb = path->nodes[level];
5160                         if (!eb || eb == reloc_root->node)
5161                                 break;
5162                         nodes[level] = eb->start;
5163                         if (level == 0)
5164                                 btrfs_item_key_to_cpu(eb, &keys[level], 0);
5165                         else
5166                                 btrfs_node_key_to_cpu(eb, &keys[level], 0);
5167                 }
5168                 if (nodes[0] &&
5169                     ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5170                         eb = path->nodes[0];
5171                         ret = replace_extents_in_leaf(trans, reloc_root, eb,
5172                                                       group, reloc_inode);
5173                         BUG_ON(ret);
5174                 }
5175                 btrfs_release_path(reloc_root, path);
5176         } else {
5177                 ret = btrfs_merge_path(trans, reloc_root, keys, nodes,
5178                                        lowest_level);
5179                 BUG_ON(ret);
5180         }
5181
5182         /*
5183          * replace tree blocks in the fs tree with tree blocks in
5184          * the reloc tree.
5185          */
5186         ret = btrfs_merge_path(trans, root, keys, nodes, lowest_level);
5187         BUG_ON(ret < 0);
5188
5189         if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5190                 ret = btrfs_search_slot(trans, reloc_root, first_key, path,
5191                                         0, 0);
5192                 BUG_ON(ret);
5193                 extent_buffer_get(path->nodes[0]);
5194                 eb = path->nodes[0];
5195                 btrfs_release_path(reloc_root, path);
5196                 ret = invalidate_extent_cache(reloc_root, eb, group, root);
5197                 BUG_ON(ret);
5198                 free_extent_buffer(eb);
5199         }
5200
5201         mutex_unlock(&root->fs_info->tree_reloc_mutex);
5202         path->lowest_level = 0;
5203         return 0;
5204 }
5205
5206 static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5207                                         struct btrfs_root *root,
5208                                         struct btrfs_path *path,
5209                                         struct btrfs_key *first_key,
5210                                         struct btrfs_ref_path *ref_path)
5211 {
5212         int ret;
5213
5214         ret = relocate_one_path(trans, root, path, first_key,
5215                                 ref_path, NULL, NULL);
5216         BUG_ON(ret);
5217
5218         if (root == root->fs_info->extent_root)
5219                 btrfs_extent_post_op(trans, root);
5220
5221         return 0;
5222 }
5223
5224 static noinline int del_extent_zero(struct btrfs_trans_handle *trans,
5225                                     struct btrfs_root *extent_root,
5226                                     struct btrfs_path *path,
5227                                     struct btrfs_key *extent_key)
5228 {
5229         int ret;
5230
5231         ret = btrfs_search_slot(trans, extent_root, extent_key, path, -1, 1);
5232         if (ret)
5233                 goto out;
5234         ret = btrfs_del_item(trans, extent_root, path);
5235 out:
5236         btrfs_release_path(extent_root, path);
5237         return ret;
5238 }
5239
5240 static noinline struct btrfs_root *read_ref_root(struct btrfs_fs_info *fs_info,
5241                                                 struct btrfs_ref_path *ref_path)
5242 {
5243         struct btrfs_key root_key;
5244
5245         root_key.objectid = ref_path->root_objectid;
5246         root_key.type = BTRFS_ROOT_ITEM_KEY;
5247         if (is_cowonly_root(ref_path->root_objectid))
5248                 root_key.offset = 0;
5249         else
5250                 root_key.offset = (u64)-1;
5251
5252         return btrfs_read_fs_root_no_name(fs_info, &root_key);
5253 }
5254
5255 static noinline int relocate_one_extent(struct btrfs_root *extent_root,
5256                                         struct btrfs_path *path,
5257                                         struct btrfs_key *extent_key,
5258                                         struct btrfs_block_group_cache *group,
5259                                         struct inode *reloc_inode, int pass)
5260 {
5261         struct btrfs_trans_handle *trans;
5262         struct btrfs_root *found_root;
5263         struct btrfs_ref_path *ref_path = NULL;
5264         struct disk_extent *new_extents = NULL;
5265         int nr_extents = 0;
5266         int loops;
5267         int ret;
5268         int level;
