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