Pileus Git - ~andy/linux/blob - fs/btrfs/extent-tree.c

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