5269         struct btrfs_key first_key;
5270         u64 prev_block = 0;
5271
5272
5273         trans = btrfs_start_transaction(extent_root, 1);
5274         BUG_ON(!trans);
5275
5276         if (extent_key->objectid == 0) {
5277                 ret = del_extent_zero(trans, extent_root, path, extent_key);
5278                 goto out;
5279         }
5280
5281         ref_path = kmalloc(sizeof(*ref_path), GFP_NOFS);
5282         if (!ref_path) {
5283                 ret = -ENOMEM;
5284                 goto out;
5285         }
5286
5287         for (loops = 0; ; loops++) {
5288                 if (loops == 0) {
5289                         ret = btrfs_first_ref_path(trans, extent_root, ref_path,
5290                                                    extent_key->objectid);
5291                 } else {
5292                         ret = btrfs_next_ref_path(trans, extent_root, ref_path);
5293                 }
5294                 if (ret < 0)
5295                         goto out;
5296                 if (ret > 0)
5297                         break;
5298
5299                 if (ref_path->root_objectid == BTRFS_TREE_LOG_OBJECTID ||
5300                     ref_path->root_objectid == BTRFS_TREE_RELOC_OBJECTID)
5301                         continue;
5302
5303                 found_root = read_ref_root(extent_root->fs_info, ref_path);
5304                 BUG_ON(!found_root);
5305                 /*
5306                  * for reference counted tree, only process reference paths
5307                  * rooted at the latest committed root.
5308                  */
5309                 if (found_root->ref_cows &&
5310                     ref_path->root_generation != found_root->root_key.offset)
5311                         continue;
5312
5313                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5314                         if (pass == 0) {
5315                                 /*
5316                                  * copy data extents to new locations
5317                                  */
5318                                 u64 group_start = group->key.objectid;
5319                                 ret = relocate_data_extent(reloc_inode,
5320                                                            extent_key,
5321                                                            group_start);
5322                                 if (ret < 0)
5323                                         goto out;
5324                                 break;
5325                         }
5326                         level = 0;
5327                 } else {
5328                         level = ref_path->owner_objectid;
5329                 }
5330
5331                 if (prev_block != ref_path->nodes[level]) {
5332                         struct extent_buffer *eb;
5333                         u64 block_start = ref_path->nodes[level];
5334                         u64 block_size = btrfs_level_size(found_root, level);
5335
5336                         eb = read_tree_block(found_root, block_start,
5337                                              block_size, 0);
5338                         btrfs_tree_lock(eb);
5339                         BUG_ON(level != btrfs_header_level(eb));
5340
5341                         if (level == 0)
5342                                 btrfs_item_key_to_cpu(eb, &first_key, 0);
5343                         else
5344                                 btrfs_node_key_to_cpu(eb, &first_key, 0);
5345
5346                         btrfs_tree_unlock(eb);
5347                         free_extent_buffer(eb);
5348                         prev_block = block_start;
5349                 }
5350
5351                 btrfs_record_root_in_trans(found_root);
5352                 if (ref_path->owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
5353                         /*
5354                          * try to update data extent references while
5355                          * keeping metadata shared between snapshots.
5356                          */
5357                         if (pass == 1) {
5358                                 ret = relocate_one_path(trans, found_root,
5359                                                 path, &first_key, ref_path,
5360                                                 group, reloc_inode);
5361                                 if (ret < 0)
5362                                         goto out;
5363                                 continue;
5364                         }
5365                         /*
5366                          * use fallback method to process the remaining
5367                          * references.
