fs/btrfs/extent-tree.c

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