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

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