5368                          */
5369                         if (!new_extents) {
5370                                 u64 group_start = group->key.objectid;
5371                                 new_extents = kmalloc(sizeof(*new_extents),
5372                                                       GFP_NOFS);
5373                                 nr_extents = 1;
5374                                 ret = get_new_locations(reloc_inode,
5375                                                         extent_key,
5376                                                         group_start, 1,
5377                                                         &new_extents,
5378                                                         &nr_extents);
5379                                 if (ret)
5380                                         goto out;
5381                         }
5382                         ret = replace_one_extent(trans, found_root,
5383                                                 path, extent_key,
5384                                                 &first_key, ref_path,
5385                                                 new_extents, nr_extents);
5386                 } else {
5387                         ret = relocate_tree_block(trans, found_root, path,
5388                                                   &first_key, ref_path);
5389                 }
5390                 if (ret < 0)
5391                         goto out;
5392         }
5393         ret = 0;
5394 out:
5395         btrfs_end_transaction(trans, extent_root);
5396         kfree(new_extents);
5397         kfree(ref_path);
5398         return ret;
5399 }
5400
5401 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
5402 {
5403         u64 num_devices;
5404         u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
5405                 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
5406
5407         num_devices = root->fs_info->fs_devices->rw_devices;
5408         if (num_devices == 1) {
5409                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5410                 stripped = flags & ~stripped;
5411
5412                 /* turn raid0 into single device chunks */
5413                 if (flags & BTRFS_BLOCK_GROUP_RAID0)
5414                         return stripped;
5415
5416                 /* turn mirroring into duplication */
5417                 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
5418                              BTRFS_BLOCK_GROUP_RAID10))
5419                         return stripped | BTRFS_BLOCK_GROUP_DUP;
5420                 return flags;
5421         } else {
5422                 /* they already had raid on here, just return */
5423                 if (flags & stripped)
5424                         return flags;
5425
5426                 stripped |= BTRFS_BLOCK_GROUP_DUP;
5427                 stripped = flags & ~stripped;
5428
5429                 /* switch duplicated blocks with raid1 */
5430                 if (flags & BTRFS_BLOCK_GROUP_DUP)
5431                         return stripped | BTRFS_BLOCK_GROUP_RAID1;
5432
5433                 /* turn single device chunks into raid0 */
5434                 return stripped | BTRFS_BLOCK_GROUP_RAID0;
5435         }
5436         return flags;
5437 }
5438
5439 static int __alloc_chunk_for_shrink(struct btrfs_root *root,
5440                      struct btrfs_block_group_cache *shrink_block_group,
5441                      int force)
5442 {
5443         struct btrfs_trans_handle *trans;
5444         u64 new_alloc_flags;
5445         u64 calc;
5446
5447         spin_lock(&shrink_block_group->lock);
5448         if (btrfs_block_group_used(&shrink_block_group->item) > 0) {
5449                 spin_unlock(&shrink_block_group->lock);
5450
5451                 trans = btrfs_start_transaction(root, 1);
5452                 spin_lock(&shrink_block_group->lock);
5453
5454                 new_alloc_flags = update_block_group_flags(root,
5455                                                    shrink_block_group->flags);
5456                 if (new_alloc_flags != shrink_block_group->flags) {
5457                         calc =
5458                              btrfs_block_group_used(&shrink_block_group->item);
5459                 } else {
5460                         calc = shrink_block_group->key.offset;
5461                 }
5462                 spin_unlock(&shrink_block_group->lock);
5463
5464                 do_chunk_alloc(trans, root->fs_info->extent_root,
5465                                calc + 2 * 1024 * 1024, new_alloc_flags, force);
5466
5467                 btrfs_end_transaction(trans, root);
5468         } else
5469                 spin_unlock(&shrink_block_group->lock);
5470         return 0;
5471 }
5472
5473 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
5474                                  struct btrfs_root *root,
5475                                  u64 objectid, u64 size)
5476 {
5477         struct btrfs_path *path;
5478         struct btrfs_inode_item *item;
5479         struct extent_buffer *leaf;
5480         int ret;
5481
5482         path = btrfs_alloc_path();
5483         if (!path)
5484                 return -ENOMEM;
5485
5486         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
5487         if (ret)
5488                 goto out;
5489
5490         leaf = path->nodes[0];
5491         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
5492         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
5493         btrfs_set_inode_generation(leaf, item, 1);
5494         btrfs_set_inode_size(leaf, item, size);
5495         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
5496         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
5497         btrfs_mark_buffer_dirty(leaf);
5498         btrfs_release_path(root, path);
5499 out:
5500         btrfs_free_path(path);
5501         return ret;
5502 }
5503
5504 static noinline struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
5505                                         struct btrfs_block_group_cache *group)
5506 {
5507         struct inode *inode = NULL;
5508         struct btrfs_trans_handle *trans;
5509         struct btrfs_root *root;
5510         struct btrfs_key root_key;
5511         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
5512         int err = 0;
5513
5514         root_key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5515         root_key.type = BTRFS_ROOT_ITEM_KEY;
5516         root_key.offset = (u64)-1;
5517         root = btrfs_read_fs_root_no_name(fs_info, &root_key);
5518         if (IS_ERR(root))
5519                 return ERR_CAST(root);
5520
5521         trans = btrfs_start_transaction(root, 1);
5522         BUG_ON(!trans);
5523
5524         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
5525         if (err)
5526                 goto out;
5527
5528         err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
5529         BUG_ON(err);
5530
5531         err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
5532                                        group->key.offset, 0, group->key.offset,
5533                                        0, 0, 0);
5534         BUG_ON(err);
5535
5536         inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
5537         if (inode->i_state & I_NEW) {
5538                 BTRFS_I(inode)->root = root;
5539                 BTRFS_I(inode)->location.objectid = objectid;
5540                 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5541                 BTRFS_I(inode)->location.offset = 0;
5542                 btrfs_read_locked_inode(inode);
5543                 unlock_new_inode(inode);
5544                 BUG_ON(is_bad_inode(inode));
5545         } else {
5546                 BUG_ON(1);
5547         }
5548         BTRFS_I(inode)->index_cnt = group->key.objectid;
5549
5550         err = btrfs_orphan_add(trans, inode);
5551 out:
5552         btrfs_end_transaction(trans, root);
5553         if (err) {
5554                 if (inode)
5555                         iput(inode);
5556                 inode = ERR_PTR(err);
5557         }
5558         return inode;
5559 }
5560
5561 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
5562 {
5563
5564         struct btrfs_ordered_sum *sums;
5565         struct btrfs_sector_sum *sector_sum;
5566         struct btrfs_ordered_extent *ordered;
5567         struct btrfs_root *root = BTRFS_I(inode)->root;
5568         struct list_head list;
5569         size_t offset;
5570         int ret;
5571         u64 disk_bytenr;
5572
5573         INIT_LIST_HEAD(&list);
5574
5575         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
5576         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
5577
5578         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
5579         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
5580                                        disk_bytenr + len - 1, &list);
5581
5582         while (!list_empty(&list)) {
5583                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
5584                 list_del_init(&sums->list);
5585
5586                 sector_sum = sums->sums;
5587                 sums->bytenr = ordered->start;
5588
5589                 offset = 0;
5590                 while (offset < sums->len) {
5591                         sector_sum->bytenr += ordered->start - disk_bytenr;
5592                         sector_sum++;
5593                         offset += root->sectorsize;
5594                 }
5595
5596                 btrfs_add_ordered_sum(inode, ordered, sums);
5597         }
5598         btrfs_put_ordered_extent(ordered);
5599         return 0;
5600 }
5601
5602 int btrfs_relocate_block_group(struct btrfs_root *root, u64 group_start)
5603 {
5604         struct btrfs_trans_handle *trans;
5605         struct btrfs_path *path;
5606         struct btrfs_fs_info *info = root->fs_info;
5607         struct extent_buffer *leaf;
5608         struct inode *reloc_inode;
5609         struct btrfs_block_group_cache *block_group;
5610         struct btrfs_key key;
5611         u64 skipped;
5612         u64 cur_byte;
5613         u64 total_found;
5614         u32 nritems;
5615         int ret;
5616         int progress;
5617         int pass = 0;
5618
5619         root = root->fs_info->extent_root;
5620
5621         block_group = btrfs_lookup_block_group(info, group_start);
5622         BUG_ON(!block_group);
5623
5624         printk(KERN_INFO "btrfs relocating block group %llu flags %llu\n",
5625                (unsigned long long)block_group->key.objectid,
5626                (unsigned long long)block_group->flags);
5627
5628         path = btrfs_alloc_path();
5629         BUG_ON(!path);
5630
5631         reloc_inode = create_reloc_inode(info, block_group);
5632         BUG_ON(IS_ERR(reloc_inode));
5633
5634         __alloc_chunk_for_shrink(root, block_group, 1);
5635         set_block_group_readonly(block_group);
5636
5637         btrfs_start_delalloc_inodes(info->tree_root);
5638         btrfs_wait_ordered_extents(info->tree_root, 0);
5639 again:
5640         skipped = 0;
5641         total_found = 0;
5642         progress = 0;
5643         key.objectid = block_group->key.objectid;
5644         key.offset = 0;
5645         key.type = 0;
5646         cur_byte = key.objectid;
5647
5648         trans = btrfs_start_transaction(info->tree_root, 1);
5649         btrfs_commit_transaction(trans, info->tree_root);
5650
5651         mutex_lock(&root->fs_info->cleaner_mutex);
5652         btrfs_clean_old_snapshots(info->tree_root);
5653         btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
5654         mutex_unlock(&root->fs_info->cleaner_mutex);
5655
5656         while (1) {
5657                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5658                 if (ret < 0)
5659                         goto out;
5660 next:
5661                 leaf = path->nodes[0];
5662                 nritems = btrfs_header_nritems(leaf);
5663                 if (path->slots[0] >= nritems) {
5664                         ret = btrfs_next_leaf(root, path);
5665                         if (ret < 0)
5666                                 goto out;
5667                         if (ret == 1) {
5668                                 ret = 0;
5669                                 break;
5670                         }
5671                         leaf = path->nodes[0];
5672                         nritems = btrfs_header_nritems(leaf);
5673                 }
5674
5675                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5676
5677                 if (key.objectid >= block_group->key.objectid +
5678                     block_group->key.offset)
5679                         break;
5680
5681                 if (progress && need_resched()) {
5682                         btrfs_release_path(root, path);
5683                         cond_resched();
5684                         progress = 0;
5685                         continue;
5686                 }
5687                 progress = 1;
5688
5689                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY ||
5690                     key.objectid + key.offset <= cur_byte) {
5691                         path->slots[0]++;
5692                         goto next;
5693                 }
5694
5695                 total_found++;
5696                 cur_byte = key.objectid + key.offset;
5697                 btrfs_release_path(root, path);
5698
5699                 __alloc_chunk_for_shrink(root, block_group, 0);
5700                 ret = relocate_one_extent(root, path, &key, block_group,
5701                                           reloc_inode, pass);
5702                 BUG_ON(ret < 0);
5703                 if (ret > 0)
5704                         skipped++;
5705
5706                 key.objectid = cur_byte;
5707                 key.type = 0;
5708                 key.offset = 0;
5709         }
5710
5711         btrfs_release_path(root, path);
5712
5713         if (pass == 0) {
5714                 btrfs_wait_ordered_range(reloc_inode, 0, (u64)-1);
5715                 invalidate_mapping_pages(reloc_inode->i_mapping, 0, -1);
5716         }
5717
5718         if (total_found > 0) {
5719                 printk(KERN_INFO "btrfs found %llu extents in pass %d\n",
5720                        (unsigned long long)total_found, pass);
5721                 pass++;
5722                 if (total_found == skipped && pass > 2) {
5723                         iput(reloc_inode);
5724                         reloc_inode = create_reloc_inode(info, block_group);
5725                         pass = 0;
5726                 }
5727                 goto again;
5728         }
5729
5730         /* delete reloc_inode */
5731         iput(reloc_inode);
5732
5733         /* unpin extents in this range */
5734         trans = btrfs_start_transaction(info->tree_root, 1);
5735         btrfs_commit_transaction(trans, info->tree_root);
5736
5737         spin_lock(&block_group->lock);
5738         WARN_ON(block_group->pinned > 0);
5739         WARN_ON(block_group->reserved > 0);
5740         WARN_ON(btrfs_block_group_used(&block_group->item) > 0);
5741         spin_unlock(&block_group->lock);
5742         put_block_group(block_group);
5743         ret = 0;
5744 out:
5745         btrfs_free_path(path);
5746         return ret;
5747 }
5748
5749 static int find_first_block_group(struct btrfs_root *root,
5750                 struct btrfs_path *path, struct btrfs_key *key)
5751 {
5752         int ret = 0;
5753         struct btrfs_key found_key;
5754         struct extent_buffer *leaf;
5755         int slot;
5756
5757         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
5758         if (ret < 0)
5759                 goto out;
5760
5761         while (1) {
5762                 slot = path->slots[0];
5763                 leaf = path->nodes[0];
5764                 if (slot >= btrfs_header_nritems(leaf)) {
5765                         ret = btrfs_next_leaf(root, path);
5766                         if (ret == 0)
5767                                 continue;
5768                         if (ret < 0)
5769                                 goto out;
5770                         break;
5771                 }
5772                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5773
5774                 if (found_key.objectid >= key->objectid &&
5775                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5776                         ret = 0;
5777                         goto out;
5778                 }
5779                 path->slots[0]++;
5780         }
5781         ret = -ENOENT;
5782 out:
5783         return ret;
5784 }
5785
5786 int btrfs_free_block_groups(struct btrfs_fs_info *info)
5787 {
5788         struct btrfs_block_group_cache *block_group;
5789         struct rb_node *n;
5790
5791         spin_lock(&info->block_group_cache_lock);
5792         while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
5793                 block_group = rb_entry(n, struct btrfs_block_group_cache,
5794                                        cache_node);
5795                 rb_erase(&block_group->cache_node,
5796                          &info->block_group_cache_tree);
5797                 spin_unlock(&info->block_group_cache_lock);
5798
5799                 btrfs_remove_free_space_cache(block_group);
5800                 down_write(&block_group->space_info->groups_sem);
5801                 list_del(&block_group->list);
5802                 up_write(&block_group->space_info->groups_sem);
5803
5804                 WARN_ON(atomic_read(&block_group->count) != 1);
5805                 kfree(block_group);
5806
5807                 spin_lock(&info->block_group_cache_lock);
5808         }
5809         spin_unlock(&info->block_group_cache_lock);
5810         return 0;
5811 }
5812
5813 int btrfs_read_block_groups(struct btrfs_root *root)
5814 {
5815         struct btrfs_path *path;
5816         int ret;
5817         struct btrfs_block_group_cache *cache;
5818         struct btrfs_fs_info *info = root->fs_info;
5819         struct btrfs_space_info *space_info;
5820         struct btrfs_key key;
5821         struct btrfs_key found_key;
5822         struct extent_buffer *leaf;
5823
5824         root = info->extent_root;
5825         key.objectid = 0;
5826         key.offset = 0;
5827         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
5828         path = btrfs_alloc_path();
5829         if (!path)
5830                 return -ENOMEM;
5831
5832         while (1) {
5833                 ret = find_first_block_group(root, path, &key);
5834                 if (ret > 0) {
5835                         ret = 0;
5836                         goto error;
5837                 }
5838                 if (ret != 0)
5839                         goto error;
5840
5841                 leaf = path->nodes[0];
5842                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5843                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
5844                 if (!cache) {
5845                         ret = -ENOMEM;
5846                         break;
5847                 }
5848
5849                 atomic_set(&cache->count, 1);
5850                 spin_lock_init(&cache->lock);
5851                 mutex_init(&cache->alloc_mutex);
5852                 mutex_init(&cache->cache_mutex);
5853                 INIT_LIST_HEAD(&cache->list);
5854                 read_extent_buffer(leaf, &cache->item,
5855                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
5856                                    sizeof(cache->item));
5857                 memcpy(&cache->key, &found_key, sizeof(found_key));
5858
5859                 key.objectid = found_key.objectid + found_key.offset;
5860                 btrfs_release_path(root, path);
5861                 cache->flags = btrfs_block_group_flags(&cache->item);
5862
5863                 ret = update_space_info(info, cache->flags, found_key.offset,
5864                                         btrfs_block_group_used(&cache->item),
5865                                         &space_info);
5866                 BUG_ON(ret);
5867                 cache->space_info = space_info;
5868                 down_write(&space_info->groups_sem);
5869                 list_add_tail(&cache->list, &space_info->block_groups);
5870                 up_write(&space_info->groups_sem);
5871
5872                 ret = btrfs_add_block_group_cache(root->fs_info, cache);
5873                 BUG_ON(ret);
5874
5875                 set_avail_alloc_bits(root->fs_info, cache->flags);
5876                 if (btrfs_chunk_readonly(root, cache->key.objectid))
5877                         set_block_group_readonly(cache);
5878         }
5879         ret = 0;
5880 error:
5881         btrfs_free_path(path);
5882         return ret;
5883 }
5884
5885 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
5886                            struct btrfs_root *root, u64 bytes_used,
5887                            u64 type, u64 chunk_objectid, u64 chunk_offset,
5888                            u64 size)
5889 {
5890         int ret;
5891         struct btrfs_root *extent_root;
5892         struct btrfs_block_group_cache *cache;
5893
5894         extent_root = root->fs_info->extent_root;
5895
5896         root->fs_info->last_trans_new_blockgroup = trans->transid;
5897
5898         cache = kzalloc(sizeof(*cache), GFP_NOFS);
5899         if (!cache)
5900                 return -ENOMEM;
5901
5902         cache->key.objectid = chunk_offset;
5903         cache->key.offset = size;
5904         cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
5905         atomic_set(&cache->count, 1);
5906         spin_lock_init(&cache->lock);
5907         mutex_init(&cache->alloc_mutex);
5908         mutex_init(&cache->cache_mutex);
5909         INIT_LIST_HEAD(&cache->list);
5910
5911         btrfs_set_block_group_used(&cache->item, bytes_used);
5912         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
5913         cache->flags = type;
5914         btrfs_set_block_group_flags(&cache->item, type);
5915
5916         ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
5917                                 &cache->space_info);
5918         BUG_ON(ret);
5919         down_write(&cache->space_info->groups_sem);
5920         list_add_tail(&cache->list, &cache->space_info->block_groups);
5921         up_write(&cache->space_info->groups_sem);
5922
5923         ret = btrfs_add_block_group_cache(root->fs_info, cache);
5924         BUG_ON(ret);
5925
5926         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
5927                                 sizeof(cache->item));
5928         BUG_ON(ret);
5929
5930         finish_current_insert(trans, extent_root, 0);
5931         ret = del_pending_extents(trans, extent_root, 0);
5932         BUG_ON(ret);
5933         set_avail_alloc_bits(extent_root->fs_info, type);
5934
5935         return 0;
5936 }
5937
5938 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
5939                              struct btrfs_root *root, u64 group_start)
5940 {
5941         struct btrfs_path *path;
5942         struct btrfs_block_group_cache *block_group;
5943         struct btrfs_key key;
5944         int ret;
5945
5946         root = root->fs_info->extent_root;
5947
5948         block_group = btrfs_lookup_block_group(root->fs_info, group_start);
5949         BUG_ON(!block_group);
5950         BUG_ON(!block_group->ro);
5951
5952         memcpy(&key, &block_group->key, sizeof(key));
5953
5954         path = btrfs_alloc_path();
5955         BUG_ON(!path);
5956
5957         spin_lock(&root->fs_info->block_group_cache_lock);
5958         rb_erase(&block_group->cache_node,
5959                  &root->fs_info->block_group_cache_tree);
5960         spin_unlock(&root->fs_info->block_group_cache_lock);
5961         btrfs_remove_free_space_cache(block_group);
5962         down_write(&block_group->space_info->groups_sem);
5963         list_del(&block_group->list);
5964         up_write(&block_group->space_info->groups_sem);
5965
5966         spin_lock(&block_group->space_info->lock);
5967         block_group->space_info->total_bytes -= block_group->key.offset;
5968         block_group->space_info->bytes_readonly -= block_group->key.offset;
5969         spin_unlock(&block_group->space_info->lock);
5970         block_group->space_info->full = 0;
5971
5972         put_block_group(block_group);
5973         put_block_group(block_group);
5974
5975         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5976         if (ret > 0)
5977                 ret = -EIO;
5978         if (ret < 0)
5979                 goto out;
5980
5981         ret = btrfs_del_item(trans, root, path);
5982 out:
5983         btrfs_free_path(path);
5984         return ret;
5985